Compare commits

...

5 Commits

Author SHA1 Message Date
Benson Wong 671c1a5a7b update deps 2025-03-13 14:00:15 -07:00
Benson Wong 52c0196e0f clean up feature list in readme 2025-03-13 13:55:20 -07:00
Benson Wong 3201a68a04 Add /v1/audio/transcriptions support (#41)
* add support for /v1/audio/transcriptions
2025-03-13 13:49:39 -07:00
Florin-Gabriel Dumitru 3ac94ad20e Adds an endpoint '/running' (#61)
* Adds an endpoint '/running' that returns either an empty JSON object if no model has been loaded so far, or the last model loaded (model key) and it's current state (state key). Possible state values are: stopped, starting, ready and stopping.

* Improves the `/running` endpoint by allowing multiple entries under the `running` key within the JSON response.
Refactors the `/running` method name (listRunningProcessesHandler).
Removes the unlisted filter implementation.

* Adds tests for:
- no model loaded
- one model loaded
- multiple models loaded

* Adds simple comments.

* Simplified code structure as per 250313 comments on PR #65.

---------

Co-authored-by: FGDumitru|B <xelotx@gmail.com>
2025-03-13 13:42:59 -07:00
Benson Wong 60355bf74a fix some potentially confusing Process.start() comment 2025-03-11 11:00:45 -07:00
7 changed files with 375 additions and 13 deletions
+7 -4
View File
@@ -11,20 +11,23 @@ Written in golang, it is very easy to install (single binary with no dependancie
- ✅ Easy to deploy: single binary with no dependencies
- ✅ Easy to config: single yaml file
- ✅ On-demand model switching
- ✅ Full control over server settings per model
- ✅ 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 `profiles` ([docs](https://github.com/mostlygeek/llama-swap/issues/53#issuecomment-2660761741))
- ✅ Remote log monitoring at `/log`
- ✅ Direct access to upstream HTTP server via `/upstream/:model_id` ([demo](https://github.com/mostlygeek/llama-swap/pull/31))
- ✅ Manually unload models via `/unload` endpoint ([#58](https://github.com/mostlygeek/llama-swap/issues/58))
- ✅ 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?
+8 -8
View File
@@ -3,7 +3,11 @@ module github.com/mostlygeek/llama-swap
go 1.23.0
require (
github.com/gin-gonic/gin v1.10.0
github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510
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
)
@@ -15,12 +19,10 @@ 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
github.com/goccy/go-json v0.10.2 // indirect
github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510 // indirect
github.com/json-iterator/go v1.1.12 // indirect
github.com/klauspost/cpuid/v2 v2.2.7 // indirect
github.com/leodido/go-urn v1.4.0 // indirect
@@ -29,16 +31,14 @@ 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/gjson v1.18.0 // indirect
github.com/tidwall/match v1.1.1 // indirect
github.com/tidwall/pretty v1.2.1 // indirect
github.com/tidwall/sjson v1.2.5 // 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 v0.31.0 // indirect
golang.org/x/net v0.33.0 // indirect
golang.org/x/sys v0.28.0 // indirect
golang.org/x/text v0.21.0 // indirect
golang.org/x/crypto v0.36.0 // indirect
golang.org/x/net v0.37.0 // 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
)
+8
View File
@@ -78,20 +78,28 @@ 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/crypto v0.31.0 h1:ihbySMvVjLAeSH1IbfcRTkD/iNscyz8rGzjF/E5hV6U=
golang.org/x/crypto v0.31.0/go.mod h1:kDsLvtWBEx7MV9tJOj9bnXsPbxwJQ6csT/x4KIN4Ssk=
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 v0.25.0 h1:d/OCCoBEUq33pjydKrGQhw7IlUPI2Oylr+8qLx49kac=
golang.org/x/net v0.25.0/go.mod h1:JkAGAh7GEvH74S6FOH42FLoXpXbE/aqXSrIQjXgsiwM=
golang.org/x/net v0.33.0 h1:74SYHlV8BIgHIFC/LrYkOGIwL19eTYXQ5wc6TBuO36I=
golang.org/x/net v0.33.0/go.mod h1:HXLR5J+9DxmrqMwG9qjGCxZ+zKXxBru04zlTvWlWuN4=
