Update app.py

app.py

@@ -1,6 +1,6 @@
 import os, re, requests, pandas as pd, gradio as gr
 from transformers import pipeline
-from langchain_huggingface.llms import HuggingFacePipeline
+from langchain_huggingface import HuggingFacePipeline, ChatHuggingFace
 from langchain.tools import tool
 from langchain_core.output_parsers import JsonOutputParser
 from langchain.agents import AgentExecutor
@@ -133,8 +133,16 @@ class BasicAgent:
         # initialize HF inference pipeline once
         if HF_TOKEN is None:
             raise ValueError("HF_TOKEN not set in environment")
-        pipe = pipeline("text-generation", model="EleutherAI/gpt-neo-125M", max_new_tokens=16)
-        self.llm = HuggingFacePipeline(pipeline=pipe).bind_tools(TOOLS)
+        #pipe = pipeline("text-generation", model="EleutherAI/gpt-neo-125M", max_new_tokens=16)
+        #self.llm = HuggingFacePipeline(pipeline=pipe).bind_tools(TOOLS)
+        hf_pipe = HuggingFacePipeline.from_model_id(
+            model_id="EleutherAI/gpt-neo-125M",
+            task="text-generation",
+            pipeline_kwargs={"max_new_tokens":16},
+        )
+        chat = ChatHuggingFace(llm=hf_pipe)           # wrap in chat‐model
+        self.llm = chat.bind_tools(TOOLS)                 # now this works :contentReference[oaicite:0]{index=0}
+        
         # The GAIA system prompt (no "FINAL ANSWER:" at the end)
         self.system_prompt = SYSTEM_MESSAGE
         print("BasicAgent initialized with LLM.")
@@ -144,7 +152,13 @@ class BasicAgent:
         prompt = f"{self.system_prompt}Q: {q}\nA:"
         #out = self.generator(prompt, max_new_tokens=16, return_full_text=False)
         #return out[0]["generated_text"].strip()
-        agent = AgentExecutor(agent=self.llm, tools=TOOLS, prompt=prompt, verbose=False, return_intermediate_steps=False)
+        agent = AgentExecutor.from_llm_and_tools(
+            llm=self.llm,
+            tools=TOOLS,
+            system_message=self.system_prompt,
+            verbose=True,
+        )
+        #agent = AgentExecutor(agent=self.llm, tools=TOOLS, prompt=prompt, verbose=False, return_intermediate_steps=False)
         result = agent.invoke({"input": question})
         return JsonOutputParser().parse(result)