small tweak to example config

The automatic ${MODEL_ID} macro includes the name of the model and can be used in Cmd and CmdStop.

README.md

@@ -7,7 +7,7 @@
 
 llama-swap is a light weight, transparent proxy server that provides automatic model swapping to llama.cpp's server.
 
-Written in golang, it is very easy to install (single binary with no dependencies) and configure (single yaml file). To get started, download a pre-built binary or use the provided docker images.
+Written in golang, it is very easy to install (single binary with no dependencies) and configure (single yaml file). To get started, download a pre-built binary, a provided docker images or Homebrew.
 
 ## Features:
 
@@ -18,9 +18,12 @@ Written in golang, it is very easy to install (single binary with no dependencie
   - `v1/completions`
   - `v1/chat/completions`
   - `v1/embeddings`
-  - `v1/rerank`, `v1/reranking`, `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-server (llama.cpp) supported endpoints:
+  - `v1/rerank`, `v1/reranking`, `/rerank`
+  - `/infill` - for code infilling
+  - `/completion` - for completion endpoint
 - ✅ llama-swap custom API endpoints
   - `/ui` - web UI
   - `/log` - remote log monitoring
@@ -204,4 +207,7 @@ For Python based inference servers like vllm or tabbyAPI it is recommended to ru
 
 ## Star History
 
+> [!NOTE]
+> ⭐️ Star this project to help others discover it! 
+
 [![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

@@ -49,8 +49,8 @@ macros:
 # - required
 # - each key is the model's ID, used in API requests
 # - model settings have default values that are used if they are not defined here
-# - below are examples of the various settings a model can have:
-# - available model settings: env, cmd, cmdStop, proxy, aliases, checkEndpoint, ttl, unlisted
+# - the model's ID is available in the ${MODEL_ID} macro, also available in macros defined above
+# - below are examples of the all the settings a model can have
 models:
 
   # keys are the model names used in API requests
@@ -129,6 +129,15 @@ models:
       # - recommended to stick to sampling parameters
       strip_params: "temperature, top_p, top_k"
 
+    # concurrencyLimit: overrides the allowed number of active parallel requests to a model
+    # - optional, default: 0
+    # - useful for limiting the number of active parallel requests a model can process
+    # - must be set per model
+    # - any number greater than 0 will override the internal default value of 10
+    # - any requests that exceeds the limit will receive an HTTP 429 Too Many Requests response
+    # - recommended to be omitted and the default used
+    concurrencyLimit: 0
+
   # Unlisted model example:
   "qwen-unlisted":
     # unlisted: boolean, true or false
@@ -139,12 +148,12 @@ models:
     cmd: llama-server --port ${PORT} -m Llama-3.2-1B-Instruct-Q4_K_M.gguf -ngl 0
 
   # Docker example:
-  # container run times like Docker and Podman can be used reliably with a
-  # a combination of cmd and cmdStop.
+  # container runtimes like Docker and Podman can be used reliably with
+  # a combination of cmd, cmdStop, and ${MODEL_ID}
   "docker-llama":
     proxy: "http://127.0.0.1:${PORT}"
     cmd: |
-      docker run --name dockertest
+      docker run --name ${MODEL_ID}
       --init --rm -p ${PORT}:8080 -v /mnt/nvme/models:/models
       ghcr.io/ggml-org/llama.cpp:server
       --model '/models/Qwen2.5-Coder-0.5B-Instruct-Q4_K_M.gguf'
@@ -158,7 +167,7 @@ models:
     # - on POSIX systems: a SIGTERM signal is sent
     # - on Windows, calls taskkill to stop the process
     # - processes have 5 seconds to shutdown until forceful termination is attempted
-    cmdStop: docker stop dockertest
+    cmdStop: docker stop ${MODEL_ID}
 
 # groups: a dictionary of group settings
 # - optional, default: empty dictionary

misc/simple-responder/simple-responder.go

@@ -153,6 +153,19 @@ func main() {
 
 	})
 
