Add stopCmd for custom stopping instructions (#136)

Allow configuration of how a model is stopped before swapping. Setting `cmdStop` in the configuration will override the default behaviour and enables better integration with other process/container managers like docker or podman.

.coderabbit.yaml

@@ -0,0 +1,15 @@
+# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
+language: "en-US"
+early_access: false
+reviews:
+  profile: "chill"
+  request_changes_workflow: false
+  high_level_summary: true
+  poem: false
+  review_status: true
+  collapse_walkthrough: false
+  auto_review:
+    enabled: true
+    drafts: false
+chat:
+  auto_reply: true

.github/ISSUE_TEMPLATE/bug-report.md

@@ -0,0 +1,37 @@
+---
+name: Bug Report
+about: Something is not working as expected...
+title: ''
+labels: bug
+assignees: ''
+
+---
+
+**Describe the bug**
+A clear and concise description of what the bug is.
+
+**Expected behaviour**
+A clear and concise description of what you expected to happen.
+
+**Operating system and version**
+
+- OS: (linux, osx, windows, freebsd, etc)
+- GPUs: (list architecture)
+
+**My Configuration**
+
+```yaml
+# copy / paste your configuration here
+```
+
+**Proxy Logs**
+
+```
+# copy / paste from /logs
+```
+
+**Upstream Logs**
+
+```
+# copy/paste from /logs
+```

.github/workflows/closeinactive.yml

@@ -0,0 +1,23 @@
+# https://docs.github.com/en/actions/use-cases-and-examples/project-management/closing-inactive-issues
+name: Close inactive issues
+on:
+  schedule:
+    - cron: "32 1 * * *"
+
+jobs:
+  close-issues:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+    steps:
+      - uses: actions/stale@v9
+        with:
+          days-before-issue-stale: 14
+          days-before-issue-close: 14
+          stale-issue-label: "stale"
+          stale-issue-message: "This issue is stale because it has been open for 2 weeks with no activity."
+          close-issue-message: "This issue was closed because it has been inactive for 2 weeks since being marked as stale."
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+          repo-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/containers.yml

@@ -0,0 +1,46 @@
+name: Build Containers
+
+on:
+  # time has no specific meaning, trying to time it after
+  # the llama.cpp daily packages are published
+  # https://github.com/ggml-org/llama.cpp/blob/master/.github/workflows/docker.yml
+  schedule:
+    - cron: "37 5 * * *"
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        platform: [intel, cuda, vulkan, cpu, musa]
+      fail-fast: false
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v2
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Run build-container
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: ./docker/build-container.sh ${{ matrix.platform }} true
+
+  # note make sure mostlygeek/llama-swap has admin rights to the llama-swap package
+  # see: https://github.com/actions/delete-package-versions/issues/74
+  delete-untagged-containers:
+    needs: build-and-push
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/delete-package-versions@v5
+        with:
+          package-name: 'llama-swap'
+          package-type: 'container'
+          delete-only-untagged-versions: 'true'

.github/workflows/go-ci-windows.yml

@@ -0,0 +1,50 @@
+name: Windows CI
+
+on:
+  push:
+    branches: [ "main" ]
+
+  pull_request:
+    branches: [ "main" ]
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+
+  run-tests:
+    runs-on: windows-latest
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v4
+      with:
+        go-version: '1.23'
+
+    # cache simple-responder to save the build time
+    - name: Restore Simple Responder
+      id: restore-simple-responder
+      uses: actions/cache/restore@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    # necessary for testing proxy/Process swapping
+    - name: Create simple-responder
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      shell: bash
+      run: make simple-responder-windows
+
+    - name: Save Simple Responder
+      # nothing new to save ... skip this step
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      id: save-simple-responder
+      uses: actions/cache/save@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    - name: Test all
+      shell: bash
+      run: make test-all

.github/workflows/go-ci.yml

@@ -0,0 +1,47 @@
+name: Linux CI
+
+on:
+  push:
+    branches: [ "main" ]
+
+  pull_request:
+    branches: [ "main" ]
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+
+  run-tests:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v4
+      with:
+        go-version: '1.23'
+
+    # cache simple-responder to save the build time
+    - name: Restore Simple Responder
+      id: restore-simple-responder
+      uses: actions/cache/restore@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    # necessary for testing proxy/Process swapping
+    - name: Create simple-responder
+      run: make simple-responder
+
+    - name: Save Simple Responder
+      # nothing new to save ... skip this step
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      id: save-simple-responder
+      uses: actions/cache/save@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    - name: Test all
+      run: make test-all

.github/workflows/release.yml

@@ -5,6 +5,9 @@ on:
     tags:
       - '*'
 
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
 permissions:
   contents: write
 
@@ -30,4 +33,4 @@ jobs:
           version: '~> v2'
           args: release --clean
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.gitignore

@@ -2,4 +2,5 @@
 .env
 build/
 dist/
-.vscode
+.vscode
+.DS_Store

.goreleaser.yaml

@@ -6,6 +6,25 @@ builds:
     goos:
       - linux
       - darwin
+      - freebsd
+      - windows
     goarch:
       - amd64
-      - arm64
+      - arm64
+    ignore:
+      - goos: freebsd
+        goarch: arm64
+      - goos: windows
+        goarch: arm64
+
+# use zip format for windows
+archives:
+  - id: default
+    format: tar.gz
+    name_template: "{{ .ProjectName }}_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
+    builds_info:
+      group: root
+      owner: root
+    format_overrides:
+      - goos: windows
+        format: zip

Makefile

@@ -2,6 +2,16 @@
 APP_NAME = llama-swap
 BUILD_DIR = build
 
+# Get the current Git hash
+GIT_HASH := $(shell git rev-parse --short HEAD)
+ifneq ($(shell git status --porcelain),)
+    # There are untracked changes
+    GIT_HASH := $(GIT_HASH)+
+endif
+
+# Capture the current build date in RFC3339 format
+BUILD_DATE := $(shell date -u +"%Y-%m-%dT%H:%M:%SZ")
+
 # Default target: Builds binaries for both OSX and Linux
 all: mac linux simple-responder
 
@@ -10,20 +20,25 @@ clean:
 	rm -rf $(BUILD_DIR)
 
 test:
-	go test -short -v ./proxy
+	go test -short -v -count=1 ./proxy
 
 test-all:
-	go test -v ./proxy
+	go test -v -count=1 ./proxy
 
 # Build OSX binary
 mac:
 	@echo "Building Mac binary..."
-	GOOS=darwin GOARCH=arm64 go build -o $(BUILD_DIR)/$(APP_NAME)-darwin-arm64
+	GOOS=darwin GOARCH=arm64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-darwin-arm64
 
 # Build Linux binary
 linux:
 	@echo "Building Linux binary..."
-	GOOS=linux GOARCH=amd64 go build -o $(BUILD_DIR)/$(APP_NAME)-linux-amd64
+	GOOS=linux GOARCH=amd64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-linux-amd64
+
+# Build Windows binary
+windows:
+	@echo "Building Windows binary..."
+	GOOS=windows GOARCH=amd64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-windows-amd64.exe
 
 # for testing proxy.Process
 simple-responder:
@@ -31,9 +46,27 @@ simple-responder:
 	GOOS=darwin GOARCH=arm64 go build -o $(BUILD_DIR)/simple-responder_darwin_arm64 misc/simple-responder/simple-responder.go
 	GOOS=linux GOARCH=amd64 go build -o $(BUILD_DIR)/simple-responder_linux_amd64 misc/simple-responder/simple-responder.go
 
+simple-responder-windows:
+	@echo "Building simple responder for windows"
+	GOOS=windows GOARCH=amd64 go build -o $(BUILD_DIR)/simple-responder.exe misc/simple-responder/simple-responder.go
+
 # Ensure build directory exists
 $(BUILD_DIR):
 	mkdir -p $(BUILD_DIR)
 
+# Create a new release tag
+release:
+	@echo "Checking for unstaged changes..."
+	@if [ -n "$(shell git status --porcelain)" ]; then \
+		echo "Error: There are unstaged changes. Please commit or stash your changes before creating a release tag." >&2; \
+		exit 1; \
+	fi
+
+# Get the highest tag in v{number} format, increment it, and create a new tag
+	@highest_tag=$$(git tag --sort=-v:refname | grep -E '^v[0-9]+$$' | head -n 1 || echo "v0"); \
+	new_tag="v$$(( $${highest_tag#v} + 1 ))"; \
+	echo "tagging new version: $$new_tag"; \
+	git tag "$$new_tag";
+
 # Phony targets
-.PHONY: all clean osx linux
+.PHONY: all clean mac linux windows simple-responder

README.md

@@ -1,42 +1,103 @@
+![llama-swap header image](header2.png)
+![GitHub Downloads (all assets, all releases)](https://img.shields.io/github/downloads/mostlygeek/llama-swap/total)
+![GitHub Actions Workflow Status](https://img.shields.io/github/actions/workflow/status/mostlygeek/llama-swap/go-ci.yml)
+![GitHub Repo stars](https://img.shields.io/github/stars/mostlygeek/llama-swap)
+
 # llama-swap
 
-![llama-swap header image](header.jpeg)
+llama-swap is a light weight, transparent proxy server that provides automatic model swapping to llama.cpp's server.
 
-llama-swap is a golang server that automatically swaps the llama.cpp server on demand. Since [llama.cpp's server](https://github.com/ggerganov/llama.cpp/tree/master/examples/server) can't swap models, let's swap the server instead!
+Written in golang, it is very easy to install (single binary with no dependancies) and configure (single yaml file). To get started, download a pre-built binary or use the provided docker images.
 
-Features:
+## Features:
 
 - ✅ Easy to deploy: single binary with no dependencies
-- ✅ Single yaml configuration file
-- ✅ Automatic switching between models
-- ✅ Full control over llama.cpp server settings per model
-- ✅ OpenAI API support (`v1/completions` and `v1/chat/completions`)
-- ✅ Multiple GPU support
-- ✅ Run multiple models at once with `profiles`
-- ✅ Remote log monitoring at `/log`
-- ✅ Automatic unloading of models from GPUs after timeout
+- ✅ Easy to config: single yaml file
+- ✅ On-demand model switching
+- ✅ OpenAI API supported endpoints:
+  - `v1/completions`
+  - `v1/chat/completions`
+  - `v1/embeddings`
+  - `v1/rerank`
+  - `v1/audio/speech` ([#36](https://github.com/mostlygeek/llama-swap/issues/36))
+  - `v1/audio/transcriptions` ([docs](https://github.com/mostlygeek/llama-swap/issues/41#issuecomment-2722637867))
+- ✅ llama-swap custom API endpoints
+  - `/log` - remote log monitoring
+  - `/upstream/:model_id` - direct access to upstream HTTP server ([demo](https://github.com/mostlygeek/llama-swap/pull/31))
+  - `/unload` - manually unload running models ([#58](https://github.com/mostlygeek/llama-swap/issues/58))
+  - `/running` - list currently running models ([#61](https://github.com/mostlygeek/llama-swap/issues/61))
+- ✅ Run multiple models at once with `Groups` ([#107](https://github.com/mostlygeek/llama-swap/issues/107))
+- ✅ Automatic unloading of models after timeout by setting a `ttl`
+- ✅ Use any local OpenAI compatible server (llama.cpp, vllm, tabbyAPI, etc)
+- ✅ Docker and Podman support
+- ✅ Full control over server settings per model
+
+## How does llama-swap work?
+
+When a request is made to an OpenAI compatible endpoint, lama-swap will extract the `model` value and load the appropriate server configuration to serve it. If the wrong upstream server is running, it will be replaced with the correct one. This is where the "swap" part comes in. The upstream server is automatically swapped to the correct one to serve the request.
+
+In the most basic configuration llama-swap handles one model at a time. For more advanced use cases, the `groups` feature allows multiple models to be loaded at the same time. You have complete control over how your system resources are used.
 
 ## config.yaml
 
 llama-swap's configuration is purposefully simple.
 
+```yaml
+models:
+  "qwen2.5":
+    proxy: "http://127.0.0.1:9999"
+    cmd: |
+      /app/llama-server
+      -hf bartowski/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M
+      --port 9999
+
+  "smollm2":
+    proxy: "http://127.0.0.1:9999"
+    cmd: |
+      /app/llama-server
+      -hf bartowski/SmolLM2-135M-Instruct-GGUF:Q4_K_M
+      --port 9999
+```
+
+<details>
+<summary>But also very powerful ...</summary>
+
 ```yaml
 # Seconds to wait for llama.cpp to load and be ready to serve requests
 # Default (and minimum) is 15 seconds
 healthCheckTimeout: 60
 
+# Valid log levels: debug, info (default), warn, error
+logLevel: info
+
+# Automatic Port Values
+# use ${PORT} in model.cmd and model.proxy to use an automatic port number
+# when you use ${PORT} you can omit a custom model.proxy value, as it will
+# default to http://localhost:${PORT}
+
+# override the default port (5800) for automatic port values
+startPort: 10001
+
 # define valid model values and the upstream server start
 models:
   "llama":
-    cmd: llama-server --port 8999 -m Llama-3.2-1B-Instruct-Q4_K_M.gguf
+    # multiline for readability
+    cmd: |
+      llama-server --port 8999
+      --model path/to/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf
+
+    # environment variables to pass to the command
+    env:
+      - "CUDA_VISIBLE_DEVICES=0"
 
     # where to reach the server started by cmd, make sure the ports match
+    # can be omitted if you use an automatic ${PORT} in cmd
     proxy: http://127.0.0.1:8999
 
     # aliases names to use this model for
     aliases:
-    - "gpt-4o-mini"
-    - "gpt-3.5-turbo"
+      - "gpt-4o-mini"
+      - "gpt-3.5-turbo"
 
     # check this path for an HTTP 200 OK before serving requests
     # default: /health to match llama.cpp
@@ -49,63 +110,198 @@ models:
     # default: 0 = never unload model
     ttl: 60
 
-  "qwen":
-    # environment variables to pass to the command
-    env:
-      - "CUDA_VISIBLE_DEVICES=0"
+    # `useModelName` overrides the model name in the request
+    # and sends a specific name to the upstream server
+    useModelName: "qwen:qwq"
 
-    # multiline for readability
-    cmd: >
-      llama-server --port 8999
-      --model path/to/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf
-    proxy: http://127.0.0.1:8999
+  # unlisted models do not show up in /v1/models or /upstream lists
+  # but they can still be requested as normal
+  "qwen-unlisted":
+    unlisted: true
+    cmd: llama-server --port ${PORT} -m Llama-3.2-1B-Instruct-Q4_K_M.gguf -ngl 0
 
-# profiles make it easy to managing multi model (and gpu) configurations.
+  # Docker Support (v26.1.4+ required!)
+  "docker-llama":
+    proxy: "http://127.0.0.1:${PORT}"
+    cmd: |
+      docker run --name dockertest
+      --init --rm -p ${PORT}:8080 -v /mnt/nvme/models:/models
+      ghcr.io/ggerganov/llama.cpp:server
+      --model '/models/Qwen2.5-Coder-0.5B-Instruct-Q4_K_M.gguf'
+
+    # use a custom command to stop the model when swapping. By default
+    # this is SIGTERM on POSIX systems, and taskkill on Windows systems
+    cmdStop: docker stop dockertest
+
+# Groups provide advanced controls over model swapping behaviour. Using groups
+# some models can be kept loaded indefinitely, while others are swapped out.
 #
 # Tips:
-#  - each model must be listening on a unique address and port
-#  - the model name is in this format: "profile_name/model", like "coding/qwen"
-#  - the profile will load and unload all models in the profile at the same time
-profiles:
-  coding:
-    - "qwen"
-    - "llama"
+#
+#  - models must be defined above in the Models section
+#  - a model can only be a member of one group
+#  - group behaviour is controlled via the `swap`, `exclusive` and `persistent` fields
+#  - see issue #109 for details
+#
+# NOTE: the example below uses model names that are not defined above for demonstration purposes
+groups:
+  # group1 is the default behaviour of llama-swap where only one model is allowed
+  # to run a time across the whole llama-swap instance
+  "group1":
+    # swap controls the model swapping behaviour in within the group
+    # - true : only one model is allowed to run at a time
+    # - false: all models can run together, no swapping
+    swap: true
+
+    # exclusive controls how the group affects other groups
+    # - true: causes all other groups to unload their models when this group runs a model
+    # - false: does not affect other groups
+    exclusive: true
+
+    # members references the models defined above
+    members:
+      - "llama"
+      - "qwen-unlisted"
+
+  # models in this group are never unloaded
+  "group2":
+    swap: false
+    exclusive: false
+    members:
+      - "docker-llama"
+      # (not defined above, here for example)
+      - "modelA"
+      - "modelB"
+
+  "forever":
+    # setting persistent to true causes the group to never be affected by the swapping behaviour of
+    # other groups. It is a shortcut to keeping some models always loaded.
+    persistent: true
+
+    # set swap/exclusive to false to prevent swapping inside the group and effect on other groups
+    swap: false
+    exclusive: false
+    members:
+      - "forever-modelA"
+      - "forever-modelB"
+      - "forever-modelc"
 ```
 
-More [examples](examples/README.md) are available for different use cases.
+### Use Case Examples
 
-## Installation
+- [config.example.yaml](config.example.yaml) includes example for supporting `v1/embeddings` and `v1/rerank` endpoints
+- [Speculative Decoding](examples/speculative-decoding/README.md) - using a small draft model can increase inference speeds from 20% to 40%. This example includes a configurations Qwen2.5-Coder-32B (2.5x increase) and Llama-3.1-70B (1.4x increase) in the best cases.
+- [Optimizing Code Generation](examples/benchmark-snakegame/README.md) - find the optimal settings for your machine. This example demonstrates defining multiple configurations and testing which one is fastest.
+- [Restart on Config Change](examples/restart-on-config-change/README.md) - automatically restart llama-swap when trying out different configurations.
+</details>
+
+## Docker Install ([download images](https://github.com/mostlygeek/llama-swap/pkgs/container/llama-swap))
+
+Docker is the quickest way to try out llama-swap:
+
+```shell
+# use CPU inference
+$ docker run -it --rm -p 9292:8080 ghcr.io/mostlygeek/llama-swap:cpu
+
+
+# qwen2.5 0.5B
+$ curl -s http://localhost:9292/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -H "Authorization: Bearer no-key" \
+    -d '{"model":"qwen2.5","messages": [{"role": "user","content": "tell me a joke"}]}' | \
+    jq -r '.choices[0].message.content'
+
+
+# SmolLM2 135M
+$ curl -s http://localhost:9292/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -H "Authorization: Bearer no-key" \
+    -d '{"model":"smollm2","messages": [{"role": "user","content": "tell me a joke"}]}' | \
+    jq -r '.choices[0].message.content'
+```
+
+<details>
+<summary>Docker images are nightly ...</summary>
+
+They include:
+
+- `ghcr.io/mostlygeek/llama-swap:cpu`
+- `ghcr.io/mostlygeek/llama-swap:cuda`
+- `ghcr.io/mostlygeek/llama-swap:intel`
+- `ghcr.io/mostlygeek/llama-swap:vulkan`
+- ROCm disabled until fixed in llama.cpp container
+
+Specific versions are also available and are tagged with the llama-swap, architecture and llama.cpp versions. For example: `ghcr.io/mostlygeek/llama-swap:v89-cuda-b4716`
+
+Beyond the demo you will likely want to run the containers with your downloaded models and custom configuration.
+
+```shell
+$ docker run -it --rm --runtime nvidia -p 9292:8080 \
+  -v /path/to/models:/models \
+  -v /path/to/custom/config.yaml:/app/config.yaml \
+  ghcr.io/mostlygeek/llama-swap:cuda
+```
+
+</details>
+
+## Bare metal Install ([download](https://github.com/mostlygeek/llama-swap/releases))
+
+Pre-built binaries are available for Linux, FreeBSD and Darwin (OSX). These are automatically published and are likely a few hours ahead of the docker releases. The baremetal install works with any OpenAI compatible server, not just llama-server.
 
 1. Create a configuration file, see [config.example.yaml](config.example.yaml)
 1. Download a [release](https://github.com/mostlygeek/llama-swap/releases) appropriate for your OS and architecture.
-    * _Note: Windows currently untested._
-1. Run the binary with `llama-swap --config path/to/config.yaml`
+1. Run the binary with `llama-swap --config path/to/config.yaml`.
+   Available flags:
+   - `--config`: Path to the configuration file (default: `config.yaml`).
+   - `--listen`: Address and port to listen on (default: `:8080`).
+   - `--version`: Show version information and exit.
+   - `--watch-config`: Automatically reload the configuration file when it changes. This will wait for in-flight requests to complete then stop all running models (default: `false`).
+
+### Building from source
+
+1. Install golang for your system
+1. `git clone git@github.com:mostlygeek/llama-swap.git`
+1. `make clean all`
+1. Binaries will be in `build/` subdirectory
 
 ## Monitoring Logs
 
-The `/logs` endpoint is available to monitor what llama-swap is doing. It will send the last 10KB of logs. Useful for monitoring the output of llama-server. It also supports streaming of logs.
+Open the `http://<host>/logs` with your browser to get a web interface with streaming logs.
 
-Usage:
+Of course, CLI access is also supported:
 
-```
+```shell
 # sends up to the last 10KB of logs
 curl http://host/logs'
 
-# streams logs using chunk encoding
+# streams combined logs
 curl -Ns 'http://host/logs/stream'
 
