From cef02408ddc79ba57dbfa5765e293c2f2f6c1328 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Sun, 5 Oct 2025 23:34:35 +0530
Subject: [PATCH 1/9] feat: add HallucinationDetector for evaluating
 hallucinations (#33191)

---
 .../evaluation/hallucination/__init__.py      |   3 +
 .../evaluation/hallucination/detector.py      |  38 ++++
 libs/langchain/pyproject.toml                 |   4 +
 .../evaluation/hallucination/__init__.py      |   0
 .../evaluation/hallucination/test_detector.py | 207 ++++++++++++++++++
 5 files changed, 252 insertions(+)
 create mode 100644 libs/langchain/langchain/evaluation/hallucination/__init__.py
 create mode 100644 libs/langchain/langchain/evaluation/hallucination/detector.py
 create mode 100644 libs/langchain/tests/integration_tests/evaluation/hallucination/__init__.py
 create mode 100644 libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py

diff --git a/libs/langchain/langchain/evaluation/hallucination/__init__.py b/libs/langchain/langchain/evaluation/hallucination/__init__.py
new file mode 100644
index 0000000000000..dde5106ca1e31
--- /dev/null
+++ b/libs/langchain/langchain/evaluation/hallucination/__init__.py
@@ -0,0 +1,3 @@
+from .detector import HallucinationDetector
+
+__all__ = ["HallucinationDetector"]
diff --git a/libs/langchain/langchain/evaluation/hallucination/detector.py b/libs/langchain/langchain/evaluation/hallucination/detector.py
new file mode 100644
index 0000000000000..6db27161886d4
--- /dev/null
+++ b/libs/langchain/langchain/evaluation/hallucination/detector.py
@@ -0,0 +1,38 @@
+from typing import List, Dict
+from transformers import pipeline
+
+class HallucinationDetector:
+    """
+    Simple Hallucination Detector using NLI models (e.g., facebook/bart-large-mnli).
+    - Extract claims (basic sentence split)
+    - Verify claims against evidence docs using NLI
+    - Compute hallucination rate
+    """
+
+    def __init__(self, model_name: str = "facebook/bart-large-mnli"):
+        self.nli_pipeline = pipeline("text-classification", model=model_name)
+
+    def extract_claims(self, text: str) -> List[str]:
+        """Naive sentence-based claim extraction"""
+        return [c.strip() for c in text.split(".") if c.strip()]
+
+    def verify_claim(self, claim: str, evidence: str) -> bool:
+        """Check if a claim is supported by given evidence"""
+        result = self.nli_pipeline(f"{claim} </s></s> {evidence}")
+        return result[0]["label"].lower() == "entailment"
+
+    def verify_claim_multi(self, claim: str, evidence_docs: List[str]) -> bool:
+        """A claim is supported if any evidence doc entails it"""
+        return any(self.verify_claim(claim, e) for e in evidence_docs)
+
+    def compute_hallucination_rate(self, text: str, evidence_docs: List[str]) -> Dict[str, float]:
+        claims = self.extract_claims(text)
+        if not claims:
+            return {"total_claims": 0, "unsupported_claims": 0, "hallucination_rate": 0.0}
+
+        unsupported = sum(not self.verify_claim_multi(c, evidence_docs) for c in claims)
+        return {
+            "total_claims": len(claims),
+            "unsupported_claims": unsupported,
+            "hallucination_rate": unsupported / len(claims),
+        }
diff --git a/libs/langchain/pyproject.toml b/libs/langchain/pyproject.toml
index 712f7190edac3..3e4db79137df1 100644
--- a/libs/langchain/pyproject.toml
+++ b/libs/langchain/pyproject.toml
@@ -39,6 +39,8 @@ together = ["langchain-together"]
 # deepseek = ["langchain-deepseek"]
 # xai = ["langchain-xai"]
 # perplexity = ["langchain-perplexity"]
+# hallucination = ["transformers", "torch"]
+
 
