Revert all langchain migration changes - back to morning state (330a1dc)

Author: anas-elghafari

backend/Dockerfile

@@ -14,9 +14,6 @@ RUN uv sync --frozen
 
 COPY . /app/
 
-# Run import test on container start and log results
-RUN uv run python test_langchain_imports.py || true
-
 ENV PYTHONUNBUFFERED='1'
 ENV PYTHONPATH="/app"
 

backend/pyproject.toml

@@ -27,7 +27,6 @@ classifiers = [
 
 dependencies = [
     "numpy >=2.0.0,<3.0",  # Pin to 2.x - requires regenerating graph.pkl.gz
-    "pydantic >=2.0.0",  # Keep v2 for decentriq-platform compatibility
     "decentriq-platform ==0.39.1",
     "thefuzz",
     "curies",
@@ -43,6 +42,7 @@ dependencies = [
     "oxrdflib",
     "SPARQLWrapper",
     "python-dotenv",
+    # "pydantic >=2.0.0",
     # "pydantic-settings",
     # NOTE: dependencies for mapping-generation are below
     "tqdm",
@@ -52,6 +52,7 @@ dependencies = [
     "qdrant-client",
     "langchain",
     "langchain_openai",
+    "pydantic",
     "langchain_qdrant",
     "typing-extensions",
     "openai",

backend/src/mapping_generation/compress.py

@@ -1,11 +1,11 @@
 from typing import Any, List
-from langchain_community.retrievers import ContextualCompressionRetriever
+from langchain.retrievers import ContextualCompressionRetriever
 from langchain_core.callbacks import (
     AsyncCallbackManagerForRetrieverRun,
     CallbackManagerForRetrieverRun,
 )
 from langchain_core.documents import Document
-from langchain_community.retrievers import MergerRetriever
+from langchain.retrievers import MergerRetriever
 from langchain_core.retrievers import BaseRetriever
 import asyncio
 

backend/src/mapping_generation/embeddingfilter.py

@@ -2,8 +2,8 @@
 from typing import List, Optional
 
 import numpy as np
-from langchain_core.embeddings import Embeddings
-from langchain_community.retrievers.document_compressors import EmbeddingsFilter
+from langchain.embeddings.base import Embeddings
+from langchain.retrievers.document_compressors import EmbeddingsFilter
 from langchain_community.document_transformers.embeddings_redundant_filter import _DocumentWithState
 from langchain_core.callbacks.manager import Callbacks
 from langchain_core.documents import Document

backend/src/mapping_generation/llm_chain.py

@@ -7,9 +7,9 @@
 )
 from .utils import load_mapping
 from .utils import global_logger as logger
-from pydantic import ValidationError
+from pydantic.v1 import ValidationError
 from langchain_core.output_parsers import JsonOutputParser
-# Note: OutputFixingParser not available in langchain 0.3.15, using fallback
+from langchain.output_parsers import OutputFixingParser
 from typing import List, Dict, Any
 import time
 from langchain_core.callbacks.base import BaseCallbackHandler
@@ -22,8 +22,9 @@
 
 parsing_llm = LLMManager.get_instance("llama3.1")
 parser = JsonOutputParser()
-# Fallback: use parser directly since OutputFixingParser is not available
-fixing_parser = parser
+fixing_parser = OutputFixingParser.from_llm(
+    parser=parser, llm=parsing_llm, max_retries=3
+)
 REQUEST_LIMIT = 30
 TIME_WINDOW = 60
 # from langchain_core.tracers.stdout import ConsoleCallbackHandler

backend/src/mapping_generation/utils.py

@@ -9,7 +9,7 @@
 # import psutil
 import pandas as pd
 from json_repair import repair_json
-from langchain_core.documents import Document
+from langchain.schema import Document
 from simstring.database.dict import DictDatabase
 from simstring.feature_extractor.character_ngram import CharacterNgramFeatureExtractor
 from simstring.measure.cosine import CosineMeasure