Update app.py

app.py

@@ -1,48 +1,61 @@
+import os
+import torch
 import spaces
 import gradio as gr
-import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from threading import Thread
 
-MODEL_ID = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
+MODEL_ID = "Qwen/Qwen2.5-Coder-32B-Instruct-AWQ"
 
+# Cargar tokenizer
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+
+# Cargar modelo cuantizado en AWQ (ocupa ~18-20GB VRAM)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
-    torch_dtype=torch.bfloat16,
+    torch_dtype=torch.float16,
     device_map="auto",
     trust_remote_code=True
 )
 
 @spaces.GPU(duration=120)
-def generate(message, history, system_prompt=""):
+def 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})
-    
-    inputs = tokenizer.apply_chat_template(
-        messages, 
-        tokenize=True, 
-        add_generation_prompt=True, 
-        return_tensors="pt"
-    ).to(model.device)
-
-    streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
+
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+
+    streamer = TextIteratorStreamer(
+        tokenizer,
+        timeout=60.0,
+        skip_prompt=True,
+        skip_special_tokens=True
+    )
+
     generate_kwargs = dict(
-        inputs=inputs,
+        model_inputs,
         streamer=streamer,
-        max_new_tokens=2048,
-        temperature=0.3,
-        top_p=0.95
+        max_new_tokens=int(max_tokens),
+        do_sample=temperature > 0.0,
+        temperature=max(temperature, 1e-2) if temperature > 0.0 else None,
+        top_p=top_p if temperature > 0.0 else None,
+        repetition_penalty=1.05
     )
 
+    # Iniciar generación en un hilo secundario para streaming fluido
     thread = Thread(target=model.generate, kwargs=generate_kwargs)
     thread.start()
 
@@ -51,13 +64,42 @@ def generate(message, history, system_prompt=""):
         partial_text += new_token
         yield partial_text
 
-demo = gr.ChatInterface(
-    fn=generate,
-    title="DeepSeek Coder V2 Lite (ZeroGPU)",
-    additional_inputs=[
-        gr.Textbox("Eres un asistente experto en programación y resolución de problemas.", label="System Prompt")
-    ]
-)
+# Interfaz de Gradio
+with gr.Blocks(title="Qwen2.5-Coder-32B-Instruct AWQ (ZeroGPU)", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🚀 Qwen2.5-Coder-32B-Instruct (AWQ en ZeroGPU)")
+    gr.Markdown("Servicio de inferencia optimizado para código y compatible con clientes vía API / Gradio.")
+
+    chat_interface = gr.ChatInterface(
+        fn=generate,
+        additional_inputs=[
+            gr.Textbox(
+                value="Eres un asistente de programación experto, preciso y conciso.",
+                label="System Prompt",
+                lines=2
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=1.0,
+                value=0.2,
+                step=0.05,
+                label="Temperature"
+            ),
+            gr.Slider(
+                minimum=0.1,
+                maximum=1.0,
+                value=0.9,
+                step=0.05,
+                label="Top-P"
+            ),
+            gr.Slider(
+                minimum=256,
+                maximum=4096,
+                value=2048,
+                step=256,
+                label="Max New Tokens"
+            )
+        ]
+    )
 
 if __name__ == "__main__":
     demo.queue().launch()