Update chatbot_rag.py

chatbot_rag.py

@@ -41,27 +41,36 @@ def build_qa():
     # 3. Load LLM (Phi-3 mini)
     print("🔹 Loading LLM...") 
 
-    model_name = "HuggingFaceTB/SmolLM2-1.7B-Instruct"
-
-    # 🔹 Load tokenizer & model
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model_id = "microsoft/phi-2"
+    
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    
+    # Load model
     model = AutoModelForCausalLM.from_pretrained(
-        model_name,
-        device_map="auto",   # auto selects GPU if available, else CPU
-        torch_dtype="auto"   # picks best dtype automatically
+        model_id,
+        device_map="auto",       # put on GPU if available, else CPU
+        torch_dtype="auto",      # auto precision
+        trust_remote_code=True   # allow custom model code
     )
     
-    # 🔹 Create HF pipeline
+    # Create pipeline
     pipe = pipeline(
         "text-generation",
         model=model,
         tokenizer=tokenizer,
-        max_new_tokens=512,     # adjust output length
-        temperature=0.2,
-        do_sample=False, # creativity vs determinism
-        top_p=0.9               # nucleus sampling
+        max_new_tokens=256,       # control length of response
+        temperature=0.2,          # more deterministic
+        do_sample=False,          # no randomness (deterministic answers)
+        top_p=0.9,                # nucleus sampling
+        repetition_penalty=1.2,   # 🚀 reduce loops/repeats
+        eos_token_id=tokenizer.eos_token_id,
+        return_full_text=False
     )
     
+    # Wrap into LangChain LLM
+    llm = HuggingFacePipeline(pipeline=pipe)
+    
     # 🔹 Wrap in LangChain LLM
     llm = HuggingFacePipeline(pipeline=pipe)