+# just llama-swap's logs
+curl -Ns 'http://host/logs/stream/proxy'
+
+# just upstream's logs
+curl -Ns 'http://host/logs/stream/upstream'
+
+# stream and filter logs with linux pipes
+curl -Ns http://host/logs/stream | grep 'eval time'
+
 # skips history and just streams new log entries
 curl -Ns 'http://host/logs/stream?no-history'
-
-# streams logs using Server Sent Events
-curl -Ns 'http://host/logs/streamSSE'
 ```
 
+## Do I need to use llama.cpp's server (llama-server)?
+
+Any OpenAI compatible server would work. llama-swap was originally designed for llama-server and it is the best supported.
+
+For Python based inference servers like vllm or tabbyAPI it is recommended to run them via podman or docker. This provides clean environment isolation as well as responding correctly to `SIGTERM` signals to shutdown.
+
 ## Systemd Unit Files
 
 Use this unit file to start llama-swap on boot. This is only tested on Ubuntu.
 
 `/etc/systemd/system/llama-swap.service`
+
 ```
 [Unit]
 Description=llama-swap
@@ -126,8 +322,6 @@ StartLimitInterval=30
 WantedBy=multi-user.target
 ```
 
-## Building from Source
+## Star History
 
-1. Install golang for your system
-1. run `make clean all`
-1. binaries will be built into `build/` directory
+[![Star History Chart](https://api.star-history.com/svg?repos=mostlygeek/llama-swap&type=Date)](https://www.star-history.com/#mostlygeek/llama-swap&Date)

config.example.yaml

@@ -1,14 +1,24 @@
 # Seconds to wait for llama.cpp to be available to serve requests
 # Default (and minimum): 15 seconds
-healthCheckTimeout: 15
+healthCheckTimeout: 90
+
+# valid log levels: debug, info (default), warn, error
+logLevel: debug
+
+# creating a coding profile with models for code generation and general questions
+groups:
+  coding:
+    swap: false
+    members:
+      - "qwen"
+      - "llama"
 
 models:
   "llama":
-    cmd: >
+    cmd: |
       models/llama-server-osx
-      --port 9001
+      --port ${PORT}
       -m models/Llama-3.2-1B-Instruct-Q4_0.gguf
-    proxy: http://127.0.0.1:9001
 
     # list of model name aliases this llama.cpp instance can serve
     aliases:
@@ -21,18 +31,47 @@ models:
     ttl: 5
 
   "qwen":
-    cmd: models/llama-server-osx --port 9002 -m models/qwen2.5-0.5b-instruct-q8_0.gguf
-    proxy: http://127.0.0.1:9002
+    cmd: models/llama-server-osx --port ${PORT} -m models/qwen2.5-0.5b-instruct-q8_0.gguf
     aliases:
-    - gpt-3.5-turbo
+      - gpt-3.5-turbo
+
+  # Embedding example with Nomic
+  # https://huggingface.co/nomic-ai/nomic-embed-text-v1.5-GGUF
+  "nomic":
+    cmd: |
+      models/llama-server-osx --port ${PORT}
+      -m models/nomic-embed-text-v1.5.Q8_0.gguf
+      --ctx-size 8192
+      --batch-size 8192
+      --rope-scaling yarn
+      --rope-freq-scale 0.75
+      -ngl 99
+      --embeddings
+
+  # Reranking example with bge-reranker
+  # https://huggingface.co/gpustack/bge-reranker-v2-m3-GGUF
+  "bge-reranker":
+    cmd: |
+      models/llama-server-osx --port ${PORT}
+      -m models/bge-reranker-v2-m3-Q4_K_M.gguf
+      --ctx-size 8192
+      --reranking
+
+  # Docker Support (v26.1.4+ required!)
+  "dockertest":
+    cmd: |
+      docker run --name dockertest
+      --init --rm -p ${PORT}:8080 -v /mnt/nvme/models:/models
+      ghcr.io/ggerganov/llama.cpp:server
+      --model '/models/Qwen2.5-Coder-0.5B-Instruct-Q4_K_M.gguf'
 
   "simple":
     # example of setting environment variables
     env:
       - CUDA_VISIBLE_DEVICES=0,1
       - env1=hello
-    cmd: build/simple-responder --port 8999
-    proxy: http://127.0.0.1:8999
+    cmd: build/simple-responder --port ${PORT}
+    unlisted: true
 
     # use "none" to skip check. Caution this may cause some requests to fail
     # until the upstream server is ready for traffic
@@ -42,12 +81,8 @@ models:
   "broken":
     cmd: models/llama-server-osx --port 8999 -m models/doesnotexist.gguf
     proxy: http://127.0.0.1:8999
+    unlisted: true
   "broken_timeout":
     cmd: models/llama-server-osx --port 8999 -m models/qwen2.5-0.5b-instruct-q8_0.gguf
     proxy: http://127.0.0.1:9000
-
-# creating a coding profile with models for code generation and general questions
-profiles:
-  coding:
-    - "qwen"
-    - "llama"
+    unlisted: true

docker/build-container.sh

@@ -0,0 +1,55 @@
+#!/bin/bash
+
+cd $(dirname "$0")
+
+ARCH=$1
+PUSH_IMAGES=${2:-false}
+
+# List of allowed architectures
+ALLOWED_ARCHS=("intel" "vulkan" "musa" "cuda" "cpu")
+
+# Check if ARCH is in the allowed list
+if [[ ! " ${ALLOWED_ARCHS[@]} " =~ " ${ARCH} " ]]; then
+  echo "Error: ARCH must be one of the following: ${ALLOWED_ARCHS[@]}"
+  exit 1
+fi
+
+# Check if GITHUB_TOKEN is set and not empty
+if [[ -z "$GITHUB_TOKEN" ]]; then
+  echo "Error: GITHUB_TOKEN is not set or is empty."
+  exit 1
+fi
+
+# the most recent llama-swap tag
+# have to strip out the 'v' due to .tar.gz file naming
+LS_VER=$(curl -s https://api.github.com/repos/mostlygeek/llama-swap/releases/latest | jq -r .tag_name | sed 's/v//')
+
+if [ "$ARCH" == "cpu" ]; then
+    # cpu only containers just use the latest available
+    CONTAINER_LATEST="ghcr.io/mostlygeek/llama-swap:cpu"
+    echo "Building ${CONTAINER_LATEST} $LS_VER"
+    docker build -f llama-swap.Containerfile --build-arg BASE_TAG=server --build-arg LS_VER=${LS_VER} -t ${CONTAINER_LATEST} .
+    if [ "$PUSH_IMAGES" == "true" ]; then
+      docker push ${CONTAINER_LATEST}
+    fi
+else
+    LCPP_TAG=$(curl -s -H "Authorization: Bearer $GITHUB_TOKEN" \
+        "https://api.github.com/users/ggml-org/packages/container/llama.cpp/versions" \
+        | jq -r --arg arch "$ARCH" '.[] | select(.metadata.container.tags[] | startswith("server-\($arch)")) | .metadata.container.tags[]' \
+        | sort -r | head -n1 | awk -F '-' '{print $3}')
+
+    # Abort if LCPP_TAG is empty.
+    if [[ -z "$LCPP_TAG" ]]; then
+        echo "Abort: Could not find llama-server container for arch: $ARCH"
+        exit 1
+    fi
+
+    CONTAINER_TAG="ghcr.io/mostlygeek/llama-swap:v${LS_VER}-${ARCH}-${LCPP_TAG}"
+    CONTAINER_LATEST="ghcr.io/mostlygeek/llama-swap:${ARCH}"
+    echo "Building ${CONTAINER_TAG} $LS_VER"
+    docker build -f llama-swap.Containerfile --build-arg BASE_TAG=server-${ARCH}-${LCPP_TAG} --build-arg LS_VER=${LS_VER} -t ${CONTAINER_TAG} -t ${CONTAINER_LATEST} .
+    if [ "$PUSH_IMAGES" == "true" ]; then
+      docker push ${CONTAINER_TAG}
+      docker push ${CONTAINER_LATEST}
+    fi
+fi

docker/config.example.yaml

@@ -0,0 +1,17 @@
+healthCheckTimeout: 300
+logRequests: true
+
+models:
+  "qwen2.5":
+    proxy: "http://127.0.0.1:9999"
+    cmd: >
+      /app/llama-server
+      -hf bartowski/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M
+      --port 9999
+
+  "smollm2":
+    proxy: "http://127.0.0.1:9999"
+    cmd: >
+      /app/llama-server
+      -hf bartowski/SmolLM2-135M-Instruct-GGUF:Q4_K_M
+      --port 9999

docker/llama-swap.Containerfile

@@ -0,0 +1,16 @@
+ARG BASE_TAG=server-cuda
+FROM ghcr.io/ggml-org/llama.cpp:${BASE_TAG}
+
+# has to be after the FROM
+ARG LS_VER=89
+
+WORKDIR /app
+RUN \
+    curl -LO https://github.com/mostlygeek/llama-swap/releases/download/v"${LS_VER}"/llama-swap_"${LS_VER}"_linux_amd64.tar.gz && \
+    tar -zxf llama-swap_"${LS_VER}"_linux_amd64.tar.gz && \
+    rm llama-swap_"${LS_VER}"_linux_amd64.tar.gz
+
+COPY config.example.yaml /app/config.yaml
+
+HEALTHCHECK CMD curl -f http://localhost:8080/ || exit 1
+ENTRYPOINT [ "/app/llama-swap", "-config", "/app/config.yaml" ]

examples/README.md

@@ -1,9 +1,6 @@
-# Example Configurations
+# Example Configs and Use Cases
 
-Learning by example is best.
-
-Here in the `examples/` folder are llama-swap configurations that can be used on your local LLM server.
-
-## List
+A collections of usecases and examples for getting the most out of llama-swap.
 
 * [Speculative Decoding](speculative-decoding/README.md) - using a small draft model can increase inference speeds from 20% to 40%. This example includes a configurations Qwen2.5-Coder-32B (2.5x increase) and Llama-3.1-70B (1.4x increase) in the best cases.
+* [Optimizing Code Generation](benchmark-snakegame/README.md) - find the optimal settings for your machine. This example demonstrates defining multiple configurations and testing which one is fastest.

examples/aider-qwq-coder/README.md

@@ -0,0 +1,153 @@
+# aider, QwQ, Qwen-Coder 2.5 and llama-swap
+
+This guide show how to use aider and llama-swap to get a 100% local coding co-pilot setup. The focus is on the trickest part which is configuring aider, llama-swap and llama-server to work together.
+
+## Here's what you you need:
+
+- aider - [installation docs](https://aider.chat/docs/install.html)
+- llama-server - [download latest release](https://github.com/ggml-org/llama.cpp/releases)
+- llama-swap - [download latest release](https://github.com/mostlygeek/llama-swap/releases)
+- [QwQ 32B](https://huggingface.co/bartowski/Qwen_QwQ-32B-GGUF) and [Qwen Coder 2.5 32B](https://huggingface.co/bartowski/Qwen2.5-Coder-32B-Instruct-GGUF) models
+- 24GB VRAM video card
+
+## Running aider
+
+The goal is getting this command line to work:
+
+```sh
+aider --architect \
+    --no-show-model-warnings \
+    --model openai/QwQ \
+    --editor-model openai/qwen-coder-32B \
+    --model-settings-file aider.model.settings.yml \
+    --openai-api-key "sk-na" \
+    --openai-api-base "http://10.0.1.24:8080/v1" \
+```
+
+Set `--openai-api-base` to the IP and port where your llama-swap is running.
+
+## Create an aider model settings file
+
+```yaml
+# aider.model.settings.yml
+
+#
+# !!! important: model names must match llama-swap configuration names !!!
+#
+
+- name: "openai/QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/qwen-coder-32B"
+  editor_model_name: "openai/qwen-coder-32B"
+
+- name: "openai/qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/qwen-coder-32B"
+```
+
+## llama-swap configuration
+
+```yaml
+# config.yaml
+
+# The parameters are tweaked to fit model+context into 24GB VRAM GPUs
+models:
+  "qwen-coder-32B":
+    proxy: "http://127.0.0.1:8999"
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 8999 --flash-attn --slots
+      --ctx-size 16000
+      --cache-type-k q8_0 --cache-type-v q8_0
+       -ngl 99
+      --model /path/to/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+
+  "QwQ":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 9503 --flash-attn --metrics--slots
+      --cache-type-k q8_0 --cache-type-v q8_0
+      --ctx-size 32000
+      --samplers "top_k;top_p;min_p;temperature;dry;typ_p;xtc"
+      --temp 0.6 --repeat-penalty 1.1 --dry-multiplier 0.5
+      --min-p 0.01 --top-k 40 --top-p 0.95
+      -ngl 99
+      --model /mnt/nvme/models/bartowski/Qwen_QwQ-32B-Q4_K_M.gguf
+```
+
+## Advanced, Dual GPU Configuration
+
+If you have _dual 24GB GPUs_ you can use llama-swap profiles to avoid swapping between QwQ and Qwen Coder.
+
+In llama-swap's configuration file:
+
+1. add a `profiles` section with `aider` as the profile name
+2. using the `env` field to specify the GPU IDs for each model
+
+```yaml
+# config.yaml
+
+# Add a profile for aider
+profiles:
+  aider:
+    - qwen-coder-32B
+    - QwQ
+
+models:
+  "qwen-coder-32B":
+    # manually set the GPU to run on
+    env:
+      - "CUDA_VISIBLE_DEVICES=0"
+    proxy: "http://127.0.0.1:8999"
+    cmd: /path/to/llama-server ...
+
+  "QwQ":
+    # manually set the GPU to run on
+    env:
+      - "CUDA_VISIBLE_DEVICES=1"
+    proxy: "http://127.0.0.1:9503"
+    cmd: /path/to/llama-server ...
+```
+
+Append the profile tag, `aider:`, to the model names in the model settings file
+
+```yaml
+# aider.model.settings.yml
+- name: "openai/aider:QwQ"
+  weak_model_name: "openai/aider:qwen-coder-32B-aider"
+  editor_model_name: "openai/aider:qwen-coder-32B-aider"
+
+- name: "openai/aider:qwen-coder-32B"
+  editor_model_name: "openai/aider:qwen-coder-32B-aider"
+```
+
+Run aider with:
+
+```sh
+$ aider --architect \
+    --no-show-model-warnings \
+    --model openai/aider:QwQ \
+    --editor-model openai/aider:qwen-coder-32B \
+    --config aider.conf.yml \
+    --model-settings-file aider.model.settings.yml
+    --openai-api-key "sk-na" \
+    --openai-api-base "http://10.0.1.24:8080/v1"
+```

examples/aider-qwq-coder/aider.model.settings.dualgpu.yml

@@ -0,0 +1,28 @@
+# this makes use of llama-swap's profile feature to
+# keep the architect and editor models in VRAM on different GPUs
+
+- name: "openai/aider:QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/aider:qwen-coder-32B"
+  editor_model_name: "openai/aider:qwen-coder-32B"
+
+- name: "openai/aider:qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/aider:qwen-coder-32B"

examples/aider-qwq-coder/aider.model.settings.yml

@@ -0,0 +1,26 @@
+- name: "openai/QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/qwen-coder-32B"
+  editor_model_name: "openai/qwen-coder-32B"
+
+- name: "openai/qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/qwen-coder-32B"
+

examples/aider-qwq-coder/llama-swap.yaml

@@ -0,0 +1,49 @@
+healthCheckTimeout: 300
+logLevel: debug
+
+profiles:
+    aider:
+      - qwen-coder-32B
+      - QwQ
+
+models:
+  "qwen-coder-32B":
+    env:
+      - "CUDA_VISIBLE_DEVICES=0"
+    aliases:
+      - coder
+    proxy: "http://127.0.0.1:8999"
+
+    # set appropriate paths for your environment
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 8999 --flash-attn --slots
+      --ctx-size 16000
+      --ctx-size-draft 16000
+      --model /path/to/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      --model-draft /path/to/Qwen2.5-Coder-1.5B-Instruct-Q8_0.gguf
+      -ngl 99 -ngld 99
+      --draft-max 16 --draft-min 4 --draft-p-min 0.4
+      --cache-type-k q8_0 --cache-type-v q8_0
+  "QwQ":
+    env:
+      - "CUDA_VISIBLE_DEVICES=1"
+    proxy: "http://127.0.0.1:9503"
+
+    # set appropriate paths for your environment
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 9503
+      --flash-attn --metrics
+      --slots
+      --model /path/to/Qwen_QwQ-32B-Q4_K_M.gguf
+      --cache-type-k q8_0 --cache-type-v q8_0
+      --ctx-size 32000
+      --samplers "top_k;top_p;min_p;temperature;dry;typ_p;xtc"
+      --temp 0.6
+      --repeat-penalty 1.1
+      --dry-multiplier 0.5
+      --min-p 0.01
+      --top-k 40
+      --top-p 0.95
+      -ngl 99 -ngld 99

examples/benchmark-snakegame/README.md

@@ -0,0 +1,123 @@
+# Optimizing Code Generation with llama-swap
+
+Finding the best mix of settings for your hardware can be time consuming. This example demonstrates using a custom configuration file to automate testing different scenarios to find the an optimal configuration.
+
+The benchmark writes a snake game in Python, TypeScript, and Swift using the Qwen 2.5 Coder models. The experiments were done using a 3090 and a P40.
+
+**Benchmark Scenarios**
+
+Three scenarios are tested:
+
+- 3090-only: Just the main model on the 3090
+- 3090-with-draft: the main and draft models on the 3090
+- 3090-P40-draft: the main model on the 3090 with the draft model offloaded to the P40
+
+**Available Devices**
+
+Use the following command to list available devices IDs for the configuration:
+
+```
+$ /mnt/nvme/llama-server/llama-server-f3252055 --list-devices
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 4 CUDA devices:
+  Device 0: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes
+  Device 1: Tesla P40, compute capability 6.1, VMM: yes
+  Device 2: Tesla P40, compute capability 6.1, VMM: yes
+  Device 3: Tesla P40, compute capability 6.1, VMM: yes
+Available devices:
+  CUDA0: NVIDIA GeForce RTX 3090 (24154 MiB, 406 MiB free)
+  CUDA1: Tesla P40 (24438 MiB, 22942 MiB free)
+  CUDA2: Tesla P40 (24438 MiB, 24144 MiB free)
+  CUDA3: Tesla P40 (24438 MiB, 24144 MiB free)
+```
+
+**Configuration**
+
+The configuration file, `benchmark-config.yaml`, defines the three scenarios:
+
+```yaml
+models:
+  "3090-only":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn
+      --slots
+
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --ctx-size 32768
+      --cache-type-k q8_0 --cache-type-v q8_0
+
+  "3090-with-draft":
+    proxy: "http://127.0.0.1:9503"
+    # --ctx-size 28500 max that can fit on 3090 after draft model
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn
+      --slots
+
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --model-draft /mnt/nvme/models/Qwen2.5-Coder-0.5B-Instruct-Q8_0.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 4
+      --draft-p-min 0.4
+      --device-draft CUDA0
+
+      --ctx-size 28500
+      --cache-type-k q8_0 --cache-type-v q8_0
+
+  "3090-P40-draft":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn --metrics
+      --slots
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --model-draft /mnt/nvme/models/Qwen2.5-Coder-0.5B-Instruct-Q8_0.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 4
+      --draft-p-min 0.4
+      --device-draft CUDA1
+
+      --ctx-size 32768
+      --cache-type-k q8_0 --cache-type-v q8_0
+```
+
+> Note in the `3090-with-draft` scenario the `--ctx-size` had to be reduced from 32768 to to accommodate the draft model.
+
+
+**Running the Benchmark**
+
+To run the benchmark, execute the following commands:
+
+1. `llama-swap -config benchmark-config.yaml`
+1. `./run-benchmark.sh http://localhost:8080 "3090-only" "3090-with-draft" "3090-P40-draft"`
+
+The [benchmark script](run-benchmark.sh) generates a CSV output of the results, which can be converted to a Markdown table for readability.
+
+**Results (tokens/second)**
+
+| model           | python | typescript | swift |
+|-----------------|--------|------------|-------|
+| 3090-only       | 34.03  | 34.01      | 34.01 |
+| 3090-with-draft | 106.65 | 70.48      | 57.89 |
+| 3090-P40-draft  | 81.54  | 60.35      | 46.50 |
+
+Many different factors, like the programming language, can have big impacts on the performance gains. However, with a custom configuration file for benchmarking it is easy to test the different variations to discover what's best for your hardware.
+
+Happy coding!

examples/benchmark-snakegame/run-benchmark.sh

@@ -0,0 +1,40 @@
+#!/usr/bin/env bash
+
+# This script generates a CSV file showing the token/second for generating a Snake Game in python, typescript and swift
+# It was created to test the effects of speculative decoding and the various draft settings on performance.
+#
+# Writing code with a low temperature seems to provide fairly consistent logic.
+#
+# Usage: ./benchmark.sh <url> <model1> [model2 ...]
+# Example: ./benchmark.sh http://localhost:8080 model1 model2
+
+if [ "$#" -lt 2 ]; then
+    echo "Usage: $0 <url> <model1> [model2 ...]"
+    exit 1
+fi
+
+url=$1; shift
+
+echo "model,python,typescript,swift"
+
+for model in "$@"; do
+
+    echo -n "$model,"
+
+    for lang in "python" "typescript" "swift"; do
+        # expects a llama.cpp after PR https://github.com/ggerganov/llama.cpp/pull/10548
+        # (Dec 3rd/2024)
+        time=$(curl -s --url "$url/v1/chat/completions" -d "{\"messages\": [{\"role\": \"system\", \"content\": \"you only write code.\"}, {\"role\": \"user\", \"content\": \"write snake game in $lang\"}], \"top_k\": 1, \"timings_per_token\":true, \"model\":\"$model\"}" | jq -r .timings.predicted_per_second)
+
+        if [ $? -ne 0 ]; then
+            time="error"
+            exit 1
+        fi
+
+        if [ "$lang" != "swift" ]; then
+            printf "%0.2f tps," $time
+        else
+            printf "%0.2f tps\n" $time
+        fi
+    done
+done

examples/restart-on-config-change/README.md

@@ -0,0 +1,51 @@
+# Restart llama-swap on config change
+
+Sometimes editing the configuration file can take a bit of trail and error to get a model configuration tuned just right. The `watch-and-restart.sh` script can be used to watch `config.yaml` for changes and restart `llama-swap` when it detects a change.
+
+```bash
+#!/bin/bash
+#
+# A simple watch and restart llama-swap when its configuration
+# file changes. Useful for trying out configuration changes
+# without manually restarting the server each time.
+if [ -z "$1" ]; then
+    echo "Usage: $0 <path to config.yaml>"
+    exit 1
+fi
+
+while true; do
+    # Start the process again
+    ./llama-swap-linux-amd64 -config $1 -listen :1867 &
+    PID=$!
+    echo "Started llama-swap with PID $PID"
+
+    # Wait for modifications in the specified directory or file
+    inotifywait -e modify "$1"
+
+    # Check if process exists before sending signal
+    if kill -0 $PID 2>/dev/null; then
+        echo "Sending SIGTERM to $PID"
+        kill -SIGTERM $PID
+        wait $PID
+    else
+        echo "Process $PID no longer exists"
+    fi
+    sleep 1
+done
+```
+
+## Usage and output example
+
+```bash
+$ ./watch-and-restart.sh config.yaml
+Started llama-swap with PID 495455
+Setting up watches.
+Watches established.
+llama-swap listening on :1867
+Sending SIGTERM to 495455
+Shutting down llama-swap
+Started llama-swap with PID 495486
+Setting up watches.
+Watches established.
+llama-swap listening on :1867
+```

go.mod

@@ -3,11 +3,16 @@ module github.com/mostlygeek/llama-swap
 go 1.23.0
 
 require (
+	github.com/fsnotify/fsnotify v1.9.0
+	github.com/gin-gonic/gin v1.10.0
 	github.com/stretchr/testify v1.9.0
+	github.com/tidwall/gjson v1.18.0
+	github.com/tidwall/sjson v1.2.5
 	gopkg.in/yaml.v3 v3.0.1
 )
 
 require (
+	github.com/billziss-gh/golib v0.2.0 // indirect
 	github.com/bytedance/sonic v1.11.6 // indirect
 	github.com/bytedance/sonic/loader v0.1.1 // indirect
 	github.com/cloudwego/base64x v0.1.4 // indirect
@@ -15,7 +20,6 @@ require (
 	github.com/davecgh/go-spew v1.1.1 // indirect
 	github.com/gabriel-vasile/mimetype v1.4.3 // indirect
 	github.com/gin-contrib/sse v0.1.0 // indirect
-	github.com/gin-gonic/gin v1.10.0 // indirect
 	github.com/go-playground/locales v0.14.1 // indirect
 	github.com/go-playground/universal-translator v0.18.1 // indirect
 	github.com/go-playground/validator/v10 v10.20.0 // indirect
@@ -28,12 +32,14 @@ require (
 	github.com/modern-go/reflect2 v1.0.2 // indirect
 	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
+	github.com/tidwall/match v1.1.1 // indirect
+	github.com/tidwall/pretty v1.2.1 // indirect
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.12 // indirect
 	golang.org/x/arch v0.8.0 // indirect
-	golang.org/x/crypto v0.23.0 // indirect
-	golang.org/x/net v0.25.0 // indirect
-	golang.org/x/sys v0.20.0 // indirect
-	golang.org/x/text v0.15.0 // indirect
+	golang.org/x/crypto v0.36.0 // indirect
+	golang.org/x/net v0.38.0 // indirect
+	golang.org/x/sys v0.31.0 // indirect
+	golang.org/x/text v0.23.0 // indirect
 	google.golang.org/protobuf v1.34.1 // indirect
 )

go.sum

@@ -1,3 +1,5 @@
+github.com/billziss-gh/golib v0.2.0 h1:NyvcAQdfvM8xokKkKotiligKjKXzuQD4PPykg1nKc/8=
+github.com/billziss-gh/golib v0.2.0/go.mod h1:mZpUYANXZkDKSnyYbX9gfnyxwe0ddRhUtfXcsD5r8dw=
 github.com/bytedance/sonic v1.11.6 h1:oUp34TzMlL+OY1OUWxHqsdkgC/Zfc85zGqw9siXjrc0=
 github.com/bytedance/sonic v1.11.6/go.mod h1:LysEHSvpvDySVdC2f87zGWf6CIKJcAvqab1ZaiQtds4=
 github.com/bytedance/sonic/loader v0.1.1 h1:c+e5Pt1k/cy5wMveRDyk2X4B9hF4g7an8N3zCYjJFNM=
