Add llama-index and llama-cpp-python with proper imports for pydantic v2 compatibility

requirements.txt

@@ -94,3 +94,6 @@ validators==0.20.0
 yarl==1.9.2
 zipp==3.15.0
 torch>=2.0.0
+llama-index-core==0.10.0
+llama-index-llms-llama-cpp==0.1.0
+llama-cpp-python>=0.2.0

src/backend/chatbot.py

@@ -9,7 +9,7 @@ def _lazy_import_llm_libs():
     """Import heavy LLM libraries only when needed"""
     global ChatOpenAI, ChatPromptTemplate, SystemMessagePromptTemplate
     global AIMessagePromptTemplate, HumanMessagePromptTemplate
-    global SimpleDirectoryReader, VectorStoreIndex, ServiceContext
+    global SimpleDirectoryReader, VectorStoreIndex
     global LlamaCPP, messages_to_prompt, completion_to_prompt
     
     from langchain_community.chat_models import ChatOpenAI
@@ -19,16 +19,21 @@ def _lazy_import_llm_libs():
         AIMessagePromptTemplate,
         HumanMessagePromptTemplate,
     )
-    from llama_index import (
+    from llama_index.core import (
         SimpleDirectoryReader,
         VectorStoreIndex,
-        ServiceContext,
-    )
-    from llama_index.llms import LlamaCPP
-    from llama_index.llms.llama_utils import (
-        messages_to_prompt,
-        completion_to_prompt,
     )
+    from llama_index.llms.llama_cpp import LlamaCPP
+    
+    # Try to import prompt utilities (may not exist in newer versions)
+    try:
+        from llama_index.llms.llama_cpp.llama_utils import (
+            messages_to_prompt,
+            completion_to_prompt,
+        )
+    except ImportError:
+        messages_to_prompt = None
+        completion_to_prompt = None
 
 # set version
 # st.session_state.demo_lite = False
@@ -157,25 +162,23 @@ def init_llm(model, demo_lite):
                 print(f"❌ Model file not found: {model_path}")
                 return None
             
-            llm = LlamaCPP(
-                # You can pass in the URL to a GGML model to download it automatically
-                # model_url=model_url,
-                # optionally, you can set the path to a pre-downloaded model instead of model_url
-                model_path=model_path,
-                temperature=0.1,
-                max_new_tokens=1000,
-                # llama2 has a context window of 4096 tokens, but we set it lower to allow for some wiggle room
-                context_window=3000,
-                # kwargs to pass to __call__()
-                generate_kwargs={},
-                # kwargs to pass to __init__()
-                # set to at least 1 to use GPU, -1 to use all layers on GPU
-                model_kwargs={"n_gpu_layers": n_gpu_layers},
-                # transform inputs into Llama2 format
-                messages_to_prompt=messages_to_prompt,
-                completion_to_prompt=completion_to_prompt,
-                verbose=True,
-            )
+            # Build kwargs for LlamaCPP
+            llm_kwargs = {
+                "model_path": model_path,
+                "temperature": 0.1,
+                "max_new_tokens": 1000,
+                "context_window": 3000,
+                "generate_kwargs": {},
+                "model_kwargs": {"n_gpu_layers": n_gpu_layers},
+                "verbose": True,
+            }
+            # Add prompt formatters if available (optional in newer versions)
+            if messages_to_prompt is not None:
+                llm_kwargs["messages_to_prompt"] = messages_to_prompt
+            if completion_to_prompt is not None:
+                llm_kwargs["completion_to_prompt"] = completion_to_prompt
+            
+            llm = LlamaCPP(**llm_kwargs)
         elif model == "deci-7b_CPP":
             model_path = os.path.join(model_base_path, "decilm-7b-uniform-gqa-q8_0.gguf")
             print("model path: ", model_path)