Update app.py

app.py

@@ -4,7 +4,6 @@ import json
 import uuid
 import torch
 import spaces
-import uvicorn
 from fastapi import FastAPI
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
@@ -16,19 +15,18 @@ import gradio as gr
 
 MODEL_ID = "Qwen/Qwen2.5-Coder-32B-Instruct-AWQ"
 
-# 1. Cargar Tokenizer
+# Cargar Tokenizer
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
 
-# 2. Cargar modelo directamente con AutoAWQ
+# Cargar Modelo AWQ
 model = AutoAWQForCausalLM.from_quantized(
     MODEL_ID,
-    fuse_layers=True,
+    fuse_layers=False,
     trust_remote_code=True,
     safetensors=True
 )
 
-app = FastAPI(title="Qwen2.5-Coder-32B OpenAI API")
-
+# Esquemas para la API OpenAI
 class ChatMessage(BaseModel):
     role: str
     content: str
@@ -41,6 +39,7 @@ class ChatCompletionRequest(BaseModel):
     max_tokens: Optional[int] = 2048
     stream: Optional[bool] = False
 
+# Generación en GPU con ZeroGPU
 @spaces.GPU(duration=120)
 def generate_stream_tokens(messages_dict: List[Dict[str, str]], temperature: float, top_p: float, max_tokens: int):
     text = tokenizer.apply_chat_template(
@@ -73,7 +72,41 @@ def generate_stream_tokens(messages_dict: List[Dict[str, str]], temperature: flo
     for new_token in streamer:
         yield new_token
 
-@app.get("/v1/models")
+# Funciones de soporte para Gradio UI
+def gradio_generate(message, history, system_prompt, temperature, top_p, max_tokens):
+    messages = []
+    if system_prompt:
+        messages.append({"role": "system", "content": system_prompt})
+    for user_msg, bot_msg in history:
+        messages.append({"role": "user", "content": user_msg})
+        if bot_msg:
+            messages.append({"role": "assistant", "content": bot_msg})
+    messages.append({"role": "user", "content": message})
+
+    partial_text = ""
+    for token in generate_stream_tokens(messages, temperature, top_p, int(max_tokens)):
+        partial_text += token
+        yield partial_text
+
+# Definición de la interfaz Gradio
+with gr.Blocks(title="Qwen2.5-Coder-32B API & UI") as demo:
+    gr.Markdown("# 🚀 Qwen2.5-Coder-32B-Instruct (AWQ en ZeroGPU)")
+    gr.Markdown("Compatible con OpenAI: `/v1/chat/completions` | Modelo: `Qwen/Qwen2.5-Coder-32B-Instruct-AWQ`")
+    
+    gr.ChatInterface(
+        fn=gradio_generate,
+        additional_inputs=[
+            gr.Textbox("Eres un asistente de programación experto.", label="System Prompt"),
+            gr.Slider(0.0, 1.0, 0.2, step=0.05, label="Temperature"),
+            gr.Slider(0.1, 1.0, 0.9, step=0.05, label="Top-P"),
+            gr.Slider(256, 4096, 2048, step=256, label="Max New Tokens")
+        ]
+    )
+
+# Definición de FastAPI y vinculación de endpoints a la app subyacente de Gradio
+fastapi_app = demo.app
+
+@fastapi_app.get("/v1/models")
 async def list_models():
     return {
         "object": "list",
@@ -87,7 +120,7 @@ async def list_models():
         ]
     }
 
-@app.post("/v1/chat/completions")
+@fastapi_app.post("/v1/chat/completions")
 async def chat_completions(req: ChatCompletionRequest):
     req_id = f"chatcmpl-{uuid.uuid4().hex[:12]}"
     created_time = int(time.time())
@@ -146,35 +179,5 @@ async def chat_completions(req: ChatCompletionRequest):
         "usage": {"prompt_tokens": -1, "completion_tokens": -1, "total_tokens": -1}
     }
 
-def gradio_generate(message, history, system_prompt, temperature, top_p, max_tokens):
-    messages = []
-    if system_prompt:
-        messages.append({"role": "system", "content": system_prompt})
-    for user_msg, bot_msg in history:
-        messages.append({"role": "user", "content": user_msg})
-        if bot_msg:
-            messages.append({"role": "assistant", "content": bot_msg})
-    messages.append({"role": "user", "content": message})
-
-    partial_text = ""
-    for token in generate_stream_tokens(messages, temperature, top_p, int(max_tokens)):
-        partial_text += token
-        yield partial_text
-
-with gr.Blocks(title="Qwen2.5-Coder-32B API & UI", theme=gr.themes.Soft()) as gradio_app:
-    gr.Markdown("# 🚀 Qwen2.5-Coder-32B-Instruct (AWQ en ZeroGPU)")
-    gr.Markdown("Endpoint OpenAI: `/v1/chat/completions` | Modelo: `Qwen/Qwen2.5-Coder-32B-Instruct-AWQ`")
-    gr.ChatInterface(
-        fn=gradio_generate,
-        additional_inputs=[
-            gr.Textbox("Eres un asistente de programación experto.", label="System Prompt"),
-            gr.Slider(0.0, 1.0, 0.2, step=0.05, label="Temperature"),
-            gr.Slider(0.1, 1.0, 0.9, step=0.05, label="Top-P"),
-            gr.Slider(256, 4096, 2048, step=256, label="Max New Tokens")
-        ]
-    )
-
-app = gr.mount_gradio_app(app, gradio_app, path="/")
-
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    demo.queue().launch(server_name="0.0.0.0", server_port=7860)