Create withoutllama

withoutllama

@@ -0,0 +1,113 @@
+import os
+from langchain_community.tools import WikipediaQueryRun, ArxivQueryRun
+from langchain_community.utilities import WikipediaAPIWrapper, ArxivAPIWrapper
+from langchain_huggingface import HuggingFacePipeline
+from langchain.agents import initialize_agent, AgentType
+from transformers import AutoTokenizer, AutoModelForCausalLM, TextGenerationPipeline
+from huggingface_hub import login
+import torch
+import traceback
+
+# ✅ Login to HF
+token = os.getenv("HF_TOKEN")
+print("🔑 HF_TOKEN available?", token is not None)
+if token:
+    login(token=token)
+else:
+    print("❌ No HF_TOKEN found in environment")
+
+
+def build_qa():
+    print("🚀 Starting QA pipeline...")
+
+    # ---- 1. Tools ----
+    try:
+        print("🔹 Initializing Wikipedia tool...")
+        wiki_wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=200)
+        wiki = WikipediaQueryRun(api_wrapper=wiki_wrapper)
+
+        print("🔹 Initializing Arxiv tool...")
+        arxiv_wrapper = ArxivAPIWrapper(top_k_results=1, doc_content_chars_max=200)
+        arxiv = ArxivQueryRun(api_wrapper=arxiv_wrapper)
+
+        tools = [wiki, arxiv]
+        print("✅ Tools initialized")
+    except Exception as e:
+        print("❌ Tools initialization failed:", e)
+        traceback.print_exc()
+        return None
+
+    # ---- 2. Model ----
+    try:
+        print("🔹 Loading Mistral 7B model...")
+        model_name = "mistralai/Mistral-7B-Instruct-v0.3"  # or your CPU-quantized version
+        tokenizer = AutoTokenizer.from_pretrained(model_name)
+        model = AutoModelForCausalLM.from_pretrained(
+            model_name,
+            device_map="auto",   # works after installing accelerate
+            dtype=torch.float16, # instead of torch_dtype
+        )
+        llm = TextGenerationPipeline(
+            model=model,
+            tokenizer=tokenizer,
+            max_new_tokens=256,
+            temperature=0.2,
+            do_sample=False,
+            top_p=0.9,
+            repetition_penalty=1.2,
+            eos_token_id=tokenizer.eos_token_id,
+            return_full_text=False,
+        )
+        hf_llm = HuggingFacePipeline(pipeline=llm)
+        print(f"✅ Model loaded: {model_name}")
+    except Exception as e:
+        print("❌ Model load failed:", e)
+        traceback.print_exc()
+        return None
+
+    # ---- 3. Agent ----
+    try:
+        print("🔹 Initializing agent...")
+        agent = initialize_agent(
+            tools=tools,
+            llm=hf_llm,
+            agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
+            verbose=True,
+            handle_parsing_errors=True,
+        )
+        print("✅ Agent initialized")
+    except Exception as e:
+        print("❌ Agent initialization failed:", e)
+        traceback.print_exc()
+        return None
+
+    print("✅ QA pipeline ready")
+    return agent
+
+
+# ---- Build once ----
+try:
+    agent = build_qa()
+    if agent:
+        print("✅ QA pipeline built successfully:", type(agent))
+    else:
+        print("❌ QA pipeline build returned None")
+except Exception as e:
+    agent = None
+    print("❌ Failed to build QA pipeline:", e)
+    traceback.print_exc()
+
+
+def get_response(user_message, history):
+    if agent is None:
+        return "⚠️ QA pipeline not initialized."
+
+    try:
+        print("💬 User query:", user_message)
+        response = agent.invoke({"input": user_message})
+        print("🤖 Agent response:", response)
+        return response
+    except Exception as e:
+        print("❌ Agent execution failed:", e)
+        traceback.print_exc()
+        return f"❌ QA run failed: {e}"