+	// llama-server compatibility: /completion
+	r.POST("/completion", func(c *gin.Context) {
+		c.Header("Content-Type", "application/json")
+		c.JSON(http.StatusOK, gin.H{
+			"responseMessage": *responseMessage,
+			"usage": gin.H{
+				"completion_tokens": 10,
+				"prompt_tokens":     25,
+				"total_tokens":      35,
+			},
+		})
+	})
+
 	// issue #41
 	r.POST("/v1/audio/transcriptions", func(c *gin.Context) {
 		// Parse the multipart form

proxy/config.go

@@ -237,7 +237,7 @@ func LoadConfigFromReader(r io.Reader) (Config, error) {
 
 	- name must fit the regex ^[a-zA-Z0-9_-]+$
 	- names must be less than 64 characters (no reason, just cause)
-	- name can not be any reserved macros: PORT
+	- name can not be any reserved macros: PORT, MODEL_ID
 	- macro values must be less than 1024 characters
 	*/
 	macroNameRegex := regexp.MustCompile(`^[a-zA-Z0-9_-]+$`)
@@ -253,6 +253,7 @@ func LoadConfigFromReader(r io.Reader) (Config, error) {
 		}
 		switch macroName {
 		case "PORT":
+		case "MODEL_ID":
 			return Config{}, fmt.Errorf("macro name '%s' is reserved and cannot be used", macroName)
 		}
 	}
@@ -296,6 +297,11 @@ func LoadConfigFromReader(r io.Reader) (Config, error) {
 			nextPort++
 		}
 
+		if strings.Contains(modelConfig.Cmd, "${MODEL_ID}") || strings.Contains(modelConfig.CmdStop, "${MODEL_ID}") {
+			modelConfig.Cmd = strings.ReplaceAll(modelConfig.Cmd, "${MODEL_ID}", modelId)
+			modelConfig.CmdStop = strings.ReplaceAll(modelConfig.CmdStop, "${MODEL_ID}", modelId)
+		}
+
 		// make sure there are no unknown macros that have not been replaced
 		macroPattern := regexp.MustCompile(`\$\{([a-zA-Z0-9_-]+)\}`)
 		fieldMap := map[string]string{

proxy/config_test.go

@@ -440,3 +440,44 @@ models:
 	expectedCmd := "/user/llama.cpp/build/bin/llama-server --port 9990 --model /path/to/model.gguf -ngl 99"
 	assert.Equal(t, expectedCmd, cmdStr, "Final command does not match expected structure")
 }
+
+func TestConfig_MacroModelId(t *testing.T) {
+	content := `
+startPort: 9000
+macros:
+  "docker-llama": docker run --name ${MODEL_ID} -p ${PORT}:8080 docker_img
+  "docker-stop": docker stop ${MODEL_ID}
+
+models:
+  model1:
+    cmd: /path/to/server -p ${PORT} -hf ${MODEL_ID}
+
+  model2:
+    cmd: ${docker-llama}
+    cmdStop: ${docker-stop}
+
+  author/model:F16:
+    cmd: /path/to/server -p ${PORT} -hf ${MODEL_ID}
+    cmdStop: stop
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	sanitizedCmd, err := SanitizeCommand(config.Models["model1"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "/path/to/server -p 9001 -hf model1", strings.Join(sanitizedCmd, " "))
+
+	assert.Equal(t, "docker stop ${MODEL_ID}", config.Macros["docker-stop"])
+
+	sanitizedCmd2, err := SanitizeCommand(config.Models["model2"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "docker run --name model2 -p 9002:8080 docker_img", strings.Join(sanitizedCmd2, " "))
+
+	sanitizedCmdStop, err := SanitizeCommand(config.Models["model2"].CmdStop)
+	assert.NoError(t, err)
+	assert.Equal(t, "docker stop model2", strings.Join(sanitizedCmdStop, " "))
+
+	sanitizedCmd3, err := SanitizeCommand(config.Models["author/model:F16"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "/path/to/server -p 9000 -hf author/model:F16", strings.Join(sanitizedCmd3, " "))
+}

proxy/metrics_middleware.go

@@ -5,12 +5,20 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"strings"
 	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/tidwall/gjson"
 )
 
+type MetricsRecorder struct {
+	metricsMonitor *MetricsMonitor
+	realModelName  string
+	//	isStreaming    bool
+	startTime time.Time
+}
+
 // MetricsMiddleware sets up the MetricsResponseWriter for capturing upstream requests
 func MetricsMiddleware(pm *ProxyManager) gin.HandlerFunc {
 	return func(c *gin.Context) {
@@ -41,49 +49,47 @@ func MetricsMiddleware(pm *ProxyManager) gin.HandlerFunc {
 			metricsRecorder: &MetricsRecorder{
 				metricsMonitor: pm.metricsMonitor,
 				realModelName:  realModelName,
-				isStreaming:    gjson.GetBytes(bodyBytes, "stream").Bool(),
 				startTime:      time.Now(),
 			},
 		}
 		c.Writer = writer
 		c.Next()
 