@@ -9,12 +11,16 @@ github.com/cloudwego/iasm v0.2.0/go.mod h1:8rXZaNYT2n95jn+zTI1sDr+IgcD2GVs0nlbbQ
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/fsnotify/fsnotify v1.9.0 h1:2Ml+OJNzbYCTzsxtv8vKSFD9PbJjmhYF14k/jKC7S9k=
+github.com/fsnotify/fsnotify v1.9.0/go.mod h1:8jBTzvmWwFyi3Pb8djgCCO5IBqzKJ/Jwo8TRcHyHii0=
 github.com/gabriel-vasile/mimetype v1.4.3 h1:in2uUcidCuFcDKtdcBxlR0rJ1+fsokWf+uqxgUFjbI0=
 github.com/gabriel-vasile/mimetype v1.4.3/go.mod h1:d8uq/6HKRL6CGdk+aubisF/M5GcPfT7nKyLpA0lbSSk=
 github.com/gin-contrib/sse v0.1.0 h1:Y/yl/+YNO8GZSjAhjMsSuLt29uWRFHdHYUb5lYOV9qE=
 github.com/gin-contrib/sse v0.1.0/go.mod h1:RHrZQHXnP2xjPF+u1gW/2HnVO7nvIa9PG3Gm+fLHvGI=
 github.com/gin-gonic/gin v1.10.0 h1:nTuyha1TYqgedzytsKYqna+DfLos46nTv2ygFy86HFU=
 github.com/gin-gonic/gin v1.10.0/go.mod h1:4PMNQiOhvDRa013RKVbsiNwoyezlm2rm0uX/T7kzp5Y=
+github.com/go-playground/assert/v2 v2.2.0 h1:JvknZsQTYeFEAhQwI4qEt9cyV5ONwRHC+lYKSsYSR8s=
+github.com/go-playground/assert/v2 v2.2.0/go.mod h1:VDjEfimB/XKnb+ZQfWdccd7VUvScMdVu0Titje2rxJ4=
 github.com/go-playground/locales v0.14.1 h1:EWaQ/wswjilfKLTECiXz7Rh+3BjFhfDFKv/oXslEjJA=
 github.com/go-playground/locales v0.14.1/go.mod h1:hxrqLVvrK65+Rwrd5Fc6F2O76J/NuW9t0sjnWqG1slY=
 github.com/go-playground/universal-translator v0.18.1 h1:Bcnm0ZwsGyWbCzImXv+pAJnYK9S473LQFuzCbDbfSFY=
@@ -23,7 +29,11 @@ github.com/go-playground/validator/v10 v10.20.0 h1:K9ISHbSaI0lyB2eWMPJo+kOS/FBEx
 github.com/go-playground/validator/v10 v10.20.0/go.mod h1:dbuPbCMFw/DrkbEynArYaCwl3amGuJotoKCe95atGMM=
 github.com/goccy/go-json v0.10.2 h1:CrxCmQqYDkv1z7lO7Wbh2HN93uovUHgrECaO5ZrCXAU=
 github.com/goccy/go-json v0.10.2/go.mod h1:6MelG93GURQebXPDq3khkgXZkazVtN9CRI+MGFi0w8I=
+github.com/google/go-cmp v0.5.5 h1:Khx7svrCpmxxtHBq5j2mp/xVjsi8hQMfNLvJFAlrGgU=
+github.com/google/go-cmp v0.5.5/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
+github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510 h1:El6M4kTTCOh6aBiKaUGG7oYTSPP8MxqL4YI3kZKwcP4=
+github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510/go.mod h1:pupxD2MaaD3pAXIBCelhxNneeOaAeabZDe5s4K6zSpQ=
 github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
 github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
 github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
@@ -55,6 +65,16 @@ github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o
 github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
 github.com/stretchr/testify v1.9.0 h1:HtqpIVDClZ4nwg75+f6Lvsy/wHu+3BoSGCbBAcpTsTg=
 github.com/stretchr/testify v1.9.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
+github.com/tidwall/gjson v1.14.2/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
+github.com/tidwall/gjson v1.18.0 h1:FIDeeyB800efLX89e5a8Y0BNH+LOngJyGrIWxG2FKQY=
+github.com/tidwall/gjson v1.18.0/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
+github.com/tidwall/match v1.1.1 h1:+Ho715JplO36QYgwN9PGYNhgZvoUSc9X2c80KVTi+GA=
+github.com/tidwall/match v1.1.1/go.mod h1:eRSPERbgtNPcGhD8UCthc6PmLEQXEWd3PRB5JTxsfmM=
+github.com/tidwall/pretty v1.2.0/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
+github.com/tidwall/pretty v1.2.1 h1:qjsOFOWWQl+N3RsoF5/ssm1pHmJJwhjlSbZ51I6wMl4=
+github.com/tidwall/pretty v1.2.1/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
+github.com/tidwall/sjson v1.2.5 h1:kLy8mja+1c9jlljvWTlSazM7cKDRfJuR/bOJhcY5NcY=
+github.com/tidwall/sjson v1.2.5/go.mod h1:Fvgq9kS/6ociJEDnK0Fk1cpYF4FIW6ZF7LAe+6jwd28=
 github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS4MhqMhdFk5YI=
 github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
 github.com/ugorji/go/codec v1.2.12 h1:9LC83zGrHhuUA9l16C9AHXAqEV/2wBQ4nkvumAE65EE=
@@ -62,16 +82,18 @@ github.com/ugorji/go/codec v1.2.12/go.mod h1:UNopzCgEMSXjBc6AOMqYvWC1ktqTAfzJZUZ
 golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
 golang.org/x/arch v0.8.0 h1:3wRIsP3pM4yUptoR96otTUOXI367OS0+c9eeRi9doIc=
 golang.org/x/arch v0.8.0/go.mod h1:FEVrYAQjsQXMVJ1nsMoVVXPZg6p2JE2mx8psSWTDQys=
-golang.org/x/crypto v0.23.0 h1:dIJU/v2J8Mdglj/8rJ6UUOM3Zc9zLZxVZwwxMooUSAI=
-golang.org/x/crypto v0.23.0/go.mod h1:CKFgDieR+mRhux2Lsu27y0fO304Db0wZe70UKqHu0v8=
-golang.org/x/net v0.25.0 h1:d/OCCoBEUq33pjydKrGQhw7IlUPI2Oylr+8qLx49kac=
-golang.org/x/net v0.25.0/go.mod h1:JkAGAh7GEvH74S6FOH42FLoXpXbE/aqXSrIQjXgsiwM=
+golang.org/x/crypto v0.36.0 h1:AnAEvhDddvBdpY+uR+MyHmuZzzNqXSe/GvuDeob5L34=
+golang.org/x/crypto v0.36.0/go.mod h1:Y4J0ReaxCR1IMaabaSMugxJES1EpwhBHhv2bDHklZvc=
+golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8=
+golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8=
 golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.20.0 h1:Od9JTbYCk261bKm4M/mw7AklTlFYIa0bIp9BgSm1S8Y=
-golang.org/x/sys v0.20.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/text v0.15.0 h1:h1V/4gjBv8v9cjcR6+AR5+/cIYK5N/WAgiv4xlsEtAk=
-golang.org/x/text v0.15.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
+golang.org/x/sys v0.31.0 h1:ioabZlmFYtWhL+TRYpcnNlLwhyxaM9kWTDEmfnprqik=
+golang.org/x/sys v0.31.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
+golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543 h1:E7g+9GITq07hpfrRu66IVDexMakfv52eLZ2CXBWiKr4=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 google.golang.org/protobuf v1.34.1 h1:9ddQBjfCyZPOHPUiPxpYESBLc+T8P3E+Vo4IbKZgFWg=
 google.golang.org/protobuf v1.34.1/go.mod h1:c6P6GXX6sHbq/GpV6MGZEdwhWPcYBgnhAHhKbcUYpos=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=

llama-swap.go

@@ -1,27 +1,52 @@
 package main
 
 import (
+	"context"
 	"flag"
 	"fmt"
+	"log"
+	"net/http"
 	"os"
+	"os/signal"
+	"path/filepath"
+	"syscall"
+	"time"
 
+	"github.com/fsnotify/fsnotify"
 	"github.com/gin-gonic/gin"
 	"github.com/mostlygeek/llama-swap/proxy"
 )
 
+var (
+	version string = "0"
+	commit  string = "abcd1234"
+	date    string = "unknown"
+)
+
 func main() {
 	// Define a command-line flag for the port
 	configPath := flag.String("config", "config.yaml", "config file name")
 	listenStr := flag.String("listen", ":8080", "listen ip/port")
+	showVersion := flag.Bool("version", false, "show version of build")
+	watchConfig := flag.Bool("watch-config", false, "Automatically reload config file on change")
 
 	flag.Parse() // Parse the command-line flags
 
+	if *showVersion {
+		fmt.Printf("version: %s (%s), built at %s\n", version, commit, date)
+		os.Exit(0)
+	}
+
 	config, err := proxy.LoadConfig(*configPath)
 	if err != nil {
 		fmt.Printf("Error loading config: %v\n", err)
 		os.Exit(1)
 	}
 
+	if len(config.Profiles) > 0 {
+		fmt.Println("WARNING: Profile functionality has been removed in favor of Groups. See the README for more information.")
+	}
+
 	if mode := os.Getenv("GIN_MODE"); mode != "" {
 		gin.SetMode(mode)
 	} else {
@@ -29,9 +54,136 @@ func main() {
 	}
 
 	proxyManager := proxy.New(config)
-	fmt.Println("llama-swap listening on " + *listenStr)
-	if err := proxyManager.Run(*listenStr); err != nil {
-		fmt.Printf("Server error: %v\n", err)
-		os.Exit(1)
+
+	// Setup channels for server management
+	reloadChan := make(chan *proxy.ProxyManager)
+	exitChan := make(chan struct{})
+	sigChan := make(chan os.Signal, 1)
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+
+	// Create server with initial handler
+	srv := &http.Server{
+		Addr:    *listenStr,
+		Handler: proxyManager,
+	}
+
+	// Start server
+	fmt.Printf("llama-swap listening on %s\n", *listenStr)
+	go func() {
+		if err := srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
+			fmt.Printf("Fatal server error: %v\n", err)
+			close(exitChan)
+		}
+	}()
+
+	// Handle config reloads and signals
+	go func() {
+		currentManager := proxyManager
+		for {
+			select {
+			case newManager := <-reloadChan:
+				log.Println("Config change detected, waiting for in-flight requests to complete...")
+				// Stop old manager processes gracefully (this waits for in-flight requests)
+				currentManager.StopProcesses(proxy.StopWaitForInflightRequest)
+				// Now do a full shutdown to clear the process map
+				currentManager.Shutdown()
+				currentManager = newManager
+				srv.Handler = newManager
+				log.Println("Server handler updated with new config")
+			case sig := <-sigChan:
+				fmt.Printf("Received signal %v, shutting down...\n", sig)
+				ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+				defer cancel()
+				currentManager.Shutdown()
+				if err := srv.Shutdown(ctx); err != nil {
+					fmt.Printf("Server shutdown error: %v\n", err)
+				}
+				close(exitChan)
+				return
+			}
+		}
+	}()
+
+	// Start file watcher if requested
+	if *watchConfig {
+		absConfigPath, err := filepath.Abs(*configPath)
+		if err != nil {
+			log.Printf("Error getting absolute path for config: %v. File watching disabled.", err)
+		} else {
+			go watchConfigFileWithReload(absConfigPath, reloadChan)
+		}
+	}
+
+	// Wait for exit signal
+	<-exitChan
+}
+
+// watchConfigFileWithReload monitors the configuration file and sends new ProxyManager instances through reloadChan.
+func watchConfigFileWithReload(configPath string, reloadChan chan<- *proxy.ProxyManager) {
+	watcher, err := fsnotify.NewWatcher()
+	if err != nil {
+		log.Printf("Error creating file watcher: %v. File watching disabled.", err)
+		return
+	}
+	defer watcher.Close()
+
+	err = watcher.Add(configPath)
+	if err != nil {
+		log.Printf("Error adding config path (%s) to watcher: %v. File watching disabled.", configPath, err)
+		return
+	}
+
+	log.Printf("Watching config file for changes: %s", configPath)
+
+	var debounceTimer *time.Timer
+	debounceDuration := 2 * time.Second
+
+	for {
+		select {
+		case event, ok := <-watcher.Events:
+			if !ok {
+				return
+			}
+			// We only care about writes to the specific config file
+			if event.Name == configPath && event.Has(fsnotify.Write) {
+				// Reset or start the debounce timer
+				if debounceTimer != nil {
+					debounceTimer.Stop()
+				}
+				debounceTimer = time.AfterFunc(debounceDuration, func() {
+					log.Printf("Config file modified: %s, reloading...", event.Name)
+
+					// Try up to 3 times with exponential backoff
+					var newConfig proxy.Config
+					var err error
+					for retries := 0; retries < 3; retries++ {
+						// Load new configuration
+						newConfig, err = proxy.LoadConfig(configPath)
+						if err == nil {
+							break
+						}
+						log.Printf("Error loading new config (attempt %d/3): %v", retries+1, err)
+						if retries < 2 {
+							time.Sleep(time.Duration(1<<retries) * time.Second)
+						}
+					}
+					if err != nil {
+						log.Printf("Failed to load new config after retries: %v", err)
+						return
+					}
+
+					// Create new ProxyManager with new config
+					newPM := proxy.New(newConfig)
+					reloadChan <- newPM
+					log.Println("Config reloaded successfully")
+				})
+			}
+		case err, ok := <-watcher.Errors:
+			if !ok {
+				log.Println("File watcher error channel closed.")
+				return
+			}
+			log.Printf("File watcher error: %v", err)
+		}
 	}
 }

misc/simple-responder/simple-responder.go

@@ -12,18 +12,22 @@ import (
 	"time"
 
 	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
 )
 
 func main() {
 	gin.SetMode(gin.TestMode)
 	// Define a command-line flag for the port
 	port := flag.String("port", "8080", "port to listen on")
+	expectedModel := flag.String("model", "TheExpectedModel", "model name to expect")
 
 	// Define a command-line flag for the response message
 	responseMessage := flag.String("respond", "hi", "message to respond with")
 
 	silent := flag.Bool("silent", false, "disable all logging")
 
+	ignoreSigTerm := flag.Bool("ignore-sig-term", false, "ignore SIGTERM signal")
+
 	flag.Parse() // Parse the command-line flags
 
 	// Create a new Gin router
@@ -31,19 +35,88 @@ func main() {
 
 	// Set up the handler function using the provided response message
 	r.POST("/v1/chat/completions", func(c *gin.Context) {
-		c.Header("Content-Type", "text/plain")
+		c.Header("Content-Type", "application/json")
 
 		// add a wait to simulate a slow query
 		if wait, err := time.ParseDuration(c.Query("wait")); err == nil {
 			time.Sleep(wait)
 		}
 
-		c.String(200, *responseMessage)
+		c.JSON(http.StatusOK, gin.H{
+			"responseMessage":  *responseMessage,
+			"h_content_length": c.Request.Header.Get("Content-Length"),
+		})
+	})
+
+	// for issue #62 to check model name strips profile slug
+	// has to be one of the openAI API endpoints that llama-swap proxies
+	// curl http://localhost:8080/v1/audio/speech -d '{"model":"profile:TheExpectedModel"}'
+	r.POST("/v1/audio/speech", func(c *gin.Context) {
+		body, err := io.ReadAll(c.Request.Body)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to read request body"})
+			return
+		}
+		defer c.Request.Body.Close()
+		modelName := gjson.GetBytes(body, "model").String()
+		if modelName != *expectedModel {
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Invalid model: %s, expected: %s", modelName, *expectedModel)})
+			return
+		} else {
+			c.JSON(http.StatusOK, gin.H{"message": "ok"})
+		}
 	})
 
 	r.POST("/v1/completions", func(c *gin.Context) {
-		c.Header("Content-Type", "text/plain")
-		c.String(200, *responseMessage)
+		c.Header("Content-Type", "application/json")
+		c.JSON(http.StatusOK, gin.H{
+			"responseMessage": *responseMessage,
+		})
+
+	})
+
+	// issue #41
+	r.POST("/v1/audio/transcriptions", func(c *gin.Context) {
+		// Parse the multipart form
+		if err := c.Request.ParseMultipartForm(10 << 20); err != nil { // 10 MB max memory
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Error parsing multipart form: %s", err)})
+			return
+		}
+
+		// Get the model from the form values
+		model := c.Request.FormValue("model")
+
+		if model == "" {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "Missing model parameter"})
+			return
+		}
+
+		// Get the file from the form
+		file, _, err := c.Request.FormFile("file")
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Error getting file: %s", err)})
+			return
+		}
+		defer file.Close()
+
+		// Read the file content to get its size
+		fileBytes, err := io.ReadAll(file)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("Error reading file: %s", err)})
+			return
+		}
+
+		fileSize := len(fileBytes)
+
+		// Return a JSON response with the model and transcription text including file size
+		c.JSON(http.StatusOK, gin.H{
+			"text":  fmt.Sprintf("The length of the file is %d bytes", fileSize),
+			"model": model,
+
+			// expose some header values for testing
+			"h_content_type":   c.GetHeader("Content-Type"),
+			"h_content_length": c.GetHeader("Content-Length"),
+		})
 	})
 
 	r.GET("/slow-respond", func(c *gin.Context) {
@@ -119,6 +192,10 @@ func main() {
 		log.SetOutput(io.Discard)
 	}
 
+	if !*silent {
+		fmt.Printf("My PID: %d\n", os.Getpid())
+	}
+
 	go func() {
 		log.Printf("simple-responder listening on %s\n", address)
 		// service connections
@@ -129,11 +206,36 @@ func main() {
 
 	// Wait for interrupt signal to gracefully shutdown the server with
 	// a timeout of 5 seconds.
-	quit := make(chan os.Signal, 1)
+	sigChan := make(chan os.Signal, 1)
 	// kill (no param) default send syscall.SIGTERM
 	// kill -2 is syscall.SIGINT
 	// kill -9 is syscall.SIGKILL but can't be catch, so don't need add it
-	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
-	<-quit
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+
+	countSigInt := 0
+
+runloop:
+	for {
+		signal := <-sigChan
+		switch signal {
+		case syscall.SIGINT:
+			countSigInt++
+			if countSigInt > 1 {
+				break runloop
+			} else {
+				log.Println("Recieved SIGINT, send another SIGINT to shutdown")
+			}
+		case syscall.SIGTERM:
+			if *ignoreSigTerm {
+				log.Println("Ignoring SIGTERM")
+			} else {
+				log.Println("Recieved SIGTERM, shutting down")
+				break runloop
+			}
+		default:
+			break runloop
+		}
+	}
+
 	log.Println("simple-responder shutting down")
 }

misc/test-rerank/README.md

@@ -0,0 +1,4 @@
+The rerank-test.json data is from https://github.com/ggerganov/llama.cpp/pull/9510
+
+To run it:
+> curl http://127.0.0.1:8080/v1/rerank -H "Content-Type: application/json" -d @reranker-test.json  -v | jq .

misc/test-rerank/reranker-test.json

@@ -0,0 +1,17 @@
+{
+  "model": "bge-reranker",
+  "query": "Organic skincare products for sensitive skin",
+  "top_n": 3,
+  "documents": [
+    "Organic skincare for sensitive skin with aloe vera and chamomile: Imagine the soothing embrace of nature with our organic skincare range, crafted specifically for sensitive skin. Infused with the calming properties of aloe vera and chamomile, each product provides gentle nourishment and protection. Say goodbye to irritation and hello to a glowing, healthy complexion.",
+    "New makeup trends focus on bold colors and innovative techniques: Step into the world of cutting-edge beauty with this seasons makeup trends. Bold, vibrant colors and groundbreaking techniques are redefining the art of makeup. From neon eyeliners to holographic highlighters, unleash your creativity and make a statement with every look.",
+    "Bio-Hautpflege für empfindliche Haut mit Aloe Vera und Kamille: Erleben Sie die wohltuende Wirkung unserer Bio-Hautpflege, speziell für empfindliche Haut entwickelt. Mit den beruhigenden Eigenschaften von Aloe Vera und Kamille pflegen und schützen unsere Produkte Ihre Haut auf natürliche Weise. Verabschieden Sie sich von Hautirritationen und genießen Sie einen strahlenden Teint.",
+    "Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken: Tauchen Sie ein in die Welt der modernen Schönheit mit den neuesten Make-up-Trends. Kräftige, lebendige Farben und innovative Techniken setzen neue Maßstäbe. Von auffälligen Eyelinern bis hin zu holografischen Highlightern – lassen Sie Ihrer Kreativität freien Lauf und setzen Sie jedes Mal ein Statement.",
+    "Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla: Descubre el poder de la naturaleza con nuestra línea de cuidado de la piel orgánico, diseñada especialmente para pieles sensibles. Enriquecidos con aloe vera y manzanilla, estos productos ofrecen una hidratación y protección suave. Despídete de las irritaciones y saluda a una piel radiante y saludable.",
+    "Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras: Entra en el fascinante mundo del maquillaje con las tendencias más actuales. Colores vivos y técnicas innovadoras están revolucionando el arte del maquillaje. Desde delineadores neón hasta iluminadores holográficos, desata tu creatividad y destaca en cada look.",
+    "针对敏感肌专门设计的天然有机护肤产品：体验由芦荟和洋甘菊提取物带来的自然呵护。我们的护肤产品特别为敏感肌设计，温和滋润，保护您的肌肤不受刺激。让您的肌肤告别不适，迎来健康光彩。",
+    "新的化妆趋势注重鲜艳的颜色和创新的技巧：进入化妆艺术的新纪元，本季的化妆趋势以大胆的颜色和创新的技巧为主。无论是霓虹眼线还是全息高光，每一款妆容都能让您脱颖而出，展现独特魅力。",
+    "敏感肌のために特別に設計された天然有機スキンケア製品: アロエベラとカモミールのやさしい力で、自然の抱擁を感じてください。敏感肌用に特別に設計された私たちのスキンケア製品は、肌に優しく栄養を与え、保護します。肌トラブルにさようなら、輝く健康な肌にこんにちは。",
+    "新しいメイクのトレンドは鮮やかな色と革新的な技術に焦点を当てています: 今シーズンのメイクアップトレンドは、大胆な色彩と革新的な技術に注目しています。ネオンアイライナーからホログラフィックハイライターまで、クリエイティビティを解き放ち、毎回ユニークなルックを演出しましょう。"
+  ]
+}

proxy/config.go

@@ -2,32 +2,76 @@ package proxy
 
 import (
 	"fmt"
+	"io"
 	"os"
+	"runtime"
+	"sort"
+	"strconv"
 	"strings"
 
+	"github.com/billziss-gh/golib/shlex"
 	"gopkg.in/yaml.v3"
 )
 
+const DEFAULT_GROUP_ID = "(default)"
+
 type ModelConfig struct {
 	Cmd           string   `yaml:"cmd"`
+	CmdStop       string   `yaml:"cmdStop"`
 	Proxy         string   `yaml:"proxy"`
 	Aliases       []string `yaml:"aliases"`
 	Env           []string `yaml:"env"`
 	CheckEndpoint string   `yaml:"checkEndpoint"`
 	UnloadAfter   int      `yaml:"ttl"`
+	Unlisted      bool     `yaml:"unlisted"`
+	UseModelName  string   `yaml:"useModelName"`
+
+	// Limit concurrency of HTTP requests to process
+	ConcurrencyLimit int `yaml:"concurrencyLimit"`
 }
 
 func (m *ModelConfig) SanitizedCommand() ([]string, error) {
 	return SanitizeCommand(m.Cmd)
 }
 
+type GroupConfig struct {
+	Swap       bool     `yaml:"swap"`
+	Exclusive  bool     `yaml:"exclusive"`
+	Persistent bool     `yaml:"persistent"`
+	Members    []string `yaml:"members"`
+}
+
+// set default values for GroupConfig
+func (c *GroupConfig) UnmarshalYAML(unmarshal func(interface{}) error) error {
+	type rawGroupConfig GroupConfig
+	defaults := rawGroupConfig{
+		Swap:       true,
+		Exclusive:  true,
+		Persistent: false,
+		Members:    []string{},
+	}
+
+	if err := unmarshal(&defaults); err != nil {
+		return err
+	}
+
+	*c = GroupConfig(defaults)
+	return nil
+}
+
 type Config struct {
 	HealthCheckTimeout int                    `yaml:"healthCheckTimeout"`
-	Models             map[string]ModelConfig `yaml:"models"`
+	LogRequests        bool                   `yaml:"logRequests"`
+	LogLevel           string                 `yaml:"logLevel"`
+	Models             map[string]ModelConfig `yaml:"models"` /* key is model ID */
 	Profiles           map[string][]string    `yaml:"profiles"`
+	Groups             map[string]GroupConfig `yaml:"groups"` /* key is group ID */
 
 	// map aliases to actual model IDs
 	aliases map[string]string
+
+	// automatic port assignments
+	StartPort int `yaml:"startPort"`
 }
 
 func (c *Config) RealModelName(search string) (string, bool) {
@@ -48,40 +92,169 @@ func (c *Config) FindConfig(modelName string) (ModelConfig, string, bool) {
 	}
 }
 
-func LoadConfig(path string) (*Config, error) {
-	data, err := os.ReadFile(path)
+func LoadConfig(path string) (Config, error) {
+	file, err := os.Open(path)
 	if err != nil {
-		return nil, err
+		return Config{}, err
+	}
+	defer file.Close()
+	return LoadConfigFromReader(file)
+}
+
+func LoadConfigFromReader(r io.Reader) (Config, error) {
+	data, err := io.ReadAll(r)
+	if err != nil {
+		return Config{}, err
 	}
 
 	var config Config
 	err = yaml.Unmarshal(data, &config)
 	if err != nil {
-		return nil, err
+		return Config{}, err
 	}
 
 	if config.HealthCheckTimeout < 15 {
 		config.HealthCheckTimeout = 15
 	}
 
+	// set default port ranges
+	if config.StartPort == 0 {
+		// default to 5800
+		config.StartPort = 5800
+	} else if config.StartPort < 1 {
+		return Config{}, fmt.Errorf("startPort must be greater than 1")
+	}
+
 	// Populate the aliases map
 	config.aliases = make(map[string]string)
 	for modelName, modelConfig := range config.Models {
 		for _, alias := range modelConfig.Aliases {
+			if _, found := config.aliases[alias]; found {
+				return Config{}, fmt.Errorf("duplicate alias %s found in model: %s", alias, modelName)
+			}
 			config.aliases[alias] = modelName
 		}
 	}
 
-	return &config, nil
+	// Get and sort all model IDs first, makes testing more consistent
+	modelIds := make([]string, 0, len(config.Models))
+	for modelId := range config.Models {
+		modelIds = append(modelIds, modelId)
+	}
+	sort.Strings(modelIds) // This guarantees stable iteration order
+
+	nextPort := config.StartPort
+	for _, modelId := range modelIds {
+		modelConfig := config.Models[modelId]
+		// iterate over the models and replace any ${PORT} with the next available port
+		if strings.Contains(modelConfig.Cmd, "${PORT}") {
+			modelConfig.Cmd = strings.ReplaceAll(modelConfig.Cmd, "${PORT}", strconv.Itoa(nextPort))
+			if modelConfig.Proxy == "" {
+				modelConfig.Proxy = fmt.Sprintf("http://localhost:%d", nextPort)
+			} else {
