Pin torch version

app.py

@@ -3,7 +3,7 @@ from main import GetChatModel
 from graph import BuildGraph
 from retriever import db_dir
 from langgraph.checkpoint.memory import MemorySaver
-
+from dotenv import load_dotenv
 from main import openai_model, model_id
 from util import get_sources, get_start_end_months
 import zipfile
@@ -15,6 +15,9 @@ import uuid
 import ast
 import os
 
+# Setup environment variables
+load_dotenv(dotenv_path=".env", override=True)
+
 # Global settings for compute_mode and search_type
 COMPUTE = "local"
 search_type = "hybrid"
@@ -224,20 +227,6 @@ with gr.Blocks(
     css=css,
 ) as demo:
 
-    # -----------------
-    # Data availability
-    # -----------------
-
-    def is_data_present():
-        """Check if the database directory is present"""
-
-        return os.path.isdir(db_dir)
-
-    def is_data_missing():
-        """Check if the database directory is missing"""
-
-        return not os.path.isdir(db_dir)
-
     # -----------------
     # Define components
     # -----------------
@@ -256,7 +245,7 @@ with gr.Blocks(
     downloading = gr.Textbox(
         lines=1,
         label="Downloading Data, Please Wait",
-        visible=is_data_missing(),
+        visible=False,
         render=False,
     )
     extracting = gr.Textbox(
@@ -392,7 +381,7 @@ with gr.Blocks(
                     intro = gr.Markdown(get_intro_text())
                 with gr.Column(scale=1):
                     compute_mode.render()
-            with gr.Group(visible=is_data_present()) as chat_interface:
+            with gr.Group(visible=False) as chat_interface:
                 chatbot.render()
                 input.render()
             # Render textboxes for data loading progress
@@ -627,23 +616,33 @@ with gr.Blocks(
 
         return None
 
+    def is_data_present():
+        """Check if the database directory is present"""
+
+        return os.path.isdir(db_dir)
+
+    def is_data_missing():
+        """Check if the database directory is missing"""
+
+        return not os.path.isdir(db_dir)
+
     false = gr.State(False)
     need_data = gr.State()
     have_data = gr.State()
 
-    # When app is launched, check if data is present, download it if necessary,
-    # show extracting step as textbox, show error textbox if needed,
-    # update database info, and restore chat interface.
-    # nb. initial visibility of chat interface components and
-    # downloading textbox are set after instantiation of ChatInterface above
+    # When app is launched: check if data is present, download and extract it
+    # if necessary, make chat interface visible, update database info, and show
+    # error textbox if data loading failed.
 
     # fmt: off
     demo.load(
+        is_data_missing, None, [need_data], api_name=False
+    ).then(
+        change_visibility, [need_data], [downloading], api_name=False
+    ).then(
         download, None, [downloading], api_name=False
     ).then(
         change_visibility, [false], [downloading], api_name=False
-    ).then(
-        is_data_missing, None, [need_data], api_name=False
     ).then(
         change_visibility, [need_data], [extracting], api_name=False
     ).then(

graph.py

@@ -4,7 +4,6 @@ from langchain_core.tools import tool
 from langgraph.prebuilt import ToolNode, tools_condition
 from langchain_huggingface import ChatHuggingFace
 from typing import Optional
-from dotenv import load_dotenv
 import datetime
 import os
 
@@ -16,8 +15,7 @@ from mods.tool_calling_llm import ToolCallingLLM
 # Local modules
 from retriever import BuildRetriever
 
-## For LANGCHAIN_API_KEY
-# load_dotenv(dotenv_path=".env", override=True)
+# For tracing (disabled)
 # os.environ["LANGSMITH_TRACING"] = "true"
 # os.environ["LANGSMITH_PROJECT"] = "R-help-chat"
 

main.py

@@ -6,6 +6,7 @@ from langgraph.checkpoint.memory import MemorySaver
 from langchain_core.messages import ToolMessage
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 from datetime import datetime
+from dotenv import load_dotenv
 import os
 import glob
 import torch
@@ -29,6 +30,9 @@ from prompts import answer_prompt
 # -----------
 # First version by Jeffrey Dick on 2025-06-29
 
+# Setup environment variables
+load_dotenv(dotenv_path=".env", override=True)
+
 # Define the remote (OpenAI) model
 openai_model = "gpt-4o-mini"
 

requirements.txt

@@ -7,18 +7,20 @@ langgraph>=0.4.7,<0.6
 sentence-transformers>=4.1.0
 # Required by Nomic embeddings
 einops==0.8.1
-# SmolLM3: >=4.53
-# Gemma 3: >=4.50
-# Gemma 3 with 4.54.0 gives:
-# ValueError: Max cache length is not consistent across layers
+torch==2.7.1
+# Stated requirements:
+#   SmolLM3: transformers>=4.53
+#   Gemma 3: transformers>=4.50
+# Gemma 3 with transformers==4.54.0 gives:
+#   ValueError: Max cache length is not consistent across layers
 transformers==4.53.3
 # Commented because we have local modifications
 #tool-calling-llm==0.1.2
 bm25s==0.2.12
 ragas==0.2.15
-# Temporary fix for ChromaDB telemetry error log messages
+# posthog<6.0.0 is temporary fix for ChromaDB telemetry error log messages
 # https://github.com/vanna-ai/vanna/issues/917
-posthog<6.0.0
+posthog==5.4.0
 # Gradio for the web interface
 gradio==5.38.2
 spaces==0.37.1