golang.org/x/net v0.37.0 h1:1zLorHbz+LYj7MQlSf1+2tPIIgibq2eL5xkrGk6f+2c=
golang.org/x/net v0.37.0/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/sys v0.28.0 h1:Fksou7UEQUWlKvIdsqzJmUmCX3cZuD2+P3XyyzwMhlA=
golang.org/x/sys v0.28.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
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 v0.15.0 h1:h1V/4gjBv8v9cjcR6+AR5+/cIYK5N/WAgiv4xlsEtAk=
golang.org/x/text v0.15.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
golang.org/x/text v0.21.0 h1:zyQAAkrwaneQ066sspRyJaG9VNi/YJ1NfzcGB3hZ/qo=
golang.org/x/text v0.21.0/go.mod h1:4IBbMaMmOPCJ8SecivzSH54+73PCFmPWxNTLm+vZkEQ=
golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=
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=
+40
View File
@@ -67,6 +67,46 @@ func main() {
c.String(200, *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,
})
})
r.GET("/slow-respond", func(c *gin.Context) {
echo := c.Query("echo")
delay := c.Query("delay")
+4 -1
View File
@@ -116,7 +116,9 @@ func (p *Process) start() error {
return fmt.Errorf("can not start(), upstream proxy missing")
}
// wait for the other start() to complete
// multiple start() calls will wait for the one that is actually starting to
// complete before proceeding.
// ===========
curState := p.CurrentState()
if curState == StateReady {
@@ -132,6 +134,7 @@ func (p *Process) start() error {
return nil
}
// ===========
// There is the possibility of a hard to replicate race condition where
// curState *WAS* StateStopped but by the time we get to the p.stateMutex.Lock()
+125
View File
@@ -5,6 +5,7 @@ import (
"encoding/json"
"fmt"
"io"
"mime/multipart"
"net/http"
"sort"
"strconv"
@@ -95,6 +96,7 @@ func New(config *Config) *ProxyManager {
// Support audio/speech endpoint
pm.ginEngine.POST("/v1/audio/speech", pm.proxyOAIHandler)
pm.ginEngine.POST("/v1/audio/transcriptions", pm.proxyOAIAudioTranscriptionHandler)
pm.ginEngine.GET("/v1/models", pm.listModelsHandler)
@@ -108,6 +110,8 @@ func New(config *Config) *ProxyManager {
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")
@@ -372,6 +376,105 @@ func (pm *ProxyManager) proxyOAIHandler(c *gin.Context) {
}
}
func (pm *ProxyManager) proxyOAIAudioTranscriptionHandler(c *gin.Context) {
// 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)
// 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
}
// Swap to the requested model
process, err := pm.swapModel(requestedModel)
if err != nil {
pm.sendErrorResponse(c, http.StatusNotFound, fmt.Sprintf("unable to swap to model, %s", err.Error()))
return
}
// Get profile name and model name from the requested model
profileName, modelName := splitRequestedModel(requestedModel)
// 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" && profileName != "" {
fieldValue = modelName
}
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
}
}
}
// 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
}
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 {
pm.sendErrorResponse(c, http.StatusInternalServerError, "error creating modified request")
return
}
// Copy the headers from the original request
modifiedReq.Header = c.Request.Header.Clone()
modifiedReq.Header.Set("Content-Type", multipartWriter.FormDataContentType())
// Use the modified request for proxying
process.ProxyRequest(c.Writer, modifiedReq)
}
func (pm *ProxyManager) sendErrorResponse(c *gin.Context, statusCode int, message string) {
acceptHeader := c.GetHeader("Accept")
@@ -387,6 +490,28 @@ func (pm *ProxyManager) unloadAllModelsHandler(c *gin.Context) {
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 _, process := range pm.currentProcesses {
// Append the process ID and State (multiple entries if profiles are being used).
runningProcesses = append(runningProcesses, gin.H{