+				modelConfig.Proxy = strings.ReplaceAll(modelConfig.Proxy, "${PORT}", strconv.Itoa(nextPort))
+			}
+			nextPort++
+			config.Models[modelId] = modelConfig
+		} else if modelConfig.Proxy == "" {
+			return Config{}, fmt.Errorf("model %s requires a proxy value when not using automatic ${PORT}", modelId)
+		}
+	}
+
+	config = AddDefaultGroupToConfig(config)
+	// check that members are all unique in the groups
+	memberUsage := make(map[string]string) // maps member to group it appears in
+	for groupID, groupConfig := range config.Groups {
+		prevSet := make(map[string]bool)
+		for _, member := range groupConfig.Members {
+			// Check for duplicates within this group
+			if _, found := prevSet[member]; found {
+				return Config{}, fmt.Errorf("duplicate model member %s found in group: %s", member, groupID)
+			}
+			prevSet[member] = true
+
+			// Check if member is used in another group
+			if existingGroup, exists := memberUsage[member]; exists {
+				return Config{}, fmt.Errorf("model member %s is used in multiple groups: %s and %s", member, existingGroup, groupID)
+			}
+			memberUsage[member] = groupID
+		}
+	}
+
+	return config, nil
+}
+
+// rewrites the yaml to include a default group with any orphaned models
+func AddDefaultGroupToConfig(config Config) Config {
+
+	if config.Groups == nil {
+		config.Groups = make(map[string]GroupConfig)
+	}
+
+	defaultGroup := GroupConfig{
+		Swap:      true,
+		Exclusive: true,
+		Members:   []string{},
+	}
+	// if groups is empty, create a default group and put
+	// all models into it
+	if len(config.Groups) == 0 {
+		for modelName := range config.Models {
+			defaultGroup.Members = append(defaultGroup.Members, modelName)
+		}
+	} else {
+		// iterate over existing group members and add non-grouped models into the default group
+		for modelName, _ := range config.Models {
+			foundModel := false
+		found:
+			// search for the model in existing groups
+			for _, groupConfig := range config.Groups {
+				for _, member := range groupConfig.Members {
+					if member == modelName {
+						foundModel = true
+						break found
+					}
+				}
+			}
+
+			if !foundModel {
+				defaultGroup.Members = append(defaultGroup.Members, modelName)
+			}
+		}
+	}
+
+	sort.Strings(defaultGroup.Members) // make consistent ordering for testing
+	config.Groups[DEFAULT_GROUP_ID] = defaultGroup
+
+	return config
 }
 
 func SanitizeCommand(cmdStr string) ([]string, error) {
-	// Remove trailing backslashes
-	cmdStr = strings.ReplaceAll(cmdStr, "\\ \n", " ")
-	cmdStr = strings.ReplaceAll(cmdStr, "\\\n", " ")
+	var cleanedLines []string
+	for _, line := range strings.Split(cmdStr, "\n") {
+		trimmed := strings.TrimSpace(line)
+		// Skip comment lines
+		if strings.HasPrefix(trimmed, "#") {
+			continue
+		}
+		// Handle trailing backslashes by replacing with space
+		if strings.HasSuffix(trimmed, "\\") {
+			cleanedLines = append(cleanedLines, strings.TrimSuffix(trimmed, "\\")+" ")
+		} else {
+			cleanedLines = append(cleanedLines, line)
+		}
+	}
+
+	// put it back together
+	cmdStr = strings.Join(cleanedLines, "\n")
 
 	// Split the command into arguments
-	args := strings.Fields(cmdStr)
+	var args []string
+	if runtime.GOOS == "windows" {
+		args = shlex.Windows.Split(cmdStr)
+	} else {
+		args = shlex.Posix.Split(cmdStr)
+	}
 
 	// Ensure the command is not empty
 	if len(args) == 0 {

proxy/config_posix_test.go

@@ -0,0 +1,42 @@
+//go:build !windows
+
+package proxy
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_SanitizeCommand(t *testing.T) {
+	// Test a command with spaces and newlines
+	args, err := SanitizeCommand(`python model1.py \
+		-a "double quotes" \
+		--arg2 'single quotes'
+		-s
+		# comment 1
+		--arg3 123 \
+
+		  # comment 2
+		--arg4 '"string in string"'
+
+
+		# this will get stripped out as well as the white space above
+		-c "'single quoted'"
+		`)
+	assert.NoError(t, err)
+	assert.Equal(t, []string{
+		"python", "model1.py",
+		"-a", "double quotes",
+		"--arg2", "single quotes",
+		"-s",
+		"--arg3", "123",
+		"--arg4", `"string in string"`,
+		"-c", `'single quoted'`,
+	}, args)
+
+	// Test an empty command
+	args, err = SanitizeCommand("")
+	assert.Error(t, err)
+	assert.Nil(t, args)
+}

proxy/config_test.go

@@ -3,6 +3,7 @@ package proxy
 import (
 	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -35,11 +36,32 @@ models:
     aliases:
       - "m2"
     checkEndpoint: "/"
+  model3:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    aliases:
+      - "mthree"
+    checkEndpoint: "/"
+  model4:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8082"
+    checkEndpoint: "/"
+
 healthCheckTimeout: 15
 profiles:
   test:
     - model1
     - model2
+groups:
+  group1:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+  forever:
+    exclusive: false
+    persistent: true
+    members:
+      - "model4"
 `
 
 	if err := os.WriteFile(tempFile, []byte(content), 0644); err != nil {
@@ -52,7 +74,8 @@ profiles:
 		t.Fatalf("Failed to load config: %v", err)
 	}
 
-	expected := &Config{
+	expected := Config{
+		StartPort: 5800,
 		Models: map[string]ModelConfig{
 			"model1": {
 				Cmd:           "path/to/cmd --arg1 one",
@@ -68,6 +91,18 @@ profiles:
 				Env:           nil,
 				CheckEndpoint: "/",
 			},
+			"model3": {
+				Cmd:           "path/to/cmd --arg1 one",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"mthree"},
+				Env:           nil,
+				CheckEndpoint: "/",
+			},
+			"model4": {
+				Cmd:           "path/to/cmd --arg1 one",
+				Proxy:         "http://localhost:8082",
+				CheckEndpoint: "/",
+			},
 		},
 		HealthCheckTimeout: 15,
 		Profiles: map[string][]string{
@@ -77,6 +112,25 @@ profiles:
 			"m1":        "model1",
 			"model-one": "model1",
 			"m2":        "model2",
+			"mthree":    "model3",
+		},
+		Groups: map[string]GroupConfig{
+			DEFAULT_GROUP_ID: {
+				Swap:      true,
+				Exclusive: true,
+				Members:   []string{"model1", "model3"},
+			},
+			"group1": {
+				Swap:      true,
+				Exclusive: false,
+				Members:   []string{"model2"},
+			},
+			"forever": {
+				Swap:       true,
+				Exclusive:  false,
+				Persistent: true,
+				Members:    []string{"model4"},
+			},
 		},
 	}
 
@@ -87,6 +141,63 @@ profiles:
 	assert.Equal(t, "model1", realname)
 }
 
+func TestConfig_GroupMemberIsUnique(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+  model2:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+  model3:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+
+healthCheckTimeout: 15
+groups:
+  group1:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+  group2:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+`
+	// Load the config and verify
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+
+	// a Contains as order of the map is not guaranteed
+	assert.Contains(t, err.Error(), "model member model2 is used in multiple groups:")
+}
+
+func TestConfig_ModelAliasesAreUnique(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+    aliases:
+      - m1
+  model2:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+    aliases:
+      - m1
+      - m2
+`
+	// Load the config and verify
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+
+	// this is a contains because it could be `model1` or `model2` depending on the order
+	// go decided on the order of the map
+	assert.Contains(t, err.Error(), "duplicate alias m1 found in model: model")
+}
+
 func TestConfig_ModelConfigSanitizedCommand(t *testing.T) {
 	config := &ModelConfig{
 		Cmd: `python model1.py \
@@ -147,21 +258,76 @@ func TestConfig_FindConfig(t *testing.T) {
 	assert.Equal(t, ModelConfig{}, modelConfig)
 }
 
-func TestConfig_SanitizeCommand(t *testing.T) {
-	// Test a simple command
-	args, err := SanitizeCommand("python model1.py")
-	assert.NoError(t, err)
-	assert.Equal(t, []string{"python", "model1.py"}, args)
+func TestConfig_AutomaticPortAssignments(t *testing.T) {
 
-	// Test a command with spaces and newlines
-	args, err = SanitizeCommand(`python model1.py \
-    --arg1 value1 \
-    --arg2 value2`)
-	assert.NoError(t, err)
-	assert.Equal(t, []string{"python", "model1.py", "--arg1", "value1", "--arg2", "value2"}, args)
+	t.Run("Default Port Ranges", func(t *testing.T) {
+		content := ``
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
 
-	// Test an empty command
-	args, err = SanitizeCommand("")
-	assert.Error(t, err)
-	assert.Nil(t, args)
+		assert.Equal(t, 5800, config.StartPort)
+	})
+	t.Run("User specific port ranges", func(t *testing.T) {
+		content := `startPort: 1000`
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
+
+		assert.Equal(t, 1000, config.StartPort)
+	})
+
+	t.Run("Invalid start port", func(t *testing.T) {
+		content := `startPort: abcd`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.NotNil(t, err)
+	})
+
+	t.Run("start port must be greater than 1", func(t *testing.T) {
+		content := `startPort: -99`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.NotNil(t, err)
+	})
+
+	t.Run("Automatic port assignments", func(t *testing.T) {
+		content := `
+startPort: 5800
+models:
+  model1:
+    cmd: svr --port ${PORT}
+  model2:
+    cmd: svr --port ${PORT}
+    proxy: "http://172.11.22.33:${PORT}"
+  model3:
+    cmd: svr --port 1999
+    proxy: "http://1.2.3.4:1999"
+`
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
+
+		assert.Equal(t, 5800, config.StartPort)
+		assert.Equal(t, "svr --port 5800", config.Models["model1"].Cmd)
+		assert.Equal(t, "http://localhost:5800", config.Models["model1"].Proxy)
+
+		assert.Equal(t, "svr --port 5801", config.Models["model2"].Cmd)
+		assert.Equal(t, "http://172.11.22.33:5801", config.Models["model2"].Proxy)
+
+		assert.Equal(t, "svr --port 1999", config.Models["model3"].Cmd)
+		assert.Equal(t, "http://1.2.3.4:1999", config.Models["model3"].Proxy)
+
+	})
+
+	t.Run("Proxy value required if no ${PORT} in cmd", func(t *testing.T) {
+		content := `
+models:
+  model1:
+    cmd: svr --port 111
+`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.Equal(t, "model model1 requires a proxy value when not using automatic ${PORT}", err.Error())
+	})
 }

proxy/config_windows_test.go

@@ -0,0 +1,41 @@
+//go:build windows
+
+package proxy
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_SanitizeCommand(t *testing.T) {
+	// does not support single quoted strings like in config_posix_test.go
+	args, err := SanitizeCommand(`python model1.py \
+
+	-a "double quotes" \
+	-s
+	--arg3 123 \
+
+	   # comment 2
+	--arg4 '"string in string"'
+
+
+
+	# this will get stripped out as well as the white space above
+	-c "'single quoted'"
+	`)
+	assert.NoError(t, err)
+	assert.Equal(t, []string{
+		"python", "model1.py",
+		"-a", "double quotes",
+		"-s",
+		"--arg3", "123",
+		"--arg4", "'string in string'", // this is a little weird but the lexer says so...?
+		"-c", `'single quoted'`,
+	}, args)
+
+	// Test an empty command
+	args, err = SanitizeCommand("")
+	assert.Error(t, err)
+	assert.Nil(t, args)
+}

proxy/helpers_test.go

@@ -14,6 +14,7 @@ import (
 var (
 	nextTestPort int = 12000
 	portMutex    sync.Mutex
+	testLogger   = NewLogMonitorWriter(os.Stdout)
 )
 
 // Check if the binary exists
@@ -26,6 +27,17 @@ func TestMain(m *testing.M) {
 
 	gin.SetMode(gin.TestMode)
 
+	switch os.Getenv("LOG_LEVEL") {
+	case "debug":
+		testLogger.SetLogLevel(LevelDebug)
+	case "warn":
+		testLogger.SetLogLevel(LevelWarn)
+	case "info":
+		testLogger.SetLogLevel(LevelInfo)
+	default:
+		testLogger.SetLogLevel(LevelWarn)
+	}
+
 	m.Run()
 }
 
@@ -33,17 +45,26 @@ func TestMain(m *testing.M) {
 func getSimpleResponderPath() string {
 	goos := runtime.GOOS
 	goarch := runtime.GOARCH
-	return filepath.Join("..", "build", fmt.Sprintf("simple-responder_%s_%s", goos, goarch))
+
+	if goos == "windows" {
+		return filepath.Join("..", "build", "simple-responder.exe")
+	} else {
+		return filepath.Join("..", "build", fmt.Sprintf("simple-responder_%s_%s", goos, goarch))
+	}
 }
 
-func getTestSimpleResponderConfig(expectedMessage string) ModelConfig {
+func getTestPort() int {
 	portMutex.Lock()
 	defer portMutex.Unlock()
 
 	port := nextTestPort
 	nextTestPort++
 
-	return getTestSimpleResponderConfigPort(expectedMessage, port)
+	return port
+}
+
+func getTestSimpleResponderConfig(expectedMessage string) ModelConfig {
+	return getTestSimpleResponderConfigPort(expectedMessage, getTestPort())
 }
 
 func getTestSimpleResponderConfigPort(expectedMessage string, port int) ModelConfig {

proxy/html/index.html

@@ -0,0 +1,14 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>llama-swap</title>
+</head>
+<body>
+    <h1>llama-swap</h1>
+    <p>
+        <a href="/logs">view logs</a> | <a href="/upstream">configured models</a> | <a href="https://github.com/mostlygeek/llama-swap">github</a>
+    </p>
+</body>
+</html>

proxy/html/logs.html

@@ -0,0 +1,259 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Logs</title>
+    <style>
+        body {
+            margin: 0;
+            height: 100vh;
+            display: flex;
+            flex-direction: column;
+            font-family: "Courier New", Courier, monospace;
+        }
+        .log-container {
+            display: flex;
+            flex: 1;
+            gap: 0.5em;
+            margin: 0.5em;
+            min-height: 0;
+        }
+        .log-column {
+            display: flex;
+            flex-direction: column;
+            flex: 1;
+            min-width: 0;
+            transition: flex 0.3s ease;
+        }
+        .log-column.minimized {
+            flex: 0.1;
+            max-width: 50px;
+            border: 1px solid #777;
+            color: green;
+        }
+        .log-controls {
+            display: grid;
+            grid-template-columns: 1fr auto;
+            gap: 0.5em;
+            margin-bottom: 0.5em;
+        }
+        .log-controls input {
+            width: 100%;
+            padding: 4px;
+        }
+        .log-controls input:focus {
+           outline: none;
+        }
+        .log-stream {
+            flex: 1;
+            padding: 1em;
+            background: #f4f4f4;
+            overflow-y: auto;
+            white-space: pre-wrap;
+            word-wrap: break-word;
+            min-height: 0;
+        }
+
+        .regex-error {
+            background-color: #ff0000 !important;
+        }
+
+        /* Make headers clickable and show pointer cursor */
+        h2 {
+            cursor: pointer;
+            user-select: none;
+            margin: 0 0 0.5em 0;
+            padding: 0.5em;
+        }
+
+        h2:hover {
+            background-color: rgba(0, 0, 0, 0.05);
+        }
+
+        /* Dark mode styles */
+        @media (prefers-color-scheme: dark) {
+            body {
+                background-color: #333;
+                color: #fff;
+            }
+
+            .log-stream {
+                background: #444;
+                color: #fff;
+            }
+
+            .log-controls input {
+                background: #555;
+                color: #fff;
+                border: 1px solid #777;
+            }
+
+            .log-controls button {
+                background: #555;
+                color: #fff;
+                border: 1px solid #777;
+            }
+
+            h2:hover {
+                background-color: rgba(255, 255, 255, 0.1);
+            }
+        }
+
+        /* Hide content when minimized */
+        .log-column.minimized .log-controls,
+        .log-column.minimized .log-stream {
+            display: none;
+        }
+
+        .log-column.minimized h2 {
+            writing-mode: vertical-rl;
+            text-orientation: mixed;
+            transform: rotate(180deg);
+            white-space: nowrap;
+            margin: auto;
+        }
+    </style>
+</head>
+<body>
+    <div class="log-container">
+        <div class="log-column">
+            <h2>Proxy Logs</h2>
+            <div class="log-controls">
+                <input type="text" id="proxy-filter-input" placeholder="proxy regex filter">
+                <button id="proxy-clear-button">clear</button>
+            </div>
+            <pre class="log-stream" id="proxy-log-stream">Waiting for proxy logs...</pre>
+        </div>
+        <div class="log-column minimized">
+            <h2>Upstream Logs</h2>
+            <div class="log-controls">
+                <input type="text" id="upstream-filter-input" placeholder="upstream regex filter">
+                <button id="upstream-clear-button">clear</button>
+            </div>
+            <pre class="log-stream" id="upstream-log-stream">Waiting for upstream logs...</pre>
+        </div>
+    </div>
+    <script>
+        class LogStream {
+            constructor(streamElement, filterInput, clearButton, endpoint) {
+                this.streamElement = streamElement;
+                this.filterInput = filterInput;
+                this.clearButton = clearButton;
+                this.endpoint = endpoint;
+                this.logData = "";
+                this.regexFilter = null;
+                this.eventSource = null;
+
+                this.initialize();
+            }
+
+            initialize() {
+                this.filterInput.addEventListener('input', () => this.updateFilter());
+                this.clearButton.addEventListener('click', () => {
+                    this.filterInput.value = "";
+                    this.regexFilter = null;
+                    this.render();
+                });
+                this.setupEventSource();
+            }
+
+            setupEventSource() {
+                if (typeof(EventSource) === "undefined") {
+                    this.logData = "SSE Not supported by this browser.";
+                    this.render();
+                    return;
+                }
+
+                const connect = () => {
+                    this.eventSource = new EventSource(this.endpoint);
+
+                    this.eventSource.onmessage = (event) => {
+                        this.logData += event.data;
+                        this.logData = this.logData.slice(-1024 * 100);
+                        this.render();
+                    };
+
+                    this.eventSource.onerror = (err) => {
+                        // Close the current connection
+                        this.eventSource.close();
+
+                        this.logData += "\nConnection lost. Retrying in 5 seconds...\n";
+                        this.render();
+
+                        // Attempt to reconnect after 5 seconds
+                        setTimeout(() => {
+                            this.logData += "Attempting to reconnect...\n";
+                            this.render();
+                            connect();
+                        }, 5000);
+                    };
+                };
+
+                // Initial connection
+                connect();
+            }
+
+            render() {
+                let content = this.logData;
+
+                if (this.regexFilter) {
+                    const lines = content.split('\n');
+                    const filteredLines = lines.filter(line => this.regexFilter.test(line));
+                    content = filteredLines.length > 0 ? filteredLines.join('\n') + '\n' : "";
+                }
+
+                this.streamElement.textContent = content;
+                this.streamElement.scrollTop = this.streamElement.scrollHeight;
+            }
+
+            updateFilter() {
+                const pattern = this.filterInput.value.trim();
+                this.filterInput.classList.remove('regex-error');
+
+                if (!pattern) {
+                    this.regexFilter = null;
+                    this.render();
+                    return;
+                }
+
+                try {
+                    this.regexFilter = new RegExp(pattern);
+                } catch (e) {
+                    console.error("Invalid regex pattern:", e);
+                    this.regexFilter = null;
+                    this.filterInput.classList.add('regex-error');
+                    return;
+                }
+
+                this.render();
+            }
+        }
+
+        // Initialize both log streams
+        document.addEventListener('DOMContentLoaded', () => {
+            new LogStream(
+                document.getElementById('proxy-log-stream'),
+                document.getElementById('proxy-filter-input'),
+                document.getElementById('proxy-clear-button'),
+                "/logs/streamSSE/proxy"
+            );
+
+            new LogStream(
+                document.getElementById('upstream-log-stream'),
+                document.getElementById('upstream-filter-input'),
+                document.getElementById('upstream-clear-button'),
+                "/logs/streamSSE/upstream"
+            );
+
+            // Initialize clickable headers
+            document.querySelectorAll('h2').forEach(header => {
+                header.addEventListener('click', () => {
+                    const column = header.closest('.log-column');
+                    column.classList.toggle('minimized');
+                });
+            });
+        });
+    </script>
+</body>
+</html>

proxy/html_files.go

@@ -0,0 +1,10 @@
+package proxy
+
+import "embed"
+
+//go:embed html
+var htmlFiles embed.FS
+
+func getHTMLFile(path string) ([]byte, error) {
+	return htmlFiles.ReadFile("html/" + path)
+}

proxy/logMonitor.go

@@ -2,11 +2,21 @@ package proxy
 
 import (
 	"container/ring"
+	"fmt"
 	"io"
 	"os"
 	"sync"
 )
 
+type LogLevel int
+
+const (
+	LevelDebug LogLevel = iota
+	LevelInfo
+	LevelWarn
+	LevelError
+)
+
 type LogMonitor struct {
 	clients  map[chan []byte]bool
 	mu       sync.RWMutex
@@ -15,6 +25,10 @@ type LogMonitor struct {
 
 	// typically this can be os.Stdout
 	stdout io.Writer
+
+	// logging levels
+	level  LogLevel
+	prefix string
 }
 
 func NewLogMonitor() *LogMonitor {
@@ -26,6 +40,8 @@ func NewLogMonitorWriter(stdout io.Writer) *LogMonitor {
 		clients: make(map[chan []byte]bool),
 		buffer:  ring.New(10 * 1024), // keep 10KB of buffered logs
 		stdout:  stdout,
+		level:   LevelInfo,
+		prefix:  "",
 	}
 }
 
@@ -46,7 +62,7 @@ func (w *LogMonitor) Write(p []byte) (n int, err error) {
 	w.buffer = w.buffer.Next()
 	w.bufferMu.Unlock()
 
-	w.broadcast(p)
+	w.broadcast(bufferCopy)
 	return n, nil
 }
 
@@ -94,3 +110,77 @@ func (w *LogMonitor) broadcast(msg []byte) {
 		}
 	}
 }
+
+func (w *LogMonitor) SetPrefix(prefix string) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	w.prefix = prefix
+}
+
+func (w *LogMonitor) SetLogLevel(level LogLevel) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	w.level = level
+}
+
+func (w *LogMonitor) formatMessage(level string, msg string) []byte {
+	prefix := ""
+	if w.prefix != "" {
+		prefix = fmt.Sprintf("[%s] ", w.prefix)
+	}
+	return []byte(fmt.Sprintf("%s[%s] %s\n", prefix, level, msg))
+}
+
+func (w *LogMonitor) log(level LogLevel, msg string) {
+	if level < w.level {
+		return
+	}
+	w.Write(w.formatMessage(level.String(), msg))
+}
+
+func (w *LogMonitor) Debug(msg string) {
+	w.log(LevelDebug, msg)
+}
+
+func (w *LogMonitor) Info(msg string) {
+	w.log(LevelInfo, msg)