-		rec := writer.metricsRecorder
-		rec.processBody(writer.body)
-	}
-}
+		// check for streaming response
+		if strings.Contains(c.Writer.Header().Get("Content-Type"), "text/event-stream") {
+			writer.metricsRecorder.processStreamingResponse(writer.body)
+		} else {
+			writer.metricsRecorder.processNonStreamingResponse(writer.body)
+		}
 
-type MetricsRecorder struct {
-	metricsMonitor *MetricsMonitor
-	realModelName  string
-	isStreaming    bool
-	startTime      time.Time
-}
-
-// processBody handles response processing after request completes
-func (rec *MetricsRecorder) processBody(body []byte) {
-	if rec.isStreaming {
-		rec.processStreamingResponse(body)
-	} else {
-		rec.processNonStreamingResponse(body)
 	}
 }
 
 func (rec *MetricsRecorder) parseAndRecordMetrics(jsonData gjson.Result) bool {
 	usage := jsonData.Get("usage")
-	if !usage.Exists() {
+	timings := jsonData.Get("timings")
+	if !usage.Exists() && !timings.Exists() {
 		return false
 	}
 
 	// default values
-	outputTokens := int(jsonData.Get("usage.completion_tokens").Int())
-	inputTokens := int(jsonData.Get("usage.prompt_tokens").Int())
+	outputTokens := 0
+	inputTokens := 0
+
+	// timings data
 	tokensPerSecond := -1.0
 	promptPerSecond := -1.0
 	durationMs := int(time.Since(rec.startTime).Milliseconds())
 
+	if usage.Exists() {
+		outputTokens = int(jsonData.Get("usage.completion_tokens").Int())
+		inputTokens = int(jsonData.Get("usage.prompt_tokens").Int())
+	}
+
 	// use llama-server's timing data for tok/sec and duration as it is more accurate
-	if timings := jsonData.Get("timings"); timings.Exists() {
+	if timings.Exists() {
+		inputTokens = int(jsonData.Get("timings.prompt_n").Int())
+		outputTokens = int(jsonData.Get("timings.predicted_n").Int())
 		promptPerSecond = jsonData.Get("timings.prompt_per_second").Float()
 		tokensPerSecond = jsonData.Get("timings.predicted_per_second").Float()
 		durationMs = int(jsonData.Get("timings.prompt_ms").Float() + jsonData.Get("timings.predicted_ms").Float())

proxy/process.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"net/url"
 	"os/exec"
@@ -363,8 +364,18 @@ func (p *Process) stopCommand() {
 }
 
 func (p *Process) checkHealthEndpoint(healthURL string) error {
+
 	client := &http.Client{
-		Timeout: 500 * time.Millisecond,
+		// wait a short time for a tcp connection to be established
+		Transport: &http.Transport{
+			DialContext: (&net.Dialer{
+				Timeout: 500 * time.Millisecond,
+			}).DialContext,
+		},
+
+		// give a long time to respond to the health check endpoint
+		// after the connection is established. See issue: 276
+		Timeout: 5000 * time.Millisecond,
 	}
 
 	req, err := http.NewRequest("GET", healthURL, nil)

proxy/processgroup.go

@@ -60,10 +60,20 @@ func (pg *ProcessGroup) ProxyRequest(modelID string, writer http.ResponseWriter,
 	if pg.swap {
 		pg.Lock()
 		if pg.lastUsedProcess != modelID {
+
+			// is there something already running?
 			if pg.lastUsedProcess != "" {
 				pg.processes[pg.lastUsedProcess].Stop()
 			}
+
+			// wait for the request to the new model to be fully handled
+			// and prevent race conditions see issue #277
+			pg.processes[modelID].ProxyRequest(writer, request)
 			pg.lastUsedProcess = modelID
+
+			// short circuit and exit
+			pg.Unlock()
+			return nil
 		}
 		pg.Unlock()
 	}

proxy/processgroup_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"net/http"
 	"net/http/httptest"
+	"sync"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -44,32 +45,49 @@ func TestProcessGroup_HasMember(t *testing.T) {
 	assert.False(t, pg.HasMember("model3"))
 }
 