 [project.urls]
 "Source Code" = "https://github.com/langchain-ai/langchain/tree/master/libs/langchain"
@@ -83,6 +85,8 @@ test_integration = [
     "langchainhub>=0.1.16,<1.0.0",
     "langchain-core",
     "langchain-text-splitters",
+    "transformers>=4.35.0,<5.0.0",
+    "torch>=2.1.0,<3.0.0",
 ]
 lint = [
     "ruff>=0.13.1,<0.14.0",
diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/__init__.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/__init__.py
new file mode 100644
index 0000000000000..e69de29bb2d1d
diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
new file mode 100644
index 0000000000000..9b42e83c7bf43
--- /dev/null
+++ b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
@@ -0,0 +1,207 @@
+# import os
+# import pytest
+# from unittest.mock import patch
+
+# from langchain.evaluation.hallucination.detector import HallucinationDetector
+
+# # -----------------------------
+# # Integration Tests (Real HF model)
+# # -----------------------------
+# skip_if_no_hf = pytest.mark.skipif(
+#     "HF_TOKEN" not in os.environ, reason="Hugging Face token not available"
+# )
+
+# @pytest.fixture(scope="module")
+# @skip_if_no_hf
+# @pytest.mark.requires("integration")
+# def detector_real():
+#     # Only runs locally if HF token is available
+#     return HallucinationDetector(model_name="facebook/bart-large-mnli")
+
+
+# @skip_if_no_hf
+# @pytest.mark.requires("integration")
+# def test_extract_claims_integration(detector_real):
+#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+#     claims = detector_real.extract_claims(text)
+#     assert isinstance(claims, list)
+#     assert len(claims) == 2
+#     assert "Barack Obama was the 44th President of the United States" in claims
+
+
+# @skip_if_no_hf
+# @pytest.mark.requires("integration")
+# def test_compute_hallucination_rate_integration(detector_real):
+#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+#     evidence = [
+#         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+#         "Barack Obama was born in Hawaii, not Kenya."
+#     ]
+#     result = detector_real.compute_hallucination_rate(text, evidence)
+#     unsupported = result["unsupported_claims"]
+#     total = result["total_claims"]
+#     hallucination_rate = result["hallucination_rate"]
+
+#     assert "total_claims" in result
+#     assert "unsupported_claims" in result
+#     assert "hallucination_rate" in result
+#     assert result["total_claims"] == 2
+#     assert unsupported in [1, 2]  # Accepts both possible outputs
+#     assert 0 <= hallucination_rate <= 1  # Just check it’s a valid rate
+
+
+# # -----------------------------
+# # Unit Tests (Mocked)
+# # -----------------------------
+# # Unit test fixture
+# @pytest.fixture(scope="module")
+# def detector_mock():
+#     with patch("langchain.evaluation.hallucination.detector.pipeline") as mock_pipeline:
+#         # mock NLI results
+#         mock_pipeline.return_value = lambda text: [
+#             {"label": "ENTAILMENT", "score": 0.9} if "President" in text else {"label": "CONTRADICTION", "score": 0.9}
+#         ]
+#         # Now constructor won't load HF model
+#         detector = HallucinationDetector(model_name="any")
+#         yield detector
+
+
+# def test_extract_claims_mock(detector_mock):
+#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+#     claims = detector_mock.extract_claims(text)
+#     assert isinstance(claims, list)
+#     assert len(claims) == 2
+
+
+# def test_verify_claim_supported_mock(detector_mock):
+#     claim = "Barack Obama was the 44th President of the United States"
+#     evidence = "Barack Obama served as the 44th President of the United States from 2009 to 2017."
+#     result = detector_mock.verify_claim(claim, evidence)
+#     assert result is True
+
+
+# def test_verify_claim_unsupported_mock(detector_mock):
+#     claim = "Barack Obama was born in Kenya"
+#     evidence = "Barack Obama was born in Hawaii, not Kenya."
+#     result = detector_mock.verify_claim(claim, evidence)
+#     assert result is False
+
+
+# def test_compute_hallucination_rate_mock(detector_mock):
+#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+#     evidence = [
+#         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+#         "Barack Obama was born in Hawaii, not Kenya."
+#     ]
+#     result = detector_mock.compute_hallucination_rate(text, evidence)
+#     assert "total_claims" in result
+#     assert "unsupported_claims" in result
+#     assert "hallucination_rate" in result
+
+
+
+import os
+import pytest
+from unittest.mock import patch
+
+from langchain.evaluation.hallucination.detector import HallucinationDetector
+
+# -----------------------------
+# Integration Tests (Real HF model)
