steve/llama-swap
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

proxy: add versionless API endpoint (#733)

Browse Source 
Add versionless endpoints under v/ to support upstream peers that 
do not use the v1/ prefix.

Fixes #728.
This commit is contained in:
Wim Vander Schelden
2026-05-03 22:47:38 +02:00
committed by GitHub
parent 11b7913287
commit e261745c66
3 changed files with 146 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
proxy/proxymanager.go
+15
View File
@@ -351,6 +351,16 @@ func (pm *ProxyManager) setupGinEngine() {
pm.ginEngine.POST("/v1/rerank", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v1/reranking", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
// Unversioned API endpoints, see issue #728
pm.ginEngine.POST("/v/chat/completions", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/responses", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/completions", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/messages", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/messages/count_tokens", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/embeddings", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/rerank", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
pm.ginEngine.POST("/v/reranking", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
// llama-server's /infill endpoint for code infilling
pm.ginEngine.POST("/infill", pm.apiKeyAuth(), pm.trackInflight(), llmHandler)
@@ -860,6 +870,11 @@ func (pm *ProxyManager) mkProxyJSONHandler(cf captureFields) func(*gin.Context)
c.Request.Header.Set("content-length", strconv.Itoa(len(bodyBytes)))
c.Request.ContentLength = int64(len(bodyBytes))
// issue #728 support versionless API requests
if strings.HasPrefix(c.Request.URL.Path, "/v/") {
c.Request.URL.Path = strings.TrimPrefix(c.Request.URL.Path, "/v")
}
// issue #366 extract values that downstream handlers may need
isStreaming := gjson.GetBytes(bodyBytes, "stream").Bool()
ctx := context.WithValue(c.Request.Context(), proxyCtxKey("streaming"), isStreaming)