Update main.py

main.py

@@ -7,23 +7,23 @@ import torch
 import os
 import asyncio
 
-# ✅ Set a safe and writable HF cache directory
-os.environ["HF_HOME"] = "./hf_home"
+# ✅ Use writable temp dir for Hugging Face cache
+os.environ["HF_HOME"] = "/tmp/hf_home"
 os.makedirs(os.environ["HF_HOME"], exist_ok=True)
 
-# ✅ Model and tokenizer (only loaded once)
+# ✅ Load model and tokenizer
 model_name = "Qwen/Qwen2.5-0.5B-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
 
-# ✅ Set device (use GPU if available)
+# ✅ Use CUDA if available
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model.to(device)
 
-# ✅ FastAPI app
+# ✅ Initialize FastAPI
 app = FastAPI()
 
-# ✅ CORS settings
+# ✅ Enable CORS
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -32,25 +32,21 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# ✅ Request schema
+# ✅ Input data model
 class Question(BaseModel):
     question: str
 
-# ✅ System prompt
+# ✅ Instructional system prompt
 SYSTEM_PROMPT = "You are Orion, an intelligent AI assistant created by Abdullah Ali, a 13-year-old from Lahore. Respond kindly and wisely."
 
-# ✅ Streaming generator
+# ✅ Streaming response generator
 async def generate_response_chunks(prompt: str):
     qwen_prompt = (
         f"<|im_start|>system\n{SYSTEM_PROMPT}<|im_end|>\n"
         f"<|im_start|>user\n{prompt}<|im_end|>\n"
         f"<|im_start|>assistant\n"
     )
-
-    # Tokenize prompt
     inputs = tokenizer(qwen_prompt, return_tensors="pt").to(device)
-
-    # Generate output
     outputs = model.generate(
         **inputs,
         max_new_tokens=256,
@@ -59,17 +55,13 @@ async def generate_response_chunks(prompt: str):
         top_p=0.9,
         pad_token_id=tokenizer.eos_token_id
     )
-
-    # Decode output
     full_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
     reply = full_output.split("<|im_start|>assistant\n")[-1].strip()
-
-    # Yield chunks word by word (simulating stream)
     for word in reply.split():
         yield word + " "
-        await asyncio.sleep(0.01)  # slight delay for streaming effect
+        await asyncio.sleep(0.01)
 
-# ✅ POST endpoint
+# ✅ API route
 @app.post("/ask")
 async def ask(question: Question):
     return StreamingResponse(generate_response_chunks(question.question), media_type="text/plain")