"model": process.ID,
"state": process.state,
})
}
// Put the results under the `running` key.
response := gin.H{
"running": runningProcesses,
}
context.JSON(http.StatusOK, response) // Always return 200 OK
}
func ProcessKeyName(groupName, modelName string) string {
return groupName + PROFILE_SPLIT_CHAR + modelName
}
+183
View File
@@ -4,6 +4,8 @@ import (
"bytes"
"encoding/json"
"fmt"
"math/rand"
"mime/multipart"
"net/http"
"net/http/httptest"
"sync"
@@ -349,3 +351,184 @@ func TestProxyManager_StripProfileSlug(t *testing.T) {
assert.Equal(t, http.StatusOK, w.Code)
assert.Contains(t, w.Body.String(), "ok")
}
// Test issue #61 `Listing the current list of models and the loaded model.`
func TestProxyManager_RunningEndpoint(t *testing.T) {
// Shared configuration
config := &Config{
HealthCheckTimeout: 15,
Models: map[string]ModelConfig{
"model1": getTestSimpleResponderConfig("model1"),
"model2": getTestSimpleResponderConfig("model2"),
},
Profiles: map[string][]string{
"test": {"model1", "model2"},
},
}
// 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()
t.Run("no models loaded", func(t *testing.T) {
req := httptest.NewRequest("GET", "/running", nil)
w := httptest.NewRecorder()
proxy.HandlerFunc(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.HandlerFunc(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.HandlerFunc(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)
})
t.Run("multiple models via profile", func(t *testing.T) {
// Load more than one model.
for _, model := range []string{"model1", "model2"} {
profileModel := ProcessKeyName("test", model)
reqBody := fmt.Sprintf(`{"model":"%s"}`, profileModel)
req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
w := httptest.NewRecorder()
proxy.HandlerFunc(w, req)
assert.Equal(t, http.StatusOK, w.Code)
}
// Simulate the browser call.
req := httptest.NewRequest("GET", "/running", nil)
w := httptest.NewRecorder()
proxy.HandlerFunc(w, req)
var response RunningResponse
// The JSON response must be valid.
assert.NoError(t, json.Unmarshal(w.Body.Bytes(), &response))
// The response should contain 2 models.
assert.Len(t, response.Running, 2)
expectedModels := map[string]struct{}{
"model1": {},
"model2": {},
}
// Iterate through the models and check their states as well.
for _, entry := range response.Running {
_, exists := expectedModels[entry.Model]
assert.True(t, exists, "unexpected model %s", entry.Model)
assert.Equal(t, "ready", entry.State)
delete(expectedModels, entry.Model)
}
// Since we deleted each model while testing for its validity we should have no more models in the response.
assert.Empty(t, expectedModels, "unexpected additional models in response")
})
}
func TestProxyManager_AudioTranscriptionHandler(t *testing.T) {
config := &Config{
HealthCheckTimeout: 15,
Profiles: map[string][]string{
"test": {"TheExpectedModel"},
},
Models: map[string]ModelConfig{
"TheExpectedModel": getTestSimpleResponderConfig("TheExpectedModel"),
},
}
proxy := New(config)
defer proxy.StopProcesses()
testCases := []struct {
name string
modelInput string
expectModel string
}{
{
name: "With Profile Prefix",
modelInput: "test:TheExpectedModel",
expectModel: "TheExpectedModel", // Profile prefix should be stripped
},
{
name: "Without Profile Prefix",
modelInput: "TheExpectedModel",
expectModel: "TheExpectedModel", // Should remain the same
},
}
for _, tc := range testCases {
t.Run(tc.name, 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(tc.modelInput))
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.HandlerFunc(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, tc.expectModel, response["model"])
assert.Equal(t, response["text"], fmt.Sprintf("The length of the file is %d bytes", contentLength)) // matches simple-responder
})
}
}