Skip to content

CHANGES.md

Latest commit

 

History

History
148 lines (108 loc) · 5.05 KB

CHANGES.md

File metadata and controls

148 lines (108 loc) · 5.05 KB

Changelog — Compatibility Fixes for LangChain 1.x, llama_index 0.10+, ChromaDB 1.x, OpenAI API 1.x

2026-02-19 — LangChain 1.x + ChromaDB Compatibility Fix

Problem

Running the original script produced:

AttributeError: 'NoneType' object has no attribute 'info'

This was caused by two separate issues: broken import paths from the LangChain 0.x → 1.x migration, and Python 3.14 being incompatible with the pydantic v1 compatibility layer used internally by ChromaDB.

1. Import Path Migrations

LangChain 1.x split its monolithic package into several smaller packages. The old import paths no longer exist.

Old (broken) New
from langchain.document_loaders import DirectoryLoader, TextLoader from langchain_community.document_loaders import DirectoryLoader, TextLoader
from langchain.embeddings import OpenAIEmbeddings from langchain_openai import OpenAIEmbeddings, OpenAI
from langchain import OpenAI merged into langchain_openai line above
from langchain.vectorstores import Chroma from langchain_community.vectorstores import Chroma
from langchain.text_splitter import CharacterTextSplitter from langchain_text_splitters import CharacterTextSplitter

2. Removed RetrievalQA Chain

RetrievalQA was removed in LangChain 1.x.

Before:

from langchain.chains import RetrievalQA

qa = RetrievalQA.from_chain_type(llm=OpenAI(), chain_type="stuff", retriever=docsearch.as_retriever())

def query(q):
    print(qa.run(q))

After — replaced with direct retriever + LLM invocation:

retriever = docsearch.as_retriever()
llm = OpenAI()

def query(q):
    print("Query: ", q)
    docs = retriever.invoke(q)
    context = "\n\n".join([doc.page_content for doc in docs])
    answer = llm.invoke(
        f"Use the following context to answer the question.\n\n"
        f"Context:\n{context}\n\nQuestion: {q}\n\nAnswer:"
    )
    print("Answer: ", answer)

3. Python Version Requirement

ChromaDB internally uses pydantic.v1.BaseSettings. Pydantic explicitly dropped support for this compatibility shim on Python 3.14+, causing a hard crash at import time.

Requirement: Python 3.12 (or 3.11/3.13 — anything below 3.14).

Create a fresh conda environment:

conda create -n llm-env python=3.12
conda activate llm-env

4. Required pip Packages

The original code assumed LangChain 0.x's all-in-one langchain package. With the split into smaller packages, the following must be installed explicitly:

pip install langchain-community langchain-openai langchain-text-splitters chromadb python-dotenv
Package Purpose
langchain-community DirectoryLoader, TextLoader, Chroma
langchain-openai OpenAIEmbeddings, OpenAI
langchain-text-splitters CharacterTextSplitter
chromadb Vector store backend
python-dotenv .env file loading for OPENAI_API_KEY

2026-02-20 — Compatibility Fixes for Current Package Versions

Files Fixed

02b_llama_chroma.py — ChromaDB 1.x

  • chromadb.Client() → chromadb.PersistentClient(path="./storage")
  • create_collection() → get_or_create_collection()

03_db.py — LangChain 1.x + Tool-calling

  • Changed OpenAI → ChatOpenAI (required for tool-calling agents)
  • SQLDatabaseChain → create_sql_agent(agent_type="openai-tools")
  • .run() → .invoke({"input": query})

04_csv.py — LangChain 1.x + Security

  • Changed OpenAI → ChatOpenAI
  • Added allow_dangerous_code=True (required in LangChain 1.x for code execution)

07_custom.py — llama_index 0.10+ + CPU compatibility

  • max_input_size → context_window
  • max_chunk_overlap → chunk_overlap_ratio
  • Removed device="cuda:0" for cross-platform compatibility

11_worldbuilding.py — Cohere Generate API → Chat API

  • Rewrote generate() function to use co.chat() (Generate API deprecated Sept 15, 2025)
  • Model updated to command-a-03-2025
  • Simplified prompts for Chat API compatibility
  • Added error handling for inconsistent outputs
  • Note: Token likelihoods not available in Chat API

13_caching_sqlite.py — LangChain 1.x legacy

  • from langchain.chains.summarize → from langchain_classic.chains.summarize

14_streamlit.py — LangChain 1.x legacy

  • from langchain.agents → from langchain_classic.agents

15_sql.py — Groq model update

  • llama3-70b-8192 → llama-3.3-70b-versatile (old model decommissioned)

16_repl.py — Groq model update

  • llama3-70b-8192 → llama-3.3-70b-versatile (old model decommissioned)

requirements.txt — Cross-platform + new packages

  • Commented out triton and uvloop (Linux-only packages)
  • Added openai-agents==0.9.2 (for files 19-24)
  • Added langchain-groq==1.1.2 (for 15_sql.py, 16_repl.py)
  • Added langchain-classic==0.3.21 (for legacy APIs)
  • Updated all package versions to tested environment

Key Package Versions

  • langchain==1.2.10
  • langchain-community==0.4.1
  • langchain-openai==1.1.10
  • chromadb==1.5.0
  • llama-index==0.14.15
  • Python 3.12 required