Update app.py

app.py

@@ -2,9 +2,14 @@ from fastapi import FastAPI
 from pydantic import BaseModel
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
+import multiprocessing
 
 app = FastAPI()
 
+# ===============================
+# MODEL CONFIG
+# ===============================
+
 MODEL_REPO = "bartowski/Qwen2.5-Coder-1.5B-Instruct-GGUF"
 MODEL_FILE = "Qwen2.5-Coder-1.5B-Instruct-Q4_K_M.gguf"
 
@@ -13,33 +18,89 @@ model_path = hf_hub_download(
     filename=MODEL_FILE
 )
 
+# ===============================
+# LLM INITIALIZATION (OPTIMIZED)
+# ===============================
+
 llm = Llama(
     model_path=model_path,
-    n_ctx=2048,
-    n_threads=2
+
+    # Larger context for coding tasks
+    n_ctx=8192,
+
+    # Use all CPU cores
+    n_threads=multiprocessing.cpu_count(),
+
+    # CPU inference
+    n_gpu_layers=0,
+
+    # PERFORMANCE BOOST
+    n_batch=512,
+    use_mmap=True,
+    use_mlock=True,
 )
 
+# ===============================
+# REQUEST MODEL
+# ===============================
+
 class ChatRequest(BaseModel):
     message: str
 
+# ===============================
+# HEALTH CHECK
+# ===============================
+
 @app.get("/")
 def root():
-    return {"status": "AI engine running"}
+    return {"status": "Coding AI engine running"}
+
+# ===============================
+# CHAT ENDPOINT
+# ===============================
 
 @app.post("/chat")
 def chat(req: ChatRequest):
+
+    # CODING SPECIALIZED SYSTEM PROMPT
+    system_prompt = (
+        "<|system|>"
+        "You are an elite senior software engineer AI. "
+        "Write clean, production-ready code. "
+        "Always include comments. "
+        "Use best practices, error handling, and optimization. "
+        "Format output in proper markdown with code blocks."
+        "<|end|>"
+    )
+
+    prompt = system_prompt + f"<|user|>{req.message}<|assistant|>"
+
     output = llm(
-        f"<|user|>{req.message}<|assistant|>",
-        max_tokens=512,
-        temperature=0.7,
+        prompt,
+
+        # Larger token output for code
+        max_tokens=800,
+
+        # Lower randomness = better code
+        temperature=0.4,
+
+        # Stable generation
         top_p=0.9,
-        repeat_penalty=1.1,
+
+        # Prevent repetition loops
+        repeat_penalty=1.2,
+
         stop=["<|end|>"]
     )
 
-    return {"reply": output["choices"][0]["text"]}
+    response_text = output["choices"][0]["text"].strip()
+
+    return {"reply": response_text}
+
+# ===============================
+# LOCAL RUN
+# ===============================
 
-# ⭐ THIS PART WAS MISSING
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run(app, host="0.0.0.0", port=7860)