Update app.py

app.py

@@ -1,4 +1,3 @@
-# [سيرفر النموذج] app.py
 import os, json, asyncio
 from fastapi import FastAPI, Request, HTTPException
 from fastapi.responses import StreamingResponse
@@ -22,7 +21,7 @@ def load_model():
         model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
         llm = Llama(
             model_path=model_path,
-            n_ctx=2048,
+            n_ctx=1536,
             n_threads=4,
             n_gpu_layers=0,
             use_mmap=True,
@@ -36,14 +35,14 @@ def startup():
 
 async def generate_stream(messages: list, mode: str):
     system_prompt = build_system_prompt(mode)
-    prompt = f"<|system|>\n{system_prompt}\n<|user|>\n{messages[-1]['content']}\n<|assistant|>\n"
+    user_msg = messages[-1]['content']
+    prompt = f"<|system|>\n{system_prompt}\n<|user|>\n{user_msg}\n<|assistant|>\n"
     
     if mode == "search":
-        query = messages[-1]['content']
-        search_res = search_web(query)
-        prompt = f"<|system|>\n{system_prompt}\n<|user|>\n{query}\n[SEARCH RESULTS]\n{search_res}\n<|assistant|>\n"
+        search_res = search_web(user_msg)
+        prompt = f"<|system|>\n{system_prompt}\n<|user|>\n{user_msg}\n[SEARCH RESULTS]\n{search_res}\n<|assistant|>\n"
 
-    for token in llm(prompt, max_tokens=1500, stop=["<|user|>", "<|end|>"], stream=True, temperature=0.7):
+    for token in llm(prompt, max_tokens=1200, stop=["<|user|>", "<|end|>"], stream=True, temperature=0.7):
         yield json.dumps({"token": token["choices"][0]["text"]}) + "\n"
         await asyncio.sleep(0.01)