.

meta_ui.py

@@ -47,10 +47,10 @@ def backend(text, sampling, api_suffix):
     url = f"http://localhost:30000{api_suffix}"
     if api_suffix == "/generate":
         payload = {"model": MODEL_NAME, "text": text, "sampling_params": sampling}
-    else:  # "/v1/chat/completions"
+    elif api_suffix == "/v1/completions":
         payload = {
             "model": MODEL_NAME,
-            "messages": [{"role": "user", "content": text}],
+            "prompt": text,
             **sampling
         }
 
@@ -70,9 +70,9 @@ def backend(text, sampling, api_suffix):
             meta = data.get("meta_info", {})
             fr = meta.get("finish_reason")
             ctok = meta.get("completion_tokens")
-        else:
+        elif api_suffix == "/v1/completions":
             choice = data.get("choices", [{}])[0]
-            txt = choice.get("message", {}).get("content", "").strip()
+            txt = choice.get("text", "").strip()
             fr = choice.get("finish_reason")
             ctok = data.get("usage", {}).get("completion_tokens")
 
@@ -100,7 +100,7 @@ def chat(
 
     stop = [s.strip() for s in stop_raw.split(",") if s.strip()] or None
     samp = {
-        ("max_tokens" if api_suffix == "/v1/chat/completions" else "max_new_tokens"): int(max_new),
+        ("max_tokens" if api_suffix == "/v1/completions" else "max_new_tokens"): int(max_new),
         "temperature": temp,
         "top_p": top_p,
         "top_k": int(top_k),
@@ -121,7 +121,7 @@ def chat(
     while True:
         try:
             result = result_q.get(timeout=0.1)
-            # ★ 不论 /generate（纯 str）还是 /v1/chat/completions（dict），都转成 {"text": ...}
+            # ★ 不论 /generate（纯 str）还是 /v1/completions（dict），都转成 {"text": ...}
             if isinstance(result, str):
                 result = {"text": result}
             elif not isinstance(result, dict) or "text" not in result:
@@ -137,7 +137,7 @@ with gr.Blocks(title="调试界面") as demo:
     gr.Markdown(f"## 💬 调试界面  \n权重 **{MODEL_PATH.name}**")
 
     with gr.Row():
-        api_choice = gr.Dropdown(choices=["/generate", "/v1/chat/completions"],
+        api_choice = gr.Dropdown(choices=["/generate", "/v1/completions"],
                                 value="/generate", label="选择推理接口")
     with gr.Row():
         max_new = gr.Slider(32, 32768, 128, label="max_new_tokens")