+}
+
+func (w *LogMonitor) Warn(msg string) {
+	w.log(LevelWarn, msg)
+}
+
+func (w *LogMonitor) Error(msg string) {
+	w.log(LevelError, msg)
+}
+
+func (w *LogMonitor) Debugf(format string, args ...interface{}) {
+	w.log(LevelDebug, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Infof(format string, args ...interface{}) {
+	w.log(LevelInfo, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Warnf(format string, args ...interface{}) {
+	w.log(LevelWarn, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Errorf(format string, args ...interface{}) {
+	w.log(LevelError, fmt.Sprintf(format, args...))
+}
+
+func (l LogLevel) String() string {
+	switch l {
+	case LevelDebug:
+		return "DEBUG"
+	case LevelInfo:
+		return "INFO"
+	case LevelWarn:
+		return "WARN"
+	case LevelError:
+		return "ERROR"
+	default:
+		return "UNKNOWN"
+	}
+}

proxy/process.go

@@ -8,6 +8,8 @@ import (
 	"net/http"
 	"net/url"
 	"os/exec"
+	"runtime"
+	"strconv"
 	"strings"
 	"sync"
 	"syscall"
@@ -17,19 +19,38 @@ import (
 type ProcessState string
 
 const (
-	StateStopped ProcessState = ProcessState("stopped")
-	StateReady   ProcessState = ProcessState("ready")
-	StateFailed  ProcessState = ProcessState("failed")
+	StateStopped  ProcessState = ProcessState("stopped")
+	StateStarting ProcessState = ProcessState("starting")
+	StateReady    ProcessState = ProcessState("ready")
+	StateStopping ProcessState = ProcessState("stopping")
+
+	// failed a health check on start and will not be recovered
+	StateFailed ProcessState = ProcessState("failed")
+
+	// process is shutdown and will not be restarted
+	StateShutdown ProcessState = ProcessState("shutdown")
+)
+
+type StopStrategy int
+
+const (
+	StopImmediately StopStrategy = iota
+	StopWaitForInflightRequest
 )
 
 type Process struct {
-	sync.Mutex
+	ID     string
+	config ModelConfig
+	cmd    *exec.Cmd
 
-	ID                 string
-	config             ModelConfig
-	cmd                *exec.Cmd
-	logMonitor         *LogMonitor
-	healthCheckTimeout int
+	// for p.cmd.Wait() select { ... }
+	cmdWaitChan chan error
+
+	processLogger *LogMonitor
+	proxyLogger   *LogMonitor
+
+	healthCheckTimeout      int
+	healthCheckLoopInterval time.Duration
 
 	lastRequestHandled time.Time
 
@@ -37,155 +58,101 @@ type Process struct {
 	state      ProcessState
 
 	inFlightRequests sync.WaitGroup
+
+	// used to block on multiple start() calls
+	waitStarting sync.WaitGroup
+
+	// for managing shutdown state
+	shutdownCtx    context.Context
+	shutdownCancel context.CancelFunc
+
+	// for managing concurrency limits
+	concurrencyLimitSemaphore chan struct{}
+
+	// stop timeout waiting for graceful shutdown
+	gracefulStopTimeout time.Duration
+
+	// track that this happened
+	upstreamWasStoppedWithKill bool
 }
 
-func NewProcess(ID string, healthCheckTimeout int, config ModelConfig, logMonitor *LogMonitor) *Process {
+func NewProcess(ID string, healthCheckTimeout int, config ModelConfig, processLogger *LogMonitor, proxyLogger *LogMonitor) *Process {
+	ctx, cancel := context.WithCancel(context.Background())
+	concurrentLimit := 10
+	if config.ConcurrencyLimit > 0 {
+		concurrentLimit = config.ConcurrencyLimit
+	} else {
+		proxyLogger.Debugf("Concurrency limit for model %s not set, defaulting to 10", ID)
+	}
 	return &Process{
-		ID:                 ID,
-		config:             config,
-		cmd:                nil,
-		logMonitor:         logMonitor,
-		healthCheckTimeout: healthCheckTimeout,
-		state:              StateStopped,
+		ID:                      ID,
+		config:                  config,
+		cmd:                     nil,
+		cmdWaitChan:             make(chan error, 1),
+		processLogger:           processLogger,
+		proxyLogger:             proxyLogger,
+		healthCheckTimeout:      healthCheckTimeout,
+		healthCheckLoopInterval: 5 * time.Second, /* default, can not be set by user - used for testing */
+		state:                   StateStopped,
+		shutdownCtx:             ctx,
+		shutdownCancel:          cancel,
+
+		// concurrency limit
+		concurrencyLimitSemaphore: make(chan struct{}, concurrentLimit),
+
+		// stop timeout
+		gracefulStopTimeout:        5 * time.Second,
+		upstreamWasStoppedWithKill: false,
 	}
 }
 
-// start the process and returns when it is ready
-func (p *Process) start() error {
-
-	p.stateMutex.Lock()
-	defer p.stateMutex.Unlock()
-
-	if p.state == StateReady {
-		return nil
-	}
-
-	if p.state == StateFailed {
-		return fmt.Errorf("process is in a failed state and can not be restarted")
-	}
-
-	args, err := p.config.SanitizedCommand()
-	if err != nil {
-		return fmt.Errorf("unable to get sanitized command: %v", err)
-	}
-
-	p.cmd = exec.Command(args[0], args[1:]...)
-	p.cmd.Stdout = p.logMonitor
-	p.cmd.Stderr = p.logMonitor
-	p.cmd.Env = p.config.Env
-
-	err = p.cmd.Start()
-
-	if err != nil {
-		return err
-	}
-
-	// One of three things can happen at this stage:
-	// 1. The command exits unexpectedly
-	// 2. The health check fails
-	// 3. The health check passes
-	//
-	// only in the third case will the process be considered Ready to accept
-	healthCheckContext, cancelHealthCheck := context.WithCancelCause(context.Background())
-	defer cancelHealthCheck(nil) // clean up
-	cmdWaitChan := make(chan error, 1)
-	healthCheckChan := make(chan error, 1)
-
-	go func() {
-		// possible cmd exits early
-		cmdWaitChan <- p.cmd.Wait()
-	}()
-
-	go func() {
-		<-time.After(250 * time.Millisecond) // give process a bit of time to start
-		healthCheckChan <- p.checkHealthEndpoint(healthCheckContext)
-	}()
-
-	select {
-	case err := <-cmdWaitChan:
-		p.state = StateFailed
-		if err != nil {
-			err = fmt.Errorf("command [%s] %s", strings.Join(p.cmd.Args, " "), err.Error())
-		} else {
-			err = fmt.Errorf("command [%s] exited unexpected", strings.Join(p.cmd.Args, " "))
-		}
-		cancelHealthCheck(err)
-		return err
-	case err := <-healthCheckChan:
-		if err != nil {
-			p.state = StateFailed
-			return err
-		}
-	}
-
-	if p.config.UnloadAfter > 0 {
-		// start a goroutine to check every second if
-		// the process should be stopped
-		go func() {
-			ticker := time.NewTicker(time.Second)
-			defer ticker.Stop()
-			maxDuration := time.Duration(p.config.UnloadAfter) * time.Second
-
-			for {
-				<-ticker.C
-				if time.Since(p.lastRequestHandled) > maxDuration {
-					fmt.Fprintf(p.logMonitor, "!!! Unloading model %s, TTL of %d reached.\n", p.ID, p.config.UnloadAfter)
-					p.Stop()
-					return
-				}
-			}
-		}()
-	}
-
-	p.state = StateReady
-	return nil
+// LogMonitor returns the log monitor associated with the process.
+func (p *Process) LogMonitor() *LogMonitor {
+	return p.processLogger
 }
 
-func (p *Process) Stop() {
-	// wait for any inflight requests before proceeding
-	p.inFlightRequests.Wait()
+// custom error types for swapping state
+var (
+	ErrExpectedStateMismatch  = errors.New("expected state mismatch")
+	ErrInvalidStateTransition = errors.New("invalid state transition")
+)
 
+// swapState performs a compare and swap of the state atomically. It returns the current state
+// and an error if the swap failed.
+func (p *Process) swapState(expectedState, newState ProcessState) (ProcessState, error) {
 	p.stateMutex.Lock()
 	defer p.stateMutex.Unlock()
 
-	if p.state != StateReady {
-		return
+	if p.state != expectedState {
+		p.proxyLogger.Warnf("<%s> swapState() Unexpected current state %s, expected %s", p.ID, p.state, expectedState)
+		return p.state, ErrExpectedStateMismatch
 	}
 
-	if p.cmd == nil || p.cmd.Process == nil {
-		// this situation should never happen... but if it does just update the state
-		fmt.Fprintf(p.logMonitor, "!!! State is Ready but Command is nil.")
-		p.state = StateStopped
-		return
+	if !isValidTransition(p.state, newState) {
+		p.proxyLogger.Warnf("<%s> swapState() Invalid state transition from %s to %s", p.ID, p.state, newState)
+		return p.state, ErrInvalidStateTransition
 	}
 
-	// Pretty sure this stopping code needs some work for windows and
-	// will be a source of pain in the future.
+	p.state = newState
+	p.proxyLogger.Debugf("<%s> swapState() State transitioned from %s to %s", p.ID, expectedState, newState)
+	return p.state, nil
+}
 
-	p.cmd.Process.Signal(syscall.SIGTERM)
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-
-	done := make(chan error, 1)
-	go func() {
-		done <- p.cmd.Wait()
-	}()
-
-	select {
-	case <-ctx.Done():
-		fmt.Printf("!!! process for %s timed out waiting to stop\n", p.ID)
-		p.cmd.Process.Kill()
-		p.cmd.Wait()
-	case err := <-done:
-		if err != nil {
-			if err.Error() != "wait: no child processes" {
-				// possible that simple-responder for testing is just not
-				// existing right, so suppress those errors.
-				fmt.Printf("!!! process for %s stopped with error > %v\n", p.ID, err)
-			}
-		}
+// Helper function to encapsulate transition rules
+func isValidTransition(from, to ProcessState) bool {
+	switch from {
+	case StateStopped:
+		return to == StateStarting
+	case StateStarting:
+		return to == StateReady || to == StateFailed || to == StateStopping
+	case StateReady:
+		return to == StateStopping
+	case StateStopping:
+		return to == StateStopped || to == StateShutdown
+	case StateFailed, StateShutdown:
+		return false // No transitions allowed from these states
 	}
-	p.state = StateStopped
+	return false
 }
 
 func (p *Process) CurrentState() ProcessState {
@@ -194,107 +161,384 @@ func (p *Process) CurrentState() ProcessState {
 	return p.state
 }
 
-func (p *Process) checkHealthEndpoint(ctxFromStart context.Context) error {
+// start starts the upstream command, checks the health endpoint, and sets the state to Ready
+// it is a private method because starting is automatic but stopping can be called
+// at any time.
+func (p *Process) start() error {
+
 	if p.config.Proxy == "" {
-		return fmt.Errorf("no upstream available to check /health")
+		return fmt.Errorf("can not start(), upstream proxy missing")
 	}
 
+	args, err := p.config.SanitizedCommand()
+	if err != nil {
+		return fmt.Errorf("unable to get sanitized command: %v", err)
+	}
+
+	if curState, err := p.swapState(StateStopped, StateStarting); err != nil {
+		if err == ErrExpectedStateMismatch {
+			// already starting, just wait for it to complete and expect
+			// it to be be in the Ready start after. If not, return an error
+			if curState == StateStarting {
+				p.waitStarting.Wait()
+				if state := p.CurrentState(); state == StateReady {
+					return nil
+				} else {
+					return fmt.Errorf("process was already starting but wound up in state %v", state)
+				}
+			} else {
+				return fmt.Errorf("processes was in state %v when start() was called", curState)
+			}
+		} else {
+			return fmt.Errorf("failed to set Process state to starting: current state: %v, error: %v", curState, err)
+		}
+	}
+
+	p.waitStarting.Add(1)
+	defer p.waitStarting.Done()
+
+	p.cmd = exec.Command(args[0], args[1:]...)
+	p.cmd.Stdout = p.processLogger
+	p.cmd.Stderr = p.processLogger
+	p.cmd.Env = p.config.Env
+
+	err = p.cmd.Start()
+
+	// Set process state to failed
+	if err != nil {
+		if curState, swapErr := p.swapState(StateStarting, StateFailed); swapErr != nil {
+			return fmt.Errorf(
+				"failed to start command and state swap failed. command error: %v, current state: %v, state swap error: %v",
+				err, curState, swapErr,
+			)
+		}
+		return fmt.Errorf("start() failed: %v", err)
+	}
+
+	// Capture the exit error for later signalling
+	go func() {
+		exitErr := p.cmd.Wait()
+		p.proxyLogger.Debugf("<%s> cmd.Wait() returned error: %v", p.ID, exitErr)
+
+		// there is a race condition when SIGKILL is used, p.cmd.Wait() returns, and then
+		// the code below fires, putting an error into cmdWaitChan. This code is to prevent this
+		if p.upstreamWasStoppedWithKill {
+			p.proxyLogger.Debugf("<%s> process was killed, NOT sending exitErr: %v", p.ID, exitErr)
+			p.upstreamWasStoppedWithKill = false
+			return
+		}
+
+		p.cmdWaitChan <- exitErr
+	}()
+
+	// One of three things can happen at this stage:
+	// 1. The command exits unexpectedly
+	// 2. The health check fails
+	// 3. The health check passes
+	//
+	// only in the third case will the process be considered Ready to accept
+	<-time.After(250 * time.Millisecond) // give process a bit of time to start
+
+	checkStartTime := time.Now()
+	maxDuration := time.Second * time.Duration(p.healthCheckTimeout)
 	checkEndpoint := strings.TrimSpace(p.config.CheckEndpoint)
 
-	if checkEndpoint == "none" {
-		return nil
-	}
-
-	// keep default behaviour
-	if checkEndpoint == "" {
-		checkEndpoint = "/health"
-	}
-
-	proxyTo := p.config.Proxy
-	maxDuration := time.Second * time.Duration(p.healthCheckTimeout)
-	healthURL, err := url.JoinPath(proxyTo, checkEndpoint)
-	if err != nil {
-		return fmt.Errorf("failed to create health url with with %s and path %s", proxyTo, checkEndpoint)
-	}
-
-	client := &http.Client{}
-	startTime := time.Now()
-
-	for {
-		req, err := http.NewRequest("GET", healthURL, nil)
-		if err != nil {
-			return err
+	// a "none" means don't check for health ... I could have picked a better word :facepalm:
+	if checkEndpoint != "none" {
+		// keep default behaviour
+		if checkEndpoint == "" {
+			checkEndpoint = "/health"
 		}
 
-		ctx, cancel := context.WithTimeout(ctxFromStart, time.Second)
-		defer cancel()
-		req = req.WithContext(ctx)
-		resp, err := client.Do(req)
-
-		ttl := (maxDuration - time.Since(startTime)).Seconds()
-
+		proxyTo := p.config.Proxy
+		healthURL, err := url.JoinPath(proxyTo, checkEndpoint)
 		if err != nil {
-			// check if the context was cancelled
+			return fmt.Errorf("failed to create health check URL proxy=%s and checkEndpoint=%s", proxyTo, checkEndpoint)
+		}
+
+		checkDeadline, cancelHealthCheck := context.WithDeadline(
+			context.Background(),
+			checkStartTime.Add(maxDuration),
+		)
+		defer cancelHealthCheck()
+
+	loop:
+		// Ready Check loop
+		for {
 			select {
-			case <-ctx.Done():
-				err := context.Cause(ctx)
-				if !errors.Is(err, context.DeadlineExceeded) {
-					return err
+			case <-checkDeadline.Done():
+				if curState, err := p.swapState(StateStarting, StateFailed); err != nil {
+					return fmt.Errorf("health check timed out after %vs AND state swap failed: %v, current state: %v", maxDuration.Seconds(), err, curState)
+				} else {
+					return fmt.Errorf("health check timed out after %vs", maxDuration.Seconds())
+				}
+			case <-p.shutdownCtx.Done():
+				return errors.New("health check interrupted due to shutdown")
+			case exitErr := <-p.cmdWaitChan:
+				if exitErr != nil {
+					p.proxyLogger.Warnf("<%s> upstream command exited prematurely with error: %v", p.ID, exitErr)
+					if curState, err := p.swapState(StateStarting, StateFailed); err != nil {
+						return fmt.Errorf("upstream command exited unexpectedly: %s AND state swap failed: %v, current state: %v", exitErr.Error(), err, curState)
+					} else {
+						return fmt.Errorf("upstream command exited unexpectedly: %s", exitErr.Error())
+					}
+				} else {
+					p.proxyLogger.Warnf("<%s> upstream command exited prematurely but successfully", p.ID)
+					if curState, err := p.swapState(StateStarting, StateFailed); err != nil {
+						return fmt.Errorf("upstream command exited prematurely but successfully AND state swap failed: %v, current state: %v", err, curState)
+					} else {
+						return fmt.Errorf("upstream command exited prematurely but successfully")
+					}
 				}
 			default:
+				if err := p.checkHealthEndpoint(healthURL); err == nil {
+					p.proxyLogger.Infof("<%s> Health check passed on %s", p.ID, healthURL)
+					cancelHealthCheck()
+					break loop
+				} else {
+					if strings.Contains(err.Error(), "connection refused") {
+						endTime, _ := checkDeadline.Deadline()
+						ttl := time.Until(endTime)
+						p.proxyLogger.Debugf("<%s> Connection refused on %s, giving up in %.0fs (normal during startup)", p.ID, healthURL, ttl.Seconds())
+					} else {
+						p.proxyLogger.Debugf("<%s> Health check error on %s, %v (normal during startup)", p.ID, healthURL, err)
+					}
+				}
 			}
 
-			// wait a bit longer for TCP connection issues
-			if strings.Contains(err.Error(), "connection refused") {
-				fmt.Fprintf(p.logMonitor, "Connection refused on %s, ttl %.0fs\n", healthURL, ttl)
-				time.Sleep(5 * time.Second)
-			} else {
-				time.Sleep(time.Second)
+			<-time.After(p.healthCheckLoopInterval)
+		}
+	}
+
+	if p.config.UnloadAfter > 0 {
+		// start a goroutine to check every second if
+		// the process should be stopped
+		go func() {
+			maxDuration := time.Duration(p.config.UnloadAfter) * time.Second
+
+			for range time.Tick(time.Second) {
+				if p.CurrentState() != StateReady {
+					return
+				}
+
+				// wait for all inflight requests to complete and ticker
+				p.inFlightRequests.Wait()
+
+				if time.Since(p.lastRequestHandled) > maxDuration {
+					p.proxyLogger.Infof("<%s> Unloading model, TTL of %ds reached", p.ID, p.config.UnloadAfter)
+					p.Stop()
+					return
+				}
 			}
+		}()
+	}
 
-			if ttl < 0 {
-				return fmt.Errorf("failed to check health from: %s", healthURL)
-			}
-
-			continue
-		}
-
-		defer resp.Body.Close()
-		if resp.StatusCode == http.StatusOK {
-			return nil
-		}
-
-		if ttl < 0 {
-			return fmt.Errorf("failed to check health from: %s", healthURL)
-		}
-
-		time.Sleep(time.Second)
+	if curState, err := p.swapState(StateStarting, StateReady); err != nil {
+		return fmt.Errorf("failed to set Process state to ready: current state: %v, error: %v", curState, err)
+	} else {
+		return nil
 	}
 }
 
-func (p *Process) ProxyRequest(w http.ResponseWriter, r *http.Request) {
+// Stop will wait for inflight requests to complete before stopping the process.
+func (p *Process) Stop() {
+	if !isValidTransition(p.CurrentState(), StateStopping) {
+		return
+	}
 
-	p.inFlightRequests.Add(1)
-	defer p.inFlightRequests.Done()
+	// wait for any inflight requests before proceeding
+	p.proxyLogger.Debugf("<%s> Stop(): Waiting for inflight requests to complete", p.ID)
+	p.inFlightRequests.Wait()
+	p.StopImmediately()
+}
 
-	if p.CurrentState() != StateReady {
-		if err := p.start(); err != nil {
-			errstr := fmt.Sprintf("unable to start process: %s", err)
-			http.Error(w, errstr, http.StatusInternalServerError)
+// StopImmediately will transition the process to the stopping state and stop the process with a SIGTERM.
+// If the process does not stop within the specified timeout, it will be forcefully stopped with a SIGKILL.
+func (p *Process) StopImmediately() {
+	if !isValidTransition(p.CurrentState(), StateStopping) {
+		return
+	}
+
+	p.proxyLogger.Debugf("<%s> Stopping process", p.ID)
+
+	// calling Stop() when state is invalid is a no-op
+	if curState, err := p.swapState(StateReady, StateStopping); err != nil {
+		p.proxyLogger.Infof("<%s> Stop() Ready -> StateStopping err: %v, current state: %v", p.ID, err, curState)
+		return
+	}
+
+	// stop the process with a graceful exit timeout
+	p.stopCommand(p.gracefulStopTimeout)
+
+	if curState, err := p.swapState(StateStopping, StateStopped); err != nil {
+		p.proxyLogger.Infof("<%s> Stop() StateStopping -> StateStopped err: %v, current state: %v", p.ID, err, curState)
+	}
+}
+
+// Shutdown is called when llama-swap is shutting down. It will give a little bit
+// of time for any inflight requests to complete before shutting down. If the Process
+// is in the state of starting, it will cancel it and shut it down. Once a process is in
+// the StateShutdown state, it can not be started again.
+func (p *Process) Shutdown() {
+	p.shutdownCancel()
+	p.stopCommand(p.gracefulStopTimeout)
+	p.state = StateShutdown
+}
+
+// stopCommand will send a SIGTERM to the process and wait for it to exit.
+// If it does not exit within 5 seconds, it will send a SIGKILL.
+func (p *Process) stopCommand(sigtermTTL time.Duration) {
+	stopStartTime := time.Now()
+	defer func() {
+		p.proxyLogger.Debugf("<%s> stopCommand took %v", p.ID, time.Since(stopStartTime))
+	}()
+
+	sigtermTimeout, cancelTimeout := context.WithTimeout(context.Background(), sigtermTTL)
+	defer cancelTimeout()
+
+	if p.cmd == nil || p.cmd.Process == nil {
+		p.proxyLogger.Debugf("<%s> cmd or cmd.Process is nil (normal during config reload)", p.ID)
+		return
+	}
+
+	// if err := p.terminateProcess(); err != nil {
+	// 	p.proxyLogger.Debugf("<%s> Process already terminated: %v (normal during shutdown)", p.ID, err)
+	// }
+	// the default cmdStop to taskkill /f /t /pid ${PID}
+	if runtime.GOOS == "windows" && strings.TrimSpace(p.config.CmdStop) == "" {
+		p.config.CmdStop = "taskkill /f /t /pid ${PID}"
+	}
+
+	if p.config.CmdStop != "" {
+		// replace ${PID} with the pid of the process
+		stopArgs, err := SanitizeCommand(strings.ReplaceAll(p.config.CmdStop, "${PID}", fmt.Sprintf("%d", p.cmd.Process.Pid)))
+		if err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to sanitize stop command: %v", p.ID, err)
+			return
+		}
+
+		p.proxyLogger.Debugf("<%s> Executing stop command: %s", p.ID, strings.Join(stopArgs, " "))
+
+		stopCmd := exec.Command(stopArgs[0], stopArgs[1:]...)
+		stopCmd.Stdout = p.processLogger
+		stopCmd.Stderr = p.processLogger
+		stopCmd.Env = p.config.Env
+
+		if err := stopCmd.Run(); err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to exec stop command: %v", p.ID, err)
+			return
+		}
+	} else {
+		if err := p.cmd.Process.Signal(syscall.SIGTERM); err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to send SIGTERM to process: %v", p.ID, err)
 			return
 		}
 	}
 
-	p.lastRequestHandled = time.Now()
+	select {
+	case <-sigtermTimeout.Done():
+		p.proxyLogger.Debugf("<%s> Process timed out waiting to stop, sending KILL signal (normal during shutdown)", p.ID)
+		p.upstreamWasStoppedWithKill = true
+		if err := p.cmd.Process.Kill(); err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to kill process: %v", p.ID, err)
+		}
+	case err := <-p.cmdWaitChan:
+		// Note: in start(), p.cmdWaitChan also has a select { ... }. That should be OK
+		// because if we make it here then the cmd has been successfully running and made it
+		// through the health check. There is a possibility that the cmd crashed after the health check
+		// succeeded but that's not a case llama-swap is handling for now.
+		if err != nil {
+			if errno, ok := err.(syscall.Errno); ok {
+				p.proxyLogger.Errorf("<%s> errno >> %v", p.ID, errno)
+			} else if exitError, ok := err.(*exec.ExitError); ok {
+				if strings.Contains(exitError.String(), "signal: terminated") {
+					p.proxyLogger.Debugf("<%s> Process stopped OK", p.ID)
+				} else if strings.Contains(exitError.String(), "signal: interrupt") {
+					p.proxyLogger.Debugf("<%s> Process interrupted OK", p.ID)
+				} else {
+					p.proxyLogger.Warnf("<%s> ExitError >> %v, exit code: %d", p.ID, exitError, exitError.ExitCode())
+				}
+			} else {
+				p.proxyLogger.Errorf("<%s> Process exited >> %v", p.ID, err)
+			}
+		}
+	}
+}
+
+func (p *Process) checkHealthEndpoint(healthURL string) error {
+	client := &http.Client{
+		Timeout: 500 * time.Millisecond,
+	}
+
+	req, err := http.NewRequest("GET", healthURL, nil)
+	if err != nil {
+		return err
+	}
+
+	resp, err := client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+
