Update app.py

app.py

@@ -187,21 +187,67 @@ def chat():
                 #prompt = prompt_template.format(context=context_text_document,table=context_text_table, question=query_text)
                 print("results------------------->",prompt)
 
-            #Model Defining and its use 
+            # #Model Defining and its use 
+            # repo_id = "mistralai/Mistral-7B-Instruct-v0.3"
+            # HFT = os.environ["HF_TOKEN"]
+            # llm = HuggingFaceEndpoint(
+            #     repo_id=repo_id,
+            #     #max_tokens=3000,
+            #     max_new_tokens=2000,
+            #     task = "text-generation",
+            #     temperature=0.8,
+            #     huggingfacehub_api_token=HFT,
+            # )
+            # data= llm.invoke(prompt)
+            # #data= llm(prompt)
+            # #data = response.choices[0].message.content
+            # ---------------------- LLM CALL (FIXED) ----------------------
             repo_id = "mistralai/Mistral-7B-Instruct-v0.3"
-            HFT = os.environ["HF_TOKEN"]
+            HFT = os.environ.get("HF_TOKEN")
+            
+            if not HFT:
+                raise RuntimeError("HF_TOKEN not found in environment")
+            
             llm = HuggingFaceEndpoint(
                 repo_id=repo_id,
-                #max_tokens=3000,
+                task="text-generation",
                 max_new_tokens=2000,
-                task = "text-generation",
                 temperature=0.8,
                 huggingfacehub_api_token=HFT,
             )
-            data= llm.invoke(prompt)
-            #data= llm(prompt)
-            #data = response.choices[0].message.content
-
+            
+            try:
+                # ALWAYS USE generate() — invoke() returns raw HF dict
+                raw_resp = llm.generate([prompt])
+            
+                print("\n=== RAW LLM RESPONSE ===")
+                print(repr(raw_resp))
+                print("========================\n")
+            
+                # --- Robust extraction ---
+                if hasattr(raw_resp, "generations"):
+                    # LangChain LLMResult: list[list[Generation]]
+                    data = raw_resp.generations[0][0].text
+                elif isinstance(raw_resp, dict) and "generated_text" in raw_resp:
+                    data = raw_resp["generated_text"]
+                else:
+                    data = str(raw_resp)
+            
+                # --- Clean unwanted instruction tokens ---
+                data = re.sub(r'<\/?s>', '', data)
+                data = re.sub(r'\[\/?INST\]', '', data, flags=re.IGNORECASE)
+                data = re.sub(r'<\|im_start\|assistant>|<\|im_end\|>', '', data)
+                data = data.strip()
+            
+                print("\n=== CLEANED LLM TEXT ===")
+                print(repr(data))
+                print("========================\n")
+            
+            except Exception as e:
+                print("LLM ERROR:", repr(e))
+                flash(f"LLM Error: {e}", "error")
+                return redirect(url_for('list_dbs'))
+            # --------------------------------------------------------------
             # filtering the uneccessary context.
             if re.search(r'\bmention\b|\bnot mention\b|\bnot mentioned\b|\bnot contain\b|\bnot include\b|\bnot provide\b|\bdoes not\b|\bnot explicitly\b|\bnot explicitly mentioned\b', data, re.IGNORECASE):
                 data = "We do not have information related to your query on our end."