From 6fe37c3abf35577ad43a9b4ddc3efab4bec06086 Mon Sep 17 00:00:00 2001
From: Benson Wong <mostlygeek@gmail.com>
Date: Tue, 17 Dec 2024 17:23:26 -0800
Subject: [PATCH] support /v1/embeddings (#4)

---
 config.example.yaml   | 12 ++++++++++++
 proxy/proxymanager.go |  3 +++
 2 files changed, 15 insertions(+)

diff --git a/config.example.yaml b/config.example.yaml
index 093ff9e..b8cd5e8 100644
--- a/config.example.yaml
+++ b/config.example.yaml
@@ -26,6 +26,18 @@ models:
     aliases:
     - gpt-3.5-turbo
 
+  "nomic":
+    proxy: http://127.0.0.1:9005
+    cmd: >
+      models/llama-server-osx --port 9005
+      -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
+
   "simple":
     # example of setting environment variables
     env:
diff --git a/proxy/proxymanager.go b/proxy/proxymanager.go
index b418243..e7f8c00 100644
--- a/proxy/proxymanager.go
+++ b/proxy/proxymanager.go
@@ -49,6 +49,9 @@ func New(config *Config) *ProxyManager {
 	// Set up routes using the Gin engine
 	pm.ginEngine.POST("/v1/chat/completions", pm.proxyChatRequestHandler)
 
+	// Support embeddings
+	pm.ginEngine.POST("/v1/embeddings", pm.proxyChatRequestHandler)
+
 	// Support legacy /v1/completions api, see issue #12
 	pm.ginEngine.POST("/v1/completions", pm.proxyChatRequestHandler)