+	// got a response but it was not an OK
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("status code: %d", resp.StatusCode)
+	}
+
+	return nil
+}
+
+func (p *Process) ProxyRequest(w http.ResponseWriter, r *http.Request) {
+	requestBeginTime := time.Now()
+	var startDuration time.Duration
+
+	// prevent new requests from being made while stopping or irrecoverable
+	currentState := p.CurrentState()
+	if currentState == StateFailed || currentState == StateShutdown || currentState == StateStopping {
+		http.Error(w, fmt.Sprintf("Process can not ProxyRequest, state is %s", currentState), http.StatusServiceUnavailable)
+		return
+	}
+
+	select {
+	case p.concurrencyLimitSemaphore <- struct{}{}:
+		defer func() { <-p.concurrencyLimitSemaphore }()
+	default:
+		http.Error(w, "Too many requests", http.StatusTooManyRequests)
+		return
+	}
+
+	p.inFlightRequests.Add(1)
+	defer func() {
+		p.lastRequestHandled = time.Now()
+		p.inFlightRequests.Done()
+	}()
+
+	// start the process on demand
+	if p.CurrentState() != StateReady {
+		beginStartTime := time.Now()
+		if err := p.start(); err != nil {
+			errstr := fmt.Sprintf("unable to start process: %s", err)
+			http.Error(w, errstr, http.StatusBadGateway)
+			return
+		}
+		startDuration = time.Since(beginStartTime)
+	}
 
 	proxyTo := p.config.Proxy
 	client := &http.Client{}
-	req, err := http.NewRequest(r.Method, proxyTo+r.URL.String(), r.Body)
+	req, err := http.NewRequestWithContext(r.Context(), r.Method, proxyTo+r.URL.String(), r.Body)
 	if err != nil {
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	req.Header = r.Header.Clone()
+
+	contentLength, err := strconv.ParseInt(req.Header.Get("content-length"), 10, 64)
+	if err == nil {
+		req.ContentLength = contentLength
+	}
+
 	resp, err := client.Do(req)
 	if err != nil {
 		http.Error(w, err.Error(), http.StatusBadGateway)
@@ -328,4 +572,8 @@ func (p *Process) ProxyRequest(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 	}
+
+	totalTime := time.Since(requestBeginTime)
+	p.proxyLogger.Debugf("<%s> request %s - start: %v, total: %v",
+		p.ID, r.RequestURI, startDuration, totalTime)
 }

proxy/process_test.go

@@ -2,10 +2,10 @@ package proxy
 
 import (
 	"fmt"
-	"io"
 	"net/http"
 	"net/http/httptest"
 	"os"
+	"runtime"
 	"sync"
 	"testing"
 	"time"
@@ -13,13 +13,26 @@ import (
 	"github.com/stretchr/testify/assert"
 )
 
+var (
+	debugLogger = NewLogMonitorWriter(os.Stdout)
+)
+
+func init() {
+	// flip to help with debugging tests
+	if false {
+		debugLogger.SetLogLevel(LevelDebug)
+	} else {
+		debugLogger.SetLogLevel(LevelError)
+	}
+}
+
 func TestProcess_AutomaticallyStartsUpstream(t *testing.T) {
-	logMonitor := NewLogMonitorWriter(io.Discard)
+
 	expectedMessage := "testing91931"
 	config := getTestSimpleResponderConfig(expectedMessage)
 
 	// Create a process
-	process := NewProcess("test-process", 5, config, logMonitor)
+	process := NewProcess("test-process", 5, config, debugLogger, debugLogger)
 	defer process.Stop()
 
 	req := httptest.NewRequest("GET", "/test", nil)
@@ -48,6 +61,32 @@ func TestProcess_AutomaticallyStartsUpstream(t *testing.T) {
 	}
 }
 
+// TestProcess_WaitOnMultipleStarts tests that multiple concurrent requests
+// are all handled successfully, even though they all may ask for the process to .start()
+func TestProcess_WaitOnMultipleStarts(t *testing.T) {
+
+	expectedMessage := "testing91931"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	process := NewProcess("test-process", 5, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	var wg sync.WaitGroup
+	for i := 0; i < 5; i++ {
+		wg.Add(1)
+		go func(reqID int) {
+			defer wg.Done()
+			req := httptest.NewRequest("GET", "/test", nil)
+			w := httptest.NewRecorder()
+			process.ProxyRequest(w, req)
+			assert.Equal(t, http.StatusOK, w.Code, "Worker %d got wrong HTTP code", reqID)
+			assert.Contains(t, w.Body.String(), expectedMessage, "Worker %d got wrong message", reqID)
+		}(i)
+	}
+	wg.Wait()
+	assert.Equal(t, StateReady, process.CurrentState())
+}
+
 // test that the automatic start returns the expected error type
 func TestProcess_BrokenModelConfig(t *testing.T) {
 	// Create a process configuration
@@ -57,17 +96,20 @@ func TestProcess_BrokenModelConfig(t *testing.T) {
 		CheckEndpoint: "/health",
 	}
 
-	process := NewProcess("broken", 1, config, NewLogMonitor())
-	defer process.Stop()
+	process := NewProcess("broken", 1, config, debugLogger, debugLogger)
 
 	req := httptest.NewRequest("GET", "/", nil)
 	w := httptest.NewRecorder()
 	process.ProxyRequest(w, req)
-	assert.Equal(t, http.StatusInternalServerError, w.Code)
+	assert.Equal(t, http.StatusBadGateway, w.Code)
 	assert.Contains(t, w.Body.String(), "unable to start process")
+
+	w = httptest.NewRecorder()
+	process.ProxyRequest(w, req)
+	assert.Equal(t, http.StatusServiceUnavailable, w.Code)
+	assert.Contains(t, w.Body.String(), "Process can not ProxyRequest, state is failed")
 }
 
-// test that the process unloads after the TTL
 func TestProcess_UnloadAfterTTL(t *testing.T) {
 	if testing.Short() {
 		t.Skip("skipping long auto unload TTL test")
@@ -79,34 +121,76 @@ func TestProcess_UnloadAfterTTL(t *testing.T) {
 	config.UnloadAfter = 3 // seconds
 	assert.Equal(t, 3, config.UnloadAfter)
 
-	process := NewProcess("ttl", 2, config, NewLogMonitorWriter(io.Discard))
+	process := NewProcess("ttl_test", 2, config, debugLogger, debugLogger)
 	defer process.Stop()
 
-	req := httptest.NewRequest("GET", "/test", nil)
+	// this should take 4 seconds
+	req1 := httptest.NewRequest("GET", "/slow-respond?echo=1234&delay=1000ms", nil)
+	req2 := httptest.NewRequest("GET", "/test", nil)
+
 	w := httptest.NewRecorder()
 
-	// Proxy the request (auto start)
-	process.ProxyRequest(w, req)
+	// Proxy the request (auto start) with a slow response that takes longer than config.UnloadAfter
+	process.ProxyRequest(w, req1)
 
+	t.Log("sending slow first request (4 seconds)")
+	assert.Equal(t, http.StatusOK, w.Code, "Expected status code %d, got %d", http.StatusOK, w.Code)
+	assert.Contains(t, w.Body.String(), "1234")
+	assert.Equal(t, StateReady, process.CurrentState())
+
+	// ensure the TTL timeout does not race slow requests (see issue #25)
+	t.Log("sending second request (1 second)")
+	time.Sleep(time.Second)
+	w = httptest.NewRecorder()
+	process.ProxyRequest(w, req2)
 	assert.Equal(t, http.StatusOK, w.Code, "Expected status code %d, got %d", http.StatusOK, w.Code)
 	assert.Contains(t, w.Body.String(), expectedMessage)
-
 	assert.Equal(t, StateReady, process.CurrentState())
 
 	// wait 5 seconds
+	t.Log("sleep 5 seconds and check if unloaded")
 	time.Sleep(5 * time.Second)
 	assert.Equal(t, StateStopped, process.CurrentState())
 }
 
+func TestProcess_LowTTLValue(t *testing.T) {
+	if true { // change this code to run this ...
+		t.Skip("skipping test, edit process_test.go to run it ")
+	}
+
+	config := getTestSimpleResponderConfig("fast_ttl")
+	assert.Equal(t, 0, config.UnloadAfter)
+	config.UnloadAfter = 1 // second
+	assert.Equal(t, 1, config.UnloadAfter)
+
+	process := NewProcess("ttl", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	for i := 0; i < 100; i++ {
+		t.Logf("Waiting before sending request %d", i)
+		time.Sleep(1500 * time.Millisecond)
+
+		expected := fmt.Sprintf("echo=test_%d", i)
+		req := httptest.NewRequest("GET", fmt.Sprintf("/slow-respond?echo=%s&delay=50ms", expected), nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+		assert.Equal(t, http.StatusOK, w.Code)
+		assert.Contains(t, w.Body.String(), expected)
+	}
+
+}
+
 // issue #19
+// This test makes sure using Process.Stop() does not affect pending HTTP
+// requests. All HTTP requests in this test should complete successfully.
 func TestProcess_HTTPRequestsHaveTimeToFinish(t *testing.T) {
 	if testing.Short() {
-		t.Skip("skipping long test")
+		t.Skip("skipping slow test")
 	}
 
 	expectedMessage := "12345"
 	config := getTestSimpleResponderConfig(expectedMessage)
-	process := NewProcess("t", 10, config, NewLogMonitorWriter(os.Stdout))
+	process := NewProcess("t", 10, config, debugLogger, debugLogger)
 	defer process.Stop()
 
 	results := map[string]string{
@@ -122,8 +206,9 @@ func TestProcess_HTTPRequestsHaveTimeToFinish(t *testing.T) {
 		wg.Add(1)
 		go func(key string) {
 			defer wg.Done()
-			// send a request that should take 5 * 200ms (1 second) to complete
-			req := httptest.NewRequest("GET", fmt.Sprintf("/slow-respond?echo=%s&delay=200ms", key), nil)
+			// send a request where simple-responder is will wait 300ms before responding
+			// this will simulate an in-progress request.
+			req := httptest.NewRequest("GET", fmt.Sprintf("/slow-respond?echo=%s&delay=300ms", key), nil)
 			w := httptest.NewRecorder()
 
 			process.ProxyRequest(w, req)
@@ -139,9 +224,9 @@ func TestProcess_HTTPRequestsHaveTimeToFinish(t *testing.T) {
 		}(key)
 	}
 
-	// stop the requests in the middle
+	// Stop the process while requests are still being processed
 	go func() {
-		<-time.After(500 * time.Millisecond)
+		<-time.After(150 * time.Millisecond)
 		process.Stop()
 	}()
 
@@ -151,3 +236,235 @@ func TestProcess_HTTPRequestsHaveTimeToFinish(t *testing.T) {
 		assert.Equal(t, key, result)
 	}
 }
+
+func TestProcess_SwapState(t *testing.T) {
+	tests := []struct {
+		name           string
+		currentState   ProcessState
+		expectedState  ProcessState
+		newState       ProcessState
+		expectedError  error
+		expectedResult ProcessState
+	}{
+		{"Stopped to Starting", StateStopped, StateStopped, StateStarting, nil, StateStarting},
+		{"Starting to Ready", StateStarting, StateStarting, StateReady, nil, StateReady},
+		{"Starting to Failed", StateStarting, StateStarting, StateFailed, nil, StateFailed},
+		{"Starting to Stopping", StateStarting, StateStarting, StateStopping, nil, StateStopping},
+		{"Ready to Stopping", StateReady, StateReady, StateStopping, nil, StateStopping},
+		{"Stopping to Stopped", StateStopping, StateStopping, StateStopped, nil, StateStopped},
+		{"Stopping to Shutdown", StateStopping, StateStopping, StateShutdown, nil, StateShutdown},
+		{"Stopped to Ready", StateStopped, StateStopped, StateReady, ErrInvalidStateTransition, StateStopped},
+		{"Starting to Stopped", StateStarting, StateStarting, StateStopped, ErrInvalidStateTransition, StateStarting},
+		{"Ready to Starting", StateReady, StateReady, StateStarting, ErrInvalidStateTransition, StateReady},
+		{"Ready to Failed", StateReady, StateReady, StateFailed, ErrInvalidStateTransition, StateReady},
+		{"Stopping to Ready", StateStopping, StateStopping, StateReady, ErrInvalidStateTransition, StateStopping},
+		{"Failed to Stopped", StateFailed, StateFailed, StateStopped, ErrInvalidStateTransition, StateFailed},
+		{"Failed to Starting", StateFailed, StateFailed, StateStarting, ErrInvalidStateTransition, StateFailed},
+		{"Shutdown to Stopped", StateShutdown, StateShutdown, StateStopped, ErrInvalidStateTransition, StateShutdown},
+		{"Shutdown to Starting", StateShutdown, StateShutdown, StateStarting, ErrInvalidStateTransition, StateShutdown},
+		{"Expected state mismatch", StateStopped, StateStarting, StateStarting, ErrExpectedStateMismatch, StateStopped},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			p := NewProcess("test", 10, getTestSimpleResponderConfig("test"), debugLogger, debugLogger)
+			p.state = test.currentState
+
+			resultState, err := p.swapState(test.expectedState, test.newState)
+			if err != nil && test.expectedError == nil {
+				t.Errorf("Unexpected error: %v", err)
+			} else if err == nil && test.expectedError != nil {
+				t.Errorf("Expected error: %v, but got none", test.expectedError)
+			} else if err != nil && test.expectedError != nil {
+				if err.Error() != test.expectedError.Error() {
+					t.Errorf("Expected error: %v, got: %v", test.expectedError, err)
+				}
+			}
+
+			if resultState != test.expectedResult {
+				t.Errorf("Expected state: %v, got: %v", test.expectedResult, resultState)
+			}
+		})
+	}
+}
+
+func TestProcess_ShutdownInterruptsHealthCheck(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping long shutdown test")
+	}
+
+	expectedMessage := "testing91931"
+
+	// make a config where the healthcheck will always fail because port is wrong
+	config := getTestSimpleResponderConfigPort(expectedMessage, 9999)
+	config.Proxy = "http://localhost:9998/test"
+
+	healthCheckTTLSeconds := 30
+	process := NewProcess("test-process", healthCheckTTLSeconds, config, debugLogger, debugLogger)
+
+	// make it a lot faster
+	process.healthCheckLoopInterval = time.Second
+
+	// start a goroutine to simulate a shutdown
+	var wg sync.WaitGroup
+	go func() {
+		defer wg.Done()
+		<-time.After(time.Millisecond * 500)
+		process.Shutdown()
+	}()
+	wg.Add(1)
+
+	// start the process, this is a blocking call
+	err := process.start()
+
+	wg.Wait()
+	assert.ErrorContains(t, err, "health check interrupted due to shutdown")
+	assert.Equal(t, StateShutdown, process.CurrentState())
+}
+
+func TestProcess_ExitInterruptsHealthCheck(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping Exit Interrupts Health Check test")
+	}
+
+	// should run and exit but interrupt the long checkHealthTimeout
+	checkHealthTimeout := 5
+	config := ModelConfig{
+		Cmd:           "sleep 1",
+		Proxy:         "http://127.0.0.1:9913",
+		CheckEndpoint: "/health",
+	}
+
+	process := NewProcess("sleepy", checkHealthTimeout, config, debugLogger, debugLogger)
+	process.healthCheckLoopInterval = time.Second // make it faster
+	err := process.start()
+	assert.Equal(t, "upstream command exited prematurely but successfully", err.Error())
+	assert.Equal(t, process.CurrentState(), StateFailed)
+}
+
+func TestProcess_ConcurrencyLimit(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping long concurrency limit test")
+	}
+
+	expectedMessage := "concurrency_limit_test"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	// only allow 1 concurrent request at a time
+	config.ConcurrencyLimit = 1
+
+	process := NewProcess("ttl_test", 2, config, debugLogger, debugLogger)
+	assert.Equal(t, 1, cap(process.concurrencyLimitSemaphore))
+	defer process.Stop()
+
+	// launch a goroutine first to take up the semaphore
+	go func() {
+		req1 := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=75ms", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req1)
+		assert.Equal(t, http.StatusOK, w.Code)
+	}()
+
+	// let the goroutine start
+	<-time.After(time.Millisecond * 25)
+
+	denied := httptest.NewRequest("GET", "/test", nil)
+	w := httptest.NewRecorder()
+	process.ProxyRequest(w, denied)
+	assert.Equal(t, http.StatusTooManyRequests, w.Code)
+}
+
+func TestProcess_StopImmediately(t *testing.T) {
+	expectedMessage := "test_stop_immediate"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	process := NewProcess("stop_immediate", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+	go func() {
+		// slow, but will get killed by StopImmediate
+		req := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=1s", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+	}()
+	<-time.After(time.Millisecond)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+}
+
+// Test that SIGKILL is sent when gracefulStopTimeout is reached and properly terminates
+// the upstream command
+func TestProcess_ForceStopWithKill(t *testing.T) {
+
+	expectedMessage := "test_sigkill"
+	binaryPath := getSimpleResponderPath()
+	port := getTestPort()
+
+	config := ModelConfig{
+		// note --ignore-sig-term which ignores the SIGTERM signal so a SIGKILL must be sent
+		// to force the process to exit
+		Cmd:           fmt.Sprintf("%s --port %d --respond %s --silent --ignore-sig-term", binaryPath, port, expectedMessage),
+		Proxy:         fmt.Sprintf("http://127.0.0.1:%d", port),
+		CheckEndpoint: "/health",
+	}
+
+	process := NewProcess("stop_immediate", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	// reduce to make testing go faster
+	process.gracefulStopTimeout = time.Second
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+
+	waitChan := make(chan struct{})
+	go func() {
+		// slow, but will get killed by StopImmediate
+		req := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=2s", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+
+		// StatusOK because that was already sent before the kill
+		assert.Equal(t, http.StatusOK, w.Code)
+
+		// unexpected EOF because the kill happened, the "1" is sent before the kill
+		// then the unexpected EOF is sent after the kill
+		if runtime.GOOS == "windows" {
+			assert.Contains(t, w.Body.String(), "wsarecv: An existing connection was forcibly closed by the remote host")
+		} else {
+			assert.Contains(t, w.Body.String(), "unexpected EOF")
+		}
+
+		close(waitChan)
+	}()
+
+	<-time.After(time.Millisecond)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+
+	// the request should have been interrupted by SIGKILL
+	<-waitChan
+}
+
+func TestProcess_StopCmd(t *testing.T) {
+	config := getTestSimpleResponderConfig("test_stop_cmd")
+
+	if runtime.GOOS == "windows" {
+		config.CmdStop = "taskkill /f /t /pid ${PID}"
+	} else {
+		config.CmdStop = "kill -TERM ${PID}"
+	}
+
+	process := NewProcess("testStopCmd", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+}

proxy/processgroup.go

@@ -0,0 +1,114 @@
+package proxy
+
+import (
+	"fmt"
+	"net/http"
+	"slices"
+	"sync"
+)
+
+type ProcessGroup struct {
+	sync.Mutex
+
+	config     Config
+	id         string
+	swap       bool
+	exclusive  bool
+	persistent bool
+
+	proxyLogger    *LogMonitor
+	upstreamLogger *LogMonitor
+
+	// map of current processes
+	processes       map[string]*Process
+	lastUsedProcess string
+}
+
+func NewProcessGroup(id string, config Config, proxyLogger *LogMonitor, upstreamLogger *LogMonitor) *ProcessGroup {
+	groupConfig, ok := config.Groups[id]
+	if !ok {
+		panic("Unable to find configuration for group id: " + id)
+	}
+
+	pg := &ProcessGroup{
+		id:             id,
+		config:         config,
+		swap:           groupConfig.Swap,
+		exclusive:      groupConfig.Exclusive,
+		persistent:     groupConfig.Persistent,
+		proxyLogger:    proxyLogger,
+		upstreamLogger: upstreamLogger,
+		processes:      make(map[string]*Process),
+	}
+
+	// Create a Process for each member in the group
+	for _, modelID := range groupConfig.Members {
+		modelConfig, modelID, _ := pg.config.FindConfig(modelID)
+		process := NewProcess(modelID, pg.config.HealthCheckTimeout, modelConfig, pg.upstreamLogger, pg.proxyLogger)
+		pg.processes[modelID] = process
+	}
+
+	return pg
+}
+
+// ProxyRequest proxies a request to the specified model
+func (pg *ProcessGroup) ProxyRequest(modelID string, writer http.ResponseWriter, request *http.Request) error {
+	if !pg.HasMember(modelID) {
+		return fmt.Errorf("model %s not part of group %s", modelID, pg.id)
+	}
+
+	if pg.swap {
+		pg.Lock()
+		if pg.lastUsedProcess != modelID {
+			if pg.lastUsedProcess != "" {
+				pg.processes[pg.lastUsedProcess].Stop()
+			}
+			pg.lastUsedProcess = modelID
+		}
+		pg.Unlock()
+	}
+
+	pg.processes[modelID].ProxyRequest(writer, request)
+	return nil
+}
+
+func (pg *ProcessGroup) HasMember(modelName string) bool {
+	return slices.Contains(pg.config.Groups[pg.id].Members, modelName)
+}
+
+func (pg *ProcessGroup) StopProcesses(strategy StopStrategy) {
+	pg.Lock()
+	defer pg.Unlock()
+
+	if len(pg.processes) == 0 {
+		return
+	}
+
+	// stop Processes in parallel
+	var wg sync.WaitGroup
+	for _, process := range pg.processes {
+		wg.Add(1)
+		go func(process *Process) {
+			defer wg.Done()
+			switch strategy {
+			case StopImmediately:
+				process.StopImmediately()
+			default:
+				process.Stop()
+			}
+		}(process)
+	}
+	wg.Wait()
+}
+
+func (pg *ProcessGroup) Shutdown() {
+	var wg sync.WaitGroup
+	for _, process := range pg.processes {
+		wg.Add(1)
+		go func(process *Process) {
+			defer wg.Done()
+			process.Shutdown()
+		}(process)
+	}
+	wg.Wait()
+}

proxy/processgroup_test.go

@@ -0,0 +1,96 @@
+package proxy
+
+import (
+	"bytes"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+var processGroupTestConfig = AddDefaultGroupToConfig(Config{
+	HealthCheckTimeout: 15,
+	Models: map[string]ModelConfig{
+		"model1": getTestSimpleResponderConfig("model1"),
+		"model2": getTestSimpleResponderConfig("model2"),
+		"model3": getTestSimpleResponderConfig("model3"),
+		"model4": getTestSimpleResponderConfig("model4"),
+		"model5": getTestSimpleResponderConfig("model5"),
+	},
+	Groups: map[string]GroupConfig{
+		"G1": {
+			Swap:      true,
+			Exclusive: true,
+			Members:   []string{"model1", "model2"},
+		},
+		"G2": {
+			Swap:      false,
+			Exclusive: true,
+			Members:   []string{"model3", "model4"},
+		},
+	},
+})
+
+func TestProcessGroup_DefaultHasCorrectModel(t *testing.T) {
+	pg := NewProcessGroup(DEFAULT_GROUP_ID, processGroupTestConfig, testLogger, testLogger)
+	assert.True(t, pg.HasMember("model5"))
+}
+
+func TestProcessGroup_HasMember(t *testing.T) {
+	pg := NewProcessGroup("G1", processGroupTestConfig, testLogger, testLogger)
+	assert.True(t, pg.HasMember("model1"))
+	assert.True(t, pg.HasMember("model2"))
+	assert.False(t, pg.HasMember("model3"))
+}
+
+func TestProcessGroup_ProxyRequestSwapIsTrue(t *testing.T) {
+	pg := NewProcessGroup("G1", processGroupTestConfig, testLogger, testLogger)
+	defer pg.StopProcesses(StopWaitForInflightRequest)
+
+	tests := []string{"model1", "model2"}
+
+	for _, modelName := range tests {
+		t.Run(modelName, func(t *testing.T) {
+			reqBody := `{"x", "y"}`
+			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+			w := httptest.NewRecorder()
+
+			assert.NoError(t, pg.ProxyRequest(modelName, w, req))
+			assert.Equal(t, http.StatusOK, w.Code)
+			assert.Contains(t, w.Body.String(), modelName)
+
+			// make sure only one process is in the running state
+			count := 0
+			for _, process := range pg.processes {
+				if process.CurrentState() == StateReady {
+					count++
+				}
+			}
+			assert.Equal(t, 1, count)
+		})
+	}
+}
+
+func TestProcessGroup_ProxyRequestSwapIsFalse(t *testing.T) {
+	pg := NewProcessGroup("G2", processGroupTestConfig, testLogger, testLogger)
+	defer pg.StopProcesses(StopWaitForInflightRequest)
+
+	tests := []string{"model3", "model4"}
+
+	for _, modelName := range tests {
+		t.Run(modelName, func(t *testing.T) {
+			reqBody := `{"x", "y"}`
+			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+			w := httptest.NewRecorder()
+			assert.NoError(t, pg.ProxyRequest(modelName, w, req))
+			assert.Equal(t, http.StatusOK, w.Code)
+			assert.Contains(t, w.Body.String(), modelName)
+		})
+	}
+
+	// make sure all the processes are running
+	for _, process := range pg.processes {
+		assert.Equal(t, StateReady, process.CurrentState())
+	}
+}

proxy/proxymanager.go

@@ -5,37 +5,154 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"mime/multipart"
 	"net/http"
+	"os"
+	"sort"
 	"strconv"
 	"strings"
 	"sync"
 	"time"
 
 	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const (
+	PROFILE_SPLIT_CHAR = ":"
 )
 
 type ProxyManager struct {
 	sync.Mutex
 
-	config           *Config
-	currentProcesses map[string]*Process
-	logMonitor       *LogMonitor
-	ginEngine        *gin.Engine
+	config    Config
+	ginEngine *gin.Engine