-func TestProcessGroup_ProxyRequestSwapIsTrue(t *testing.T) {
+// TestProcessGroup_ProxyRequestSwapIsTrueParallel tests that when swap is true
+// and multiple requests are made in parallel, only one process is running at a time.
+func TestProcessGroup_ProxyRequestSwapIsTrueParallel(t *testing.T) {
+	var processGroupTestConfig = AddDefaultGroupToConfig(Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]ModelConfig{
+			// use the same listening so if a model is already running, it will fail
+			// this is a way to test that swap isolation is working
+			// properly when there are parallel requests made at the
+			// same time.
+			"model1": getTestSimpleResponderConfigPort("model1", 9832),
+			"model2": getTestSimpleResponderConfigPort("model2", 9832),
+			"model3": getTestSimpleResponderConfigPort("model3", 9832),
+			"model4": getTestSimpleResponderConfigPort("model4", 9832),
+			"model5": getTestSimpleResponderConfigPort("model5", 9832),
+		},
+		Groups: map[string]GroupConfig{
+			"G1": {
+				Swap:    true,
+				Members: []string{"model1", "model2", "model3", "model4", "model5"},
+			},
+		},
+	})
+
 	pg := NewProcessGroup("G1", processGroupTestConfig, testLogger, testLogger)
 	defer pg.StopProcesses(StopWaitForInflightRequest)
 
-	tests := []string{"model1", "model2"}
+	tests := []string{"model1", "model2", "model3", "model4", "model5"}
 
+	var wg sync.WaitGroup
+
+	wg.Add(len(tests))
 	for _, modelName := range tests {
-		t.Run(modelName, func(t *testing.T) {
-			reqBody := `{"x", "y"}`
-			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+		go func(modelName string) {
+			defer wg.Done()
+			req := httptest.NewRequest("POST", "/v1/chat/completions", nil)
 			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)
-		})
+		}(modelName)
 	}
+	wg.Wait()
 }
 
 func TestProcessGroup_ProxyRequestSwapIsFalse(t *testing.T) {

proxy/proxymanager.go

@@ -191,11 +191,20 @@ func (pm *ProxyManager) setupGinEngine() {
 	// Support legacy /v1/completions api, see issue #12
 	pm.ginEngine.POST("/v1/completions", mm, pm.proxyOAIHandler)
 
-	// Support embeddings
+	// Support embeddings and reranking
 	pm.ginEngine.POST("/v1/embeddings", mm, pm.proxyOAIHandler)
+
+	// llama-server's /reranking endpoint + aliases
+	pm.ginEngine.POST("/reranking", mm, pm.proxyOAIHandler)
+	pm.ginEngine.POST("/rerank", mm, pm.proxyOAIHandler)
 	pm.ginEngine.POST("/v1/rerank", mm, pm.proxyOAIHandler)
 	pm.ginEngine.POST("/v1/reranking", mm, pm.proxyOAIHandler)
-	pm.ginEngine.POST("/rerank", mm, pm.proxyOAIHandler)
+
+	// llama-server's /infill endpoint for code infilling
+	pm.ginEngine.POST("/infill", mm, pm.proxyOAIHandler)
+
+	// llama-server's /completion endpoint
+	pm.ginEngine.POST("/completion", mm, pm.proxyOAIHandler)
 
 	// Support audio/speech endpoint
 	pm.ginEngine.POST("/v1/audio/speech", pm.proxyOAIHandler)

proxy/proxymanager_test.go

@@ -42,7 +42,6 @@ func TestProxyManager_SwapProcessCorrectly(t *testing.T) {
 		assert.Contains(t, w.Body.String(), modelName)
 	}
 }
-
 func TestProxyManager_SwapMultiProcess(t *testing.T) {
 	config := AddDefaultGroupToConfig(Config{
 		HealthCheckTimeout: 15,
@@ -834,6 +833,28 @@ func TestProxyManager_HealthEndpoint(t *testing.T) {
 	assert.Equal(t, "OK", rec.Body.String())
 }
 
+// Ensure the custom llama-server /completion endpoint proxies correctly
+func TestProxyManager_CompletionEndpoint(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 := `{"model":"model1"}`
+	req := httptest.NewRequest("POST", "/completion", bytes.NewBufferString(reqBody))
+	w := httptest.NewRecorder()
+
+	proxy.ServeHTTP(w, req)
+	assert.Equal(t, http.StatusOK, w.Code)
+	assert.Contains(t, w.Body.String(), "model1")
+}
+
 func TestProxyManager_StartupHooks(t *testing.T) {
 
 	// using real YAML as the configuration has gotten more complex