Update chatbot_rag.py

chatbot_rag.py

@@ -38,34 +38,24 @@ def build_qa():
     print("📂 Docs in DB:", vectorstore._collection.count())
 
     # 3. Load LLM (Phi-3 mini)
-    print("🔹 Loading LLM...")
-    #model_id = "meta-llama/Llama-3.2-1B-Instruct"   # or "meta-llama/Llama-3.1-1B-Instruct"
-    model_id = "PleIAs/Pleias-RAG-1B"
-    
-    # Load tokenizer & model
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
-    model = AutoModelForCausalLM.from_pretrained(
-        model_id,
-        device_map="auto",
-        trust_remote_code=True   # important for some HF models
-    )
-    
-    # Build pipeline
-    pipe = pipeline(
-        "text-generation",
-        model=model,
-        tokenizer=tokenizer,
-        max_new_tokens=256,
-        temperature=0.2,          # more deterministic
-        do_sample=False,          # no random sampling
-        top_p=0.9,                # nucleus sampling (safety if do_sample=True)
-        repetition_penalty=1.2,   # avoid loops
-        eos_token_id=tokenizer.eos_token_id,
-        return_full_text=False
-    )
+    print("🔹 Loading LLM...") 
     
+    model_id = "meta-llama/Llama-3.2-3B-Instruct" 
+    tokenizer = AutoTokenizer.from_pretrained(model_id) 
+    model = AutoModelForCausalLM.from_pretrained( model_id, device_map="auto", trust_remote_code=True # ensures it runs on available CPU ) 
+    pipe = pipeline( 
+        "text-generation", 
+        model=model, 
+        tokenizer=tokenizer, 
+        max_new_tokens=256, 
+        temperature=0.2, # keeps answers deterministic but less rigid than 0 
+        do_sample=True, # allow some randomness 
+        top_p=0.9, # nucleus sampling to avoid loops 
+        repetition_penalty=1.2, # 🚀 penalize repeats 
+        eos_token_id=tokenizer.eos_token_id, # stop at EOS 
+        return_full_text=False 
+    ) 
     llm = HuggingFacePipeline(pipeline=pipe)
-
     # 4. Retriever
     retriever = vectorstore.as_retriever(search_kwargs={"k": 3})