+
+	// logging
+	proxyLogger    *LogMonitor
+	upstreamLogger *LogMonitor
+	muxLogger      *LogMonitor
+
+	processGroups map[string]*ProcessGroup
 }
 
-func New(config *Config) *ProxyManager {
-	pm := &ProxyManager{
-		config:           config,
-		currentProcesses: make(map[string]*Process),
-		logMonitor:       NewLogMonitor(),
-		ginEngine:        gin.New(),
+func New(config Config) *ProxyManager {
+	// set up loggers
+	stdoutLogger := NewLogMonitorWriter(os.Stdout)
+	upstreamLogger := NewLogMonitorWriter(stdoutLogger)
+	proxyLogger := NewLogMonitorWriter(stdoutLogger)
+
+	if config.LogRequests {
+		proxyLogger.Warn("LogRequests configuration is deprecated. Use logLevel instead.")
 	}
 
-	// Set up routes using the Gin engine
-	pm.ginEngine.POST("/v1/chat/completions", pm.proxyChatRequestHandler)
+	switch strings.ToLower(strings.TrimSpace(config.LogLevel)) {
+	case "debug":
+		proxyLogger.SetLogLevel(LevelDebug)
+		upstreamLogger.SetLogLevel(LevelDebug)
+	case "info":
+		proxyLogger.SetLogLevel(LevelInfo)
+		upstreamLogger.SetLogLevel(LevelInfo)
+	case "warn":
+		proxyLogger.SetLogLevel(LevelWarn)
+		upstreamLogger.SetLogLevel(LevelWarn)
+	case "error":
+		proxyLogger.SetLogLevel(LevelError)
+		upstreamLogger.SetLogLevel(LevelError)
+	default:
+		proxyLogger.SetLogLevel(LevelInfo)
+		upstreamLogger.SetLogLevel(LevelInfo)
+	}
 
+	pm := &ProxyManager{
+		config:    config,
+		ginEngine: gin.New(),
+
+		proxyLogger:    proxyLogger,
+		muxLogger:      stdoutLogger,
+		upstreamLogger: upstreamLogger,
+
+		processGroups: make(map[string]*ProcessGroup),
+	}
+
+	// create the process groups
+	for groupID := range config.Groups {
+		processGroup := NewProcessGroup(groupID, config, proxyLogger, upstreamLogger)
+		pm.processGroups[groupID] = processGroup
+	}
+
+	pm.setupGinEngine()
+	return pm
+}
+
+func (pm *ProxyManager) setupGinEngine() {
+	pm.ginEngine.Use(func(c *gin.Context) {
+		// Start timer
+		start := time.Now()
+
+		// capture these because /upstream/:model rewrites them in c.Next()
+		clientIP := c.ClientIP()
+		method := c.Request.Method
+		path := c.Request.URL.Path
+
+		// Process request
+		c.Next()
+
+		// Stop timer
+		duration := time.Since(start)
+
+		statusCode := c.Writer.Status()
+		bodySize := c.Writer.Size()
+
+		pm.proxyLogger.Infof("Request %s \"%s %s %s\" %d %d \"%s\" %v",
+			clientIP,
+			method,
+			path,
+			c.Request.Proto,
+			statusCode,
+			bodySize,
+			c.Request.UserAgent(),
+			duration,
+		)
+	})
+
+	// see: issue: #81, #77 and #42 for CORS issues
+	// respond with permissive OPTIONS for any endpoint
+	pm.ginEngine.Use(func(c *gin.Context) {
+		if c.Request.Method == "OPTIONS" {
+			c.Header("Access-Control-Allow-Origin", "*")
+			c.Header("Access-Control-Allow-Methods", "GET, POST, PUT, PATCH, DELETE, OPTIONS")
+
+			// allow whatever the client requested by default
+			if headers := c.Request.Header.Get("Access-Control-Request-Headers"); headers != "" {
+				sanitized := SanitizeAccessControlRequestHeaderValues(headers)
+				c.Header("Access-Control-Allow-Headers", sanitized)
+			} else {
+				c.Header(
+					"Access-Control-Allow-Headers",
+					"Content-Type, Authorization, Accept, X-Requested-With",
+				)
+			}
+			c.Header("Access-Control-Max-Age", "86400")
+			c.AbortWithStatus(http.StatusNoContent)
+			return
+		}
+		c.Next()
+	})
+
+	// Set up routes using the Gin engine
+	pm.ginEngine.POST("/v1/chat/completions", pm.proxyOAIHandler)
 	// Support legacy /v1/completions api, see issue #12
-	pm.ginEngine.POST("/v1/completions", pm.proxyChatRequestHandler)
+	pm.ginEngine.POST("/v1/completions", pm.proxyOAIHandler)
+
+	// Support embeddings
+	pm.ginEngine.POST("/v1/embeddings", pm.proxyOAIHandler)
+	pm.ginEngine.POST("/v1/rerank", pm.proxyOAIHandler)
+
+	// Support audio/speech endpoint
+	pm.ginEngine.POST("/v1/audio/speech", pm.proxyOAIHandler)
+	pm.ginEngine.POST("/v1/audio/transcriptions", pm.proxyOAIPostFormHandler)
 
 	pm.ginEngine.GET("/v1/models", pm.listModelsHandler)
 
@@ -43,46 +160,121 @@ func New(config *Config) *ProxyManager {
 	pm.ginEngine.GET("/logs", pm.sendLogsHandlers)
 	pm.ginEngine.GET("/logs/stream", pm.streamLogsHandler)
 	pm.ginEngine.GET("/logs/streamSSE", pm.streamLogsHandlerSSE)
+	pm.ginEngine.GET("/logs/stream/:logMonitorID", pm.streamLogsHandler)
+	pm.ginEngine.GET("/logs/streamSSE/:logMonitorID", pm.streamLogsHandlerSSE)
 
-	pm.ginEngine.NoRoute(pm.proxyNoRouteHandler)
+	pm.ginEngine.GET("/upstream", pm.upstreamIndex)
+	pm.ginEngine.Any("/upstream/:model_id/*upstreamPath", pm.proxyToUpstream)
+
+	pm.ginEngine.GET("/unload", pm.unloadAllModelsHandler)
+
+	pm.ginEngine.GET("/running", pm.listRunningProcessesHandler)
+
+	pm.ginEngine.GET("/", func(c *gin.Context) {
+		// Set the Content-Type header to text/html
+		c.Header("Content-Type", "text/html")
+
+		// Write the embedded HTML content to the response
+		htmlData, err := getHTMLFile("index.html")
+		if err != nil {
+			c.String(http.StatusInternalServerError, err.Error())
+			return
+		}
+		_, err = c.Writer.Write(htmlData)
+		if err != nil {
+			c.String(http.StatusInternalServerError, fmt.Sprintf("failed to write response: %v", err))
+			return
+		}
+	})
+
+	pm.ginEngine.GET("/favicon.ico", func(c *gin.Context) {
+		if data, err := getHTMLFile("favicon.ico"); err == nil {
+			c.Data(http.StatusOK, "image/x-icon", data)
+		} else {
+			c.String(http.StatusInternalServerError, err.Error())
+		}
+	})
 
 	// Disable console color for testing
 	gin.DisableConsoleColor()
-
-	return pm
 }
 
-func (pm *ProxyManager) Run(addr ...string) error {
-	return pm.ginEngine.Run(addr...)
-}
-
-func (pm *ProxyManager) HandlerFunc(w http.ResponseWriter, r *http.Request) {
+// ServeHTTP implements http.Handler interface
+func (pm *ProxyManager) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	pm.ginEngine.ServeHTTP(w, r)
 }
 
-func (pm *ProxyManager) StopProcesses() {
+// StopProcesses acquires a lock and stops all running upstream processes.
+// This is the public method safe for concurrent calls.
+// Unlike Shutdown, this method only stops the processes but doesn't perform
+// a complete shutdown, allowing for process replacement without full termination.
+func (pm *ProxyManager) StopProcesses(strategy StopStrategy) {
 	pm.Lock()
 	defer pm.Unlock()
 
-	pm.stopProcesses()
+	// stop Processes in parallel
+	var wg sync.WaitGroup
+	for _, processGroup := range pm.processGroups {
+		wg.Add(1)
+		go func(processGroup *ProcessGroup) {
+			defer wg.Done()
+			processGroup.StopProcesses(strategy)
+		}(processGroup)
+	}
+
+	wg.Wait()
 }
 
-// for internal usage
-func (pm *ProxyManager) stopProcesses() {
-	if len(pm.currentProcesses) == 0 {
-		return
+// Shutdown stops all processes managed by this ProxyManager
+func (pm *ProxyManager) Shutdown() {
+	pm.Lock()
+	defer pm.Unlock()
+
+	pm.proxyLogger.Debug("Shutdown() called in proxy manager")
+
+	var wg sync.WaitGroup
+	// Send shutdown signal to all process in groups
+	for _, processGroup := range pm.processGroups {
+		wg.Add(1)
+		go func(processGroup *ProcessGroup) {
+			defer wg.Done()
+			processGroup.Shutdown()
+		}(processGroup)
+	}
+	wg.Wait()
+}
+
+func (pm *ProxyManager) swapProcessGroup(requestedModel string) (*ProcessGroup, string, error) {
+	// de-alias the real model name and get a real one
+	realModelName, found := pm.config.RealModelName(requestedModel)
+	if !found {
+		return nil, realModelName, fmt.Errorf("could not find real modelID for %s", requestedModel)
 	}
 
-	for _, process := range pm.currentProcesses {
-		process.Stop()
+	processGroup := pm.findGroupByModelName(realModelName)
+	if processGroup == nil {
+		return nil, realModelName, fmt.Errorf("could not find process group for model %s", requestedModel)
 	}
 
-	pm.currentProcesses = make(map[string]*Process)
+	if processGroup.exclusive {
+		pm.proxyLogger.Debugf("Exclusive mode for group %s, stopping other process groups", processGroup.id)
+		for groupId, otherGroup := range pm.processGroups {
+			if groupId != processGroup.id && !otherGroup.persistent {
+				otherGroup.StopProcesses(StopWaitForInflightRequest)
+			}
+		}
+	}
+
+	return processGroup, realModelName, nil
 }
 
 func (pm *ProxyManager) listModelsHandler(c *gin.Context) {
 	data := []interface{}{}
-	for id := range pm.config.Models {
+	for id, modelConfig := range pm.config.Models {
+		if modelConfig.Unlisted {
+			continue
+		}
+
 		data = append(data, map[string]interface{}{
 			"id":       id,
 			"object":   "model",
@@ -94,110 +286,281 @@ func (pm *ProxyManager) listModelsHandler(c *gin.Context) {
 	// Set the Content-Type header to application/json
 	c.Header("Content-Type", "application/json")
 
+	if origin := c.Request.Header.Get("Origin"); origin != "" {
+		c.Header("Access-Control-Allow-Origin", origin)
+	}
+
 	// Encode the data as JSON and write it to the response writer
 	if err := json.NewEncoder(c.Writer).Encode(map[string]interface{}{"data": data}); err != nil {
-		c.AbortWithError(http.StatusInternalServerError, fmt.Errorf("error encoding JSON"))
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error encoding JSON %s", err.Error()))
 		return
 	}
 }
 
-func (pm *ProxyManager) swapModel(requestedModel string) (*Process, error) {
-	pm.Lock()
-	defer pm.Unlock()
+func (pm *ProxyManager) proxyToUpstream(c *gin.Context) {
+	requestedModel := c.Param("model_id")
 
-	// Check if requestedModel contains a /
-	groupName, modelName := "", requestedModel
-	if idx := strings.Index(requestedModel, "/"); idx != -1 {
-		groupName = requestedModel[:idx]
-		modelName = requestedModel[idx+1:]
+	if requestedModel == "" {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "model id required in path")
+		return
 	}
 
-	if groupName != "" {
-		if _, found := pm.config.Profiles[groupName]; !found {
-			return nil, fmt.Errorf("model group not found %s", groupName)
-		}
+	processGroup, _, err := pm.swapProcessGroup(requestedModel)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
 	}
 
-	// de-alias the real model name and get a real one
-	realModelName, found := pm.config.RealModelName(modelName)
-	if !found {
-		return nil, fmt.Errorf("could not find modelID for %s", requestedModel)
-	}
+	// rewrite the path
+	c.Request.URL.Path = c.Param("upstreamPath")
+	processGroup.ProxyRequest(requestedModel, c.Writer, c.Request)
+}
 
-	// exit early when already running, otherwise stop everything and swap
-	requestedProcessKey := groupName + "/" + realModelName
-	if process, found := pm.currentProcesses[requestedProcessKey]; found {
-		return process, nil
-	}
+func (pm *ProxyManager) upstreamIndex(c *gin.Context) {
+	var html strings.Builder
 
-	// stop all running models
-	pm.stopProcesses()
+	html.WriteString("<!doctype HTML>\n<html><body><h1>Available Models</h1><ul>")
 
-	if groupName == "" {
-		modelConfig, modelID, found := pm.config.FindConfig(realModelName)
-		if !found {
-			return nil, fmt.Errorf("could not find configuration for %s", realModelName)
+	// Extract keys and sort them
+	var modelIDs []string
+	for modelID, modelConfig := range pm.config.Models {
+		if modelConfig.Unlisted {
+			continue
 		}
 
-		process := NewProcess(modelID, pm.config.HealthCheckTimeout, modelConfig, pm.logMonitor)
-		processKey := groupName + "/" + modelID
-		pm.currentProcesses[processKey] = process
-	} else {
-		for _, modelName := range pm.config.Profiles[groupName] {
-			if realModelName, found := pm.config.RealModelName(modelName); found {
-				modelConfig, modelID, found := pm.config.FindConfig(realModelName)
-				if !found {
-					return nil, fmt.Errorf("could not find configuration for %s in group %s", realModelName, groupName)
+		modelIDs = append(modelIDs, modelID)
+	}
+	sort.Strings(modelIDs)
+
+	// Iterate over sorted keys
+	for _, modelID := range modelIDs {
+		// Get process state
+		processGroup := pm.findGroupByModelName(modelID)
+		var state string
+		if processGroup != nil {
+			process := processGroup.processes[modelID]
+			if process != nil {
+				var stateStr string
+				switch process.CurrentState() {
+				case StateReady:
+					stateStr = "Ready"
+				case StateStarting:
+					stateStr = "Starting"
+				case StateStopping:
+					stateStr = "Stopping"
+				case StateFailed:
+					stateStr = "Failed"
+				case StateShutdown:
+					stateStr = "Shutdown"
+				default:
+					stateStr = "Unknown"
 				}
+				state = stateStr
+			}
+		}
+		html.WriteString(fmt.Sprintf("<li><a href=\"/upstream/%s\">%s</a> - %s</li>", modelID, modelID, state))
+	}
+	html.WriteString("</ul></body></html>")
+	c.Header("Content-Type", "text/html")
+	c.String(http.StatusOK, html.String())
+}
 
-				process := NewProcess(modelID, pm.config.HealthCheckTimeout, modelConfig, pm.logMonitor)
-				processKey := groupName + "/" + modelID
-				pm.currentProcesses[processKey] = process
+func (pm *ProxyManager) proxyOAIHandler(c *gin.Context) {
+	bodyBytes, err := io.ReadAll(c.Request.Body)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "could not ready request body")
+		return
+	}
+
+	requestedModel := gjson.GetBytes(bodyBytes, "model").String()
+	if requestedModel == "" {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "missing or invalid 'model' key")
+		return
+	}
+
+	processGroup, realModelName, err := pm.swapProcessGroup(requestedModel)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
+	}
+
+	// issue #69 allow custom model names to be sent to upstream
+	useModelName := pm.config.Models[realModelName].UseModelName
+	if useModelName != "" {
+		bodyBytes, err = sjson.SetBytes(bodyBytes, "model", useModelName)
+		if err != nil {
+			pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error rewriting model name in JSON: %s", err.Error()))
+			return
+		}
+	}
+
+	c.Request.Body = io.NopCloser(bytes.NewBuffer(bodyBytes))
+
+	// dechunk it as we already have all the body bytes see issue #11
+	c.Request.Header.Del("transfer-encoding")
+	c.Request.Header.Set("content-length", strconv.Itoa(len(bodyBytes)))
+	c.Request.ContentLength = int64(len(bodyBytes))
+
+	if err := processGroup.ProxyRequest(realModelName, c.Writer, c.Request); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error proxying request: %s", err.Error()))
+		pm.proxyLogger.Errorf("Error Proxying Request for processGroup %s and model %s", processGroup.id, realModelName)
+		return
+	}
+}
+
+func (pm *ProxyManager) proxyOAIPostFormHandler(c *gin.Context) {
+	// Parse multipart form
+	if err := c.Request.ParseMultipartForm(32 << 20); err != nil { // 32MB max memory, larger files go to tmp disk
+		pm.sendErrorResponse(c, http.StatusBadRequest, fmt.Sprintf("error parsing multipart form: %s", err.Error()))
+		return
+	}
+
+	// Get model parameter from the form
+	requestedModel := c.Request.FormValue("model")
+	if requestedModel == "" {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "missing or invalid 'model' parameter in form data")
+		return
+	}
+
+	processGroup, realModelName, err := pm.swapProcessGroup(requestedModel)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
+	}
+
+	// We need to reconstruct the multipart form in any case since the body is consumed
+	// Create a new buffer for the reconstructed request
+	var requestBuffer bytes.Buffer
+	multipartWriter := multipart.NewWriter(&requestBuffer)
+
+	// Copy all form values
+	for key, values := range c.Request.MultipartForm.Value {
+		for _, value := range values {
+			fieldValue := value
+			// If this is the model field and we have a profile, use just the model name
+			if key == "model" {
+				// # issue #69 allow custom model names to be sent to upstream
+				useModelName := pm.config.Models[realModelName].UseModelName
+
+				if useModelName != "" {
+					fieldValue = useModelName
+				} else {
+					fieldValue = requestedModel
+				}
+			}
+			field, err := multipartWriter.CreateFormField(key)
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error recreating form field")
+				return
+			}
+			if _, err = field.Write([]byte(fieldValue)); err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error writing form field")
+				return
 			}
 		}
 	}
 
-	// requestedProcessKey should exist due to swap
-	return pm.currentProcesses[requestedProcessKey], nil
-}
+	// Copy all files from the original request
+	for key, fileHeaders := range c.Request.MultipartForm.File {
+		for _, fileHeader := range fileHeaders {
+			formFile, err := multipartWriter.CreateFormFile(key, fileHeader.Filename)
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error recreating form file")
+				return
+			}
 
-func (pm *ProxyManager) proxyChatRequestHandler(c *gin.Context) {
-	bodyBytes, err := io.ReadAll(c.Request.Body)
+			file, err := fileHeader.Open()
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error opening uploaded file")
+				return
+			}
+
+			if _, err = io.Copy(formFile, file); err != nil {
+				file.Close()
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error copying file data")
+				return
+			}
+			file.Close()
+		}
+	}
+
+	// Close the multipart writer to finalize the form
+	if err := multipartWriter.Close(); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, "error finalizing multipart form")
+		return
+	}
+
+	// Create a new request with the reconstructed form data
+	modifiedReq, err := http.NewRequestWithContext(
+		c.Request.Context(),
+		c.Request.Method,
+		c.Request.URL.String(),
+		&requestBuffer,
+	)
 	if err != nil {
-		c.AbortWithError(http.StatusBadRequest, fmt.Errorf("invalid JSON"))
-		return
-	}
-	var requestBody map[string]interface{}
-	if err := json.Unmarshal(bodyBytes, &requestBody); err != nil {
-		c.AbortWithError(http.StatusBadRequest, fmt.Errorf("invalid JSON"))
-		return
-	}
-	model, ok := requestBody["model"].(string)
-	if !ok {
-		c.AbortWithError(http.StatusBadRequest, fmt.Errorf("missing or invalid 'model' key"))
+		pm.sendErrorResponse(c, http.StatusInternalServerError, "error creating modified request")
 		return
 	}
 
-	if process, err := pm.swapModel(model); err != nil {
-		c.AbortWithError(http.StatusNotFound, fmt.Errorf("unable to swap to model, %s", err.Error()))
+	// Copy the headers from the original request
+	modifiedReq.Header = c.Request.Header.Clone()
+	modifiedReq.Header.Set("Content-Type", multipartWriter.FormDataContentType())
+
+	// set the content length of the body
+	modifiedReq.Header.Set("Content-Length", strconv.Itoa(requestBuffer.Len()))
+	modifiedReq.ContentLength = int64(requestBuffer.Len())
+
+	// Use the modified request for proxying
+	if err := processGroup.ProxyRequest(realModelName, c.Writer, modifiedReq); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error proxying request: %s", err.Error()))
+		pm.proxyLogger.Errorf("Error Proxying Request for processGroup %s and model %s", processGroup.id, realModelName)
 		return
+	}
+}
+
+func (pm *ProxyManager) sendErrorResponse(c *gin.Context, statusCode int, message string) {
+	acceptHeader := c.GetHeader("Accept")
+
+	if strings.Contains(acceptHeader, "application/json") {
+		c.JSON(statusCode, gin.H{"error": message})
 	} else {
-		c.Request.Body = io.NopCloser(bytes.NewBuffer(bodyBytes))
-
-		// dechunk it as we already have all the body bytes see issue #11
-		c.Request.Header.Del("transfer-encoding")
-		c.Request.Header.Add("content-length", strconv.Itoa(len(bodyBytes)))
-
-		process.ProxyRequest(c.Writer, c.Request)
+		c.String(statusCode, message)
 	}
 }
 
-func (pm *ProxyManager) proxyNoRouteHandler(c *gin.Context) {
-	// since maps are unordered, just use the first available process if one exists
-	for _, process := range pm.currentProcesses {
-		process.ProxyRequest(c.Writer, c.Request)
-		return
+func (pm *ProxyManager) unloadAllModelsHandler(c *gin.Context) {
+	pm.StopProcesses(StopImmediately)
+	c.String(http.StatusOK, "OK")
+}
+
+func (pm *ProxyManager) listRunningProcessesHandler(context *gin.Context) {
+	context.Header("Content-Type", "application/json")
+	runningProcesses := make([]gin.H, 0) // Default to an empty response.
+
+	for _, processGroup := range pm.processGroups {
+		for _, process := range processGroup.processes {
+			if process.CurrentState() == StateReady {
+				runningProcesses = append(runningProcesses, gin.H{
+					"model": process.ID,
+					"state": process.state,
+				})
+			}
+		}
 	}
 
-	c.AbortWithError(http.StatusBadRequest, fmt.Errorf("no strategy to handle request"))
+	// Put the results under the `running` key.
+	response := gin.H{
+		"running": runningProcesses,
+	}
+
+	context.JSON(http.StatusOK, response) // Always return 200 OK
+}
+
+func (pm *ProxyManager) findGroupByModelName(modelName string) *ProcessGroup {
+	for _, group := range pm.processGroups {
+		if group.HasMember(modelName) {
+			return group
+		}
+	}
+	return nil
 }

proxy/proxymanager_loghandlers.go

@@ -3,17 +3,36 @@ package proxy
 import (
 	"fmt"
 	"net/http"
+	"strings"
 
 	"github.com/gin-gonic/gin"
 )
 
 func (pm *ProxyManager) sendLogsHandlers(c *gin.Context) {
-	c.Header("Content-Type", "text/plain")
-	history := pm.logMonitor.GetHistory()
-	_, err := c.Writer.Write(history)
-	if err != nil {
-		c.AbortWithError(http.StatusInternalServerError, err)
-		return
+	accept := c.GetHeader("Accept")
+	if strings.Contains(accept, "text/html") {
+		// Set the Content-Type header to text/html
+		c.Header("Content-Type", "text/html")
+
+		// Write the embedded HTML content to the response
+		logsHTML, err := getHTMLFile("logs.html")
+		if err != nil {
+			c.String(http.StatusInternalServerError, err.Error())
+			return
+		}
+		_, err = c.Writer.Write(logsHTML)
+		if err != nil {
+			c.String(http.StatusInternalServerError, fmt.Sprintf("failed to write response: %v", err))
+			return
+		}
+	} else {
+		c.Header("Content-Type", "text/plain")
+		history := pm.muxLogger.GetHistory()
+		_, err := c.Writer.Write(history)
+		if err != nil {
+			c.AbortWithError(http.StatusInternalServerError, err)
+			return
+		}
 	}
 }
 
@@ -22,13 +41,19 @@ func (pm *ProxyManager) streamLogsHandler(c *gin.Context) {
 	c.Header("Transfer-Encoding", "chunked")
 	c.Header("X-Content-Type-Options", "nosniff")
 
-	ch := pm.logMonitor.Subscribe()
-	defer pm.logMonitor.Unsubscribe(ch)
+	logMonitorId := c.Param("logMonitorID")
+	logger, err := pm.getLogger(logMonitorId)
+	if err != nil {
+		c.String(http.StatusBadRequest, err.Error())
+		return
+	}
+	ch := logger.Subscribe()
+	defer logger.Unsubscribe(ch)
 
 	notify := c.Request.Context().Done()
 	flusher, ok := c.Writer.(http.Flusher)
 	if !ok {
-		c.AbortWithError(http.StatusInternalServerError, fmt.Errorf("Streaming unsupported"))
+		c.AbortWithError(http.StatusInternalServerError, fmt.Errorf("streaming unsupported"))
 		return
 	}
 
@@ -36,13 +61,9 @@ func (pm *ProxyManager) streamLogsHandler(c *gin.Context) {
 	// Send history first if not skipped
 
 	if !skipHistory {
-		history := pm.logMonitor.GetHistory()
+		history := logger.GetHistory()
 		if len(history) != 0 {
-			_, err := c.Writer.Write(history)
-			if err != nil {
-				c.AbortWithError(http.StatusInternalServerError, err)
-				return
-			}
+			c.Writer.Write(history)
 			flusher.Flush()
 		}
 	}
@@ -53,7 +74,7 @@ func (pm *ProxyManager) streamLogsHandler(c *gin.Context) {
 		case msg := <-ch:
 			_, err := c.Writer.Write(msg)
 			if err != nil {
-				c.AbortWithError(http.StatusInternalServerError, err)
+				// just break the loop if we can't write for some reason
 				return
 			}
 			flusher.Flush()
@@ -69,15 +90,21 @@ func (pm *ProxyManager) streamLogsHandlerSSE(c *gin.Context) {
 	c.Header("Connection", "keep-alive")
 	c.Header("X-Content-Type-Options", "nosniff")
 
-	ch := pm.logMonitor.Subscribe()
-	defer pm.logMonitor.Unsubscribe(ch)
+	logMonitorId := c.Param("logMonitorID")
+	logger, err := pm.getLogger(logMonitorId)
+	if err != nil {
+		c.String(http.StatusBadRequest, err.Error())
+		return
+	}
+	ch := logger.Subscribe()
+	defer logger.Unsubscribe(ch)
 
 	notify := c.Request.Context().Done()
 
 	// Send history first if not skipped
 	_, skipHistory := c.GetQuery("no-history")
 	if !skipHistory {
-		history := pm.logMonitor.GetHistory()
+		history := logger.GetHistory()
 		if len(history) != 0 {
 			c.SSEvent("message", string(history))
 			c.Writer.Flush()
@@ -95,3 +122,21 @@ func (pm *ProxyManager) streamLogsHandlerSSE(c *gin.Context) {
 		}
 	}
 }
+
+// getLogger searches for the appropriate logger based on the logMonitorId
+func (pm *ProxyManager) getLogger(logMonitorId string) (*LogMonitor, error) {
+	var logger *LogMonitor
+
+	if logMonitorId == "" {
+		// maintain the default
+		logger = pm.muxLogger
+	} else if logMonitorId == "proxy" {
+		logger = pm.proxyLogger
+	} else if logMonitorId == "upstream" {
+		logger = pm.upstreamLogger
+	} else {
+		return nil, fmt.Errorf("invalid logger. Use 'proxy' or 'upstream'")
+	}
+
+	return logger, nil
+}

proxy/proxymanager_test.go

@@ -2,78 +2,126 @@ package proxy
 
 import (
 	"bytes"
+	"encoding/json"
 	"fmt"
+	"math/rand"
+	"mime/multipart"
 	"net/http"
 	"net/http/httptest"
+	"strconv"
 	"sync"
 	"testing"
 	"time"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/tidwall/gjson"
 )
 
 func TestProxyManager_SwapProcessCorrectly(t *testing.T) {
-	config := &Config{
+	config := AddDefaultGroupToConfig(Config{
 		HealthCheckTimeout: 15,
 		Models: map[string]ModelConfig{
 			"model1": getTestSimpleResponderConfig("model1"),
 			"model2": getTestSimpleResponderConfig("model2"),
 		},
-	}
+		LogLevel: "error",
+	})
 
 	proxy := New(config)
-	defer proxy.StopProcesses()
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
 
 	for _, modelName := range []string{"model1", "model2"} {
 		reqBody := fmt.Sprintf(`{"model":"%s"}`, modelName)
 		req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
 		w := httptest.NewRecorder()
 
-		proxy.HandlerFunc(w, req)
+		proxy.ServeHTTP(w, req)
 		assert.Equal(t, http.StatusOK, w.Code)
 		assert.Contains(t, w.Body.String(), modelName)
-
-		_, exists := proxy.currentProcesses["/"+modelName]
-		assert.True(t, exists, "expected %s key in currentProcesses", modelName)
-
 	}
-
-	// make sure there's only one loaded model
-	assert.Len(t, proxy.currentProcesses, 1)
 }
 
 func TestProxyManager_SwapMultiProcess(t *testing.T) {
-	config := &Config{
+	config := AddDefaultGroupToConfig(Config{
 		HealthCheckTimeout: 15,
 		Models: map[string]ModelConfig{
 			"model1": getTestSimpleResponderConfig("model1"),
 			"model2": getTestSimpleResponderConfig("model2"),
 		},
-		Profiles: map[string][]string{
-			"test": {"model1", "model2"},
+		LogLevel: "error",
+		Groups: map[string]GroupConfig{
+			"G1": {
+				Swap:      true,
+				Exclusive: false,
+				Members:   []string{"model1"},
+			},
+			"G2": {
+				Swap:      true,
+				Exclusive: false,
+				Members:   []string{"model2"},
+			},
 		},
-	}
+	})
 
 	proxy := New(config)
-	defer proxy.StopProcesses()
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
 
-	for modelID, requestedModel := range map[string]string{"model1": "test/model1", "model2": "test/model2"} {
+	tests := []string{"model1", "model2"}
+	for _, requestedModel := range tests {
+		t.Run(requestedModel, func(t *testing.T) {
+			reqBody := fmt.Sprintf(`{"model":"%s"}`, requestedModel)
+			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+			w := httptest.NewRecorder()
+
+			proxy.ServeHTTP(w, req)
+			assert.Equal(t, http.StatusOK, w.Code)
+			assert.Contains(t, w.Body.String(), requestedModel)
+		})
+	}
+
+	// make sure there's two loaded models
+	assert.Equal(t, proxy.findGroupByModelName("model1").processes["model1"].CurrentState(), StateReady)
+	assert.Equal(t, proxy.findGroupByModelName("model2").processes["model2"].CurrentState(), StateReady)
+}
+
+// Test that a persistent group is not affected by the swapping behaviour of
+// other groups.
+func TestProxyManager_PersistentGroupsAreNotSwapped(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"), // goes into the default group
+			"model2": getTestSimpleResponderConfig("model2"),
+		},
+		LogLevel: "error",
+		Groups: map[string]GroupConfig{
+			// the forever group is persistent and should not be affected by model1
+			"forever": {
+				Swap:       true,
+				Exclusive:  false,
+				Persistent: true,
+				Members:    []string{"model2"},
+			},
+		},
+	})
+
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+	// make requests to load all models, loading model1 should not affect model2
+	tests := []string{"model2", "model1"}
+	for _, requestedModel := range tests {
 		reqBody := fmt.Sprintf(`{"model":"%s"}`, requestedModel)
 		req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
 		w := httptest.NewRecorder()
 
-		proxy.HandlerFunc(w, req)
+		proxy.ServeHTTP(w, req)
 		assert.Equal(t, http.StatusOK, w.Code)
-		assert.Contains(t, w.Body.String(), modelID)
+		assert.Contains(t, w.Body.String(), requestedModel)
 	}
 
-	// make sure there's two loaded models
-	assert.Len(t, proxy.currentProcesses, 2)
-	_, exists := proxy.currentProcesses["test/model1"]
-	assert.True(t, exists, "expected test/model1 key in currentProcesses")
-
-	_, exists = proxy.currentProcesses["test/model2"]
-	assert.True(t, exists, "expected test/model2 key in currentProcesses")
+	assert.Equal(t, proxy.findGroupByModelName("model2").processes["model2"].CurrentState(), StateReady)
+	assert.Equal(t, proxy.findGroupByModelName("model1").processes["model1"].CurrentState(), StateReady)
 }
 
 // When a request for a different model comes in ProxyManager should wait until
@@ -83,17 +131,18 @@ func TestProxyManager_SwapMultiProcessParallelRequests(t *testing.T) {
 		t.Skip("skipping slow test")
 	}
 
-	config := &Config{
+	config := AddDefaultGroupToConfig(Config{
 		HealthCheckTimeout: 15,
 		Models: map[string]ModelConfig{
 			"model1": getTestSimpleResponderConfig("model1"),
 			"model2": getTestSimpleResponderConfig("model2"),
 			"model3": getTestSimpleResponderConfig("model3"),
 		},
-	}
+		LogLevel: "error",
+	})
 
 	proxy := New(config)
-	defer proxy.StopProcesses()
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
 
 	results := map[string]string{}
 
@@ -109,15 +158,16 @@ func TestProxyManager_SwapMultiProcessParallelRequests(t *testing.T) {
 			req := httptest.NewRequest("POST", "/v1/chat/completions?wait=1000ms", bytes.NewBufferString(reqBody))
 			w := httptest.NewRecorder()
 
-			proxy.HandlerFunc(w, req)
+			proxy.ServeHTTP(w, req)
 
 			if w.Code != http.StatusOK {
 				t.Errorf("Expected status OK, got %d for key %s", w.Code, key)
 			}
 
 			mu.Lock()
-
-			results[key] = w.Body.String()
+			var response map[string]string
+			assert.NoError(t, json.Unmarshal(w.Body.Bytes(), &response))
+			results[key] = response["responseMessage"]
 			mu.Unlock()
 		}(key)
 
@@ -131,3 +181,445 @@ func TestProxyManager_SwapMultiProcessParallelRequests(t *testing.T) {
 		assert.Equal(t, key, result)
 	}
 }
+
+func TestProxyManager_ListModelsHandler(t *testing.T) {
+	config := Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+			"model2": getTestSimpleResponderConfig("model2"),
+			"model3": getTestSimpleResponderConfig("model3"),
+		},
+		LogLevel: "error",
+	}
+
+	proxy := New(config)
+
+	// Create a test request
+	req := httptest.NewRequest("GET", "/v1/models", nil)
+	req.Header.Add("Origin", "i-am-the-origin")
+	w := httptest.NewRecorder()
+
+	// Call the listModelsHandler
+	proxy.ServeHTTP(w, req)
+
+	// Check the response status code
+	assert.Equal(t, http.StatusOK, w.Code)
+
+	// Check for Access-Control-Allow-Origin
+	assert.Equal(t, req.Header.Get("Origin"), w.Result().Header.Get("Access-Control-Allow-Origin"))
+
+	// Parse the JSON response
+	var response struct {
+		Data []map[string]interface{} `json:"data"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to parse JSON response: %v", err)
+	}
+
+	// Check the number of models returned
+	assert.Len(t, response.Data, 3)
+
+	// Check the details of each model
+	expectedModels := map[string]struct{}{
+		"model1": {},
+		"model2": {},
+		"model3": {},
+	}
+
+	for _, model := range response.Data {
+		modelID, ok := model["id"].(string)
+		assert.True(t, ok, "model ID should be a string")
+		_, exists := expectedModels[modelID]
+		assert.True(t, exists, "unexpected model ID: %s", modelID)
+		delete(expectedModels, modelID)
+
+		object, ok := model["object"].(string)
+		assert.True(t, ok, "object should be a string")
+		assert.Equal(t, "model", object)
+
+		created, ok := model["created"].(float64)
+		assert.True(t, ok, "created should be a number")
+		assert.Greater(t, created, float64(0)) // Assuming the timestamp is positive
+
+		ownedBy, ok := model["owned_by"].(string)
+		assert.True(t, ok, "owned_by should be a string")
+		assert.Equal(t, "llama-swap", ownedBy)
+	}
+
+	// Ensure all expected models were returned
+	assert.Empty(t, expectedModels, "not all expected models were returned")
+}
+
+func TestProxyManager_Shutdown(t *testing.T) {
+	// make broken model configurations
+	model1Config := getTestSimpleResponderConfigPort("model1", 9991)
+	model1Config.Proxy = "http://localhost:10001/"
+
+	model2Config := getTestSimpleResponderConfigPort("model2", 9992)
+	model2Config.Proxy = "http://localhost:10002/"
+
+	model3Config := getTestSimpleResponderConfigPort("model3", 9993)
+	model3Config.Proxy = "http://localhost:10003/"
+
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": model1Config,
+			"model2": model2Config,
+			"model3": model3Config,
+		},
+		LogLevel: "error",
+		Groups: map[string]GroupConfig{
+			"test": {
+				Swap:    false,
+				Members: []string{"model1", "model2", "model3"},
+			},
+		},
+	})
+
+	proxy := New(config)
+
+	// Start all the processes
+	var wg sync.WaitGroup
+	for _, modelName := range []string{"model1", "model2", "model3"} {
+		wg.Add(1)
+		go func(modelName string) {
+			defer wg.Done()
+			reqBody := fmt.Sprintf(`{"model":"%s"}`, modelName)
+			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+			w := httptest.NewRecorder()
+
+			// send a request to trigger the proxy to load ... this should hang waiting for start up
+			proxy.ServeHTTP(w, req)
+			assert.Equal(t, http.StatusBadGateway, w.Code)
+			assert.Contains(t, w.Body.String(), "health check interrupted due to shutdown")
+		}(modelName)
+	}
+
+	go func() {
+		<-time.After(time.Second)
+		proxy.Shutdown()
+	}()
+	wg.Wait()
+}
+
+func TestProxyManager_Unload(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+		},
+		LogLevel: "error",
+	})
+
+	proxy := New(config)
+	reqBody := fmt.Sprintf(`{"model":"%s"}`, "model1")
+	req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+	w := httptest.NewRecorder()
+	proxy.ServeHTTP(w, req)
+
+	assert.Equal(t, proxy.processGroups[DEFAULT_GROUP_ID].processes["model1"].CurrentState(), StateReady)
+	req = httptest.NewRequest("GET", "/unload", nil)
+	w = httptest.NewRecorder()
+	proxy.ServeHTTP(w, req)
+	assert.Equal(t, http.StatusOK, w.Code)
+	assert.Equal(t, w.Body.String(), "OK")
+
+	// give it a bit of time to stop
+	<-time.After(time.Millisecond * 250)
+	assert.Equal(t, proxy.processGroups[DEFAULT_GROUP_ID].processes["model1"].CurrentState(), StateStopped)
+}
+
+// Test issue #61 `Listing the current list of models and the loaded model.`
+func TestProxyManager_RunningEndpoint(t *testing.T) {
+	// Shared configuration
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+			"model2": getTestSimpleResponderConfig("model2"),
+		},
+		LogLevel: "warn",
+	})
+
+	// Define a helper struct to parse the JSON response.
+	type RunningResponse struct {
+		Running []struct {
+			Model string `json:"model"`
+			State string `json:"state"`
+		} `json:"running"`
+	}
+
+	// Create proxy once for all tests
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+	t.Run("no models loaded", func(t *testing.T) {
+		req := httptest.NewRequest("GET", "/running", nil)
+		w := httptest.NewRecorder()
+		proxy.ServeHTTP(w, req)
+
+		assert.Equal(t, http.StatusOK, w.Code)
+
+		var response RunningResponse
+
+		// Check if this is a valid JSON object.
+		assert.NoError(t, json.Unmarshal(w.Body.Bytes(), &response))
+
+		// We should have an empty running array here.
+		assert.Empty(t, response.Running, "expected no running models")
+	})
+
+	t.Run("single model loaded", func(t *testing.T) {
+		// Load just a model.
+		reqBody := `{"model":"model1"}`
+		req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+		w := httptest.NewRecorder()
+		proxy.ServeHTTP(w, req)
+		assert.Equal(t, http.StatusOK, w.Code)
+
+		// Simulate browser call for the `/running` endpoint.
+		req = httptest.NewRequest("GET", "/running", nil)
+		w = httptest.NewRecorder()
+		proxy.ServeHTTP(w, req)
+
+		var response RunningResponse
+		assert.NoError(t, json.Unmarshal(w.Body.Bytes(), &response))
+
+		// Check if we have a single array element.
+		assert.Len(t, response.Running, 1)
+
+		// Is this the right model?
+		assert.Equal(t, "model1", response.Running[0].Model)
+
+		// Is the model loaded?
+		assert.Equal(t, "ready", response.Running[0].State)
+	})
+}
+
+func TestProxyManager_AudioTranscriptionHandler(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"TheExpectedModel": getTestSimpleResponderConfig("TheExpectedModel"),
+		},
+		LogLevel: "error",
+	})
+
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+	// Create a buffer with multipart form data
+	var b bytes.Buffer
+	w := multipart.NewWriter(&b)
+
+	// Add the model field
+	fw, err := w.CreateFormField("model")
+	assert.NoError(t, err)
+	_, err = fw.Write([]byte("TheExpectedModel"))
+	assert.NoError(t, err)
+
+	// Add a file field
+	fw, err = w.CreateFormFile("file", "test.mp3")
+	assert.NoError(t, err)
+	// Generate random content length between 10 and 20
+	contentLength := rand.Intn(11) + 10 // 10 to 20
+	content := make([]byte, contentLength)
+	_, err = fw.Write(content)
+	assert.NoError(t, err)
+	w.Close()
+
+	// Create the request with the multipart form data
+	req := httptest.NewRequest("POST", "/v1/audio/transcriptions", &b)
+	req.Header.Set("Content-Type", w.FormDataContentType())
+	rec := httptest.NewRecorder()
+	proxy.ServeHTTP(rec, req)
+
+	// Verify the response
+	assert.Equal(t, http.StatusOK, rec.Code)
+	var response map[string]string
+	err = json.Unmarshal(rec.Body.Bytes(), &response)
+	assert.NoError(t, err)
+	assert.Equal(t, "TheExpectedModel", response["model"])
+	assert.Equal(t, response["text"], fmt.Sprintf("The length of the file is %d bytes", contentLength)) // matches simple-responder
+	assert.Equal(t, strconv.Itoa(370+contentLength), response["h_content_length"])
+}
+
+// Test useModelName in configuration sends overrides what is sent to upstream
+func TestProxyManager_UseModelName(t *testing.T) {
+	upstreamModelName := "upstreamModel"
+	modelConfig := getTestSimpleResponderConfig(upstreamModelName)
+	modelConfig.UseModelName = upstreamModelName
+
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": modelConfig,
+		},
+		LogLevel: "error",
+	})
+
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+	requestedModel := "model1"
+
+	t.Run("useModelName over rides requested model: /v1/chat/completions", func(t *testing.T) {
+		reqBody := fmt.Sprintf(`{"model":"%s"}`, requestedModel)
+		req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+		w := httptest.NewRecorder()
+
+		proxy.ServeHTTP(w, req)
+		assert.Equal(t, http.StatusOK, w.Code)
+		assert.Contains(t, w.Body.String(), upstreamModelName)
+
+		// make sure the content length was set correctly
+		// simple-responder will return the content length it got in the response
+		body := w.Body.Bytes()
+		contentLength := int(gjson.GetBytes(body, "h_content_length").Int())
+		assert.Equal(t, len(fmt.Sprintf(`{"model":"%s"}`, upstreamModelName)), contentLength)
+	})
+
+	t.Run("useModelName over rides requested model: /v1/audio/transcriptions", func(t *testing.T) {
+		// Create a buffer with multipart form data
+		var b bytes.Buffer
+		w := multipart.NewWriter(&b)
+
+		// Add the model field
+		fw, err := w.CreateFormField("model")
+		assert.NoError(t, err)
+		_, err = fw.Write([]byte(requestedModel))
+		assert.NoError(t, err)
+
+		// Add a file field
+		fw, err = w.CreateFormFile("file", "test.mp3")
+		assert.NoError(t, err)
+		_, err = fw.Write([]byte("test"))
+		assert.NoError(t, err)
+		w.Close()
+
+		// Create the request with the multipart form data
+		req := httptest.NewRequest("POST", "/v1/audio/transcriptions", &b)
+		req.Header.Set("Content-Type", w.FormDataContentType())
+		rec := httptest.NewRecorder()
+		proxy.ServeHTTP(rec, req)
+
+		// Verify the response
+		assert.Equal(t, http.StatusOK, rec.Code)
+		var response map[string]string
+		err = json.Unmarshal(rec.Body.Bytes(), &response)
+		assert.NoError(t, err)
+		assert.Equal(t, upstreamModelName, response["model"])
+	})
+}
+
+func TestProxyManager_CORSOptionsHandler(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+		},
+		LogLevel: "error",
+	})
+
+	tests := []struct {
+		name            string
+		method          string
+		requestHeaders  map[string]string
+		expectedStatus  int
+		expectedHeaders map[string]string
+	}{
+		{
+			name:           "OPTIONS with no headers",
+			method:         "OPTIONS",
+			expectedStatus: http.StatusNoContent,
+			expectedHeaders: map[string]string{
+				"Access-Control-Allow-Origin":  "*",
+				"Access-Control-Allow-Methods": "GET, POST, PUT, PATCH, DELETE, OPTIONS",
+				"Access-Control-Allow-Headers": "Content-Type, Authorization, Accept, X-Requested-With",
+			},
+		},
+		{
+			name:   "OPTIONS with specific headers",
+			method: "OPTIONS",
+			requestHeaders: map[string]string{
+				"Access-Control-Request-Headers": "X-Custom-Header, Some-Other-Header",
+			},
+			expectedStatus: http.StatusNoContent,
+			expectedHeaders: map[string]string{
+				"Access-Control-Allow-Origin":  "*",
+				"Access-Control-Allow-Methods": "GET, POST, PUT, PATCH, DELETE, OPTIONS",
+				"Access-Control-Allow-Headers": "X-Custom-Header, Some-Other-Header",
+			},
+		},
+		{
+			name:           "Non-OPTIONS request",
+			method:         "GET",
+			expectedStatus: http.StatusNotFound, // Since we don't have a GET route defined
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			proxy := New(config)
+			defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+			req := httptest.NewRequest(tt.method, "/v1/chat/completions", nil)
+			for k, v := range tt.requestHeaders {
+				req.Header.Set(k, v)
+			}
+
+			w := httptest.NewRecorder()
+			proxy.ServeHTTP(w, req)
+
+			assert.Equal(t, tt.expectedStatus, w.Code)
+
+			for header, expectedValue := range tt.expectedHeaders {
+				assert.Equal(t, expectedValue, w.Header().Get(header))
+			}
+		})
+	}
+}
+
+func TestProxyManager_Upstream(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+		},
+		LogLevel: "error",
+	})
+
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+	req := httptest.NewRequest("GET", "/upstream/model1/test", nil)
+	rec := httptest.NewRecorder()
+	proxy.ServeHTTP(rec, req)
+	assert.Equal(t, http.StatusOK, rec.Code)
+	assert.Equal(t, "model1", rec.Body.String())
+}
+
+func TestProxyManager_ChatContentLength(t *testing.T) {
+	config := AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			"model1": getTestSimpleResponderConfig("model1"),
+		},
+		LogLevel: "error",
+	})
+
+	proxy := New(config)
+	defer proxy.StopProcesses(StopWaitForInflightRequest)
+
+	reqBody := fmt.Sprintf(`{"model":"%s", "x": "this is just some content to push the length out a bit"}`, "model1")
+	req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+	w := httptest.NewRecorder()
+
+	proxy.ServeHTTP(w, req)
+	assert.Equal(t, http.StatusOK, w.Code)
+	var response map[string]string
+	assert.NoError(t, json.Unmarshal(w.Body.Bytes(), &response))
+	assert.Equal(t, "81", response["h_content_length"])
+	assert.Equal(t, "model1", response["responseMessage"])
+}

proxy/sanitize_cors.go

@@ -0,0 +1,43 @@
+package proxy
+
+import (
+	"strings"
+)
+
+func isTokenChar(r rune) bool {
+	switch {
+	case r >= 'a' && r <= 'z':
+	case r >= 'A' && r <= 'Z':
+	case r >= '0' && r <= '9':
+	case strings.ContainsRune("!#$%&'*+-.^_`|~", r):
+	default:
+		return false
+	}
+	return true
+}
+
+func SanitizeAccessControlRequestHeaderValues(headerValues string) string {
+	parts := strings.Split(headerValues, ",")
+	valid := make([]string, 0, len(parts))
+
+	for _, p := range parts {
+		v := strings.TrimSpace(p)
+		if v == "" {
+			continue
+		}
+
+		validPart := true
+		for _, c := range v {
+			if !isTokenChar(c) {
+				validPart = false
+				break
+			}
+		}
+
+		if validPart {
+			valid = append(valid, v)
+		}
+	}
+
+	return strings.Join(valid, ", ")
+}

proxy/sanitize_cors_test.go

@@ -0,0 +1,77 @@
+package proxy
+
+import "testing"
+
+func TestSanitizeAccessControlRequestHeaderValues(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "empty string",
+			input:    "",
+			expected: "",
+		},
+		{
+			name:     "whitespace only",
+			input:    "   ",
+			expected: "",
+		},
+		{
+			name:     "single valid value",
+			input:    "content-type",
+			expected: "content-type",
+		},
+		{
+			name:     "multiple valid values",
+			input:    "content-type, authorization, x-requested-with",
+			expected: "content-type, authorization, x-requested-with",
+		},
+		{
+			name:     "values with extra spaces",
+			input:    "  content-type  ,  authorization  ",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "values with tabs",
+			input:    "content-type,\tauthorization",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "values with invalid characters",
+			input:    "content-type, auth\n, x-requested-with\r",
+			expected: "content-type, auth, x-requested-with",
+		},
+		{
+			name:     "empty values in list",
+			input:    "content-type,,authorization",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "leading and trailing commas",
+			input:    ",content-type,authorization,",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "mixed valid and invalid values",
+			input:    "content-type, \x00invalid, x-requested-with",
+			expected: "content-type, x-requested-with",
+		},
+		{
+			name:     "mixed case values",
+			input:    "Content-Type, my-Valid-Header, Another-hEader",
+			expected: "Content-Type, my-Valid-Header, Another-hEader",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := SanitizeAccessControlRequestHeaderValues(tt.input)
+			if got != tt.expected {
+				t.Errorf("SanitizeAccessControlRequestHeaderValues(%q) = %q, want %q",
+					tt.input, got, tt.expected)
+			}
+		})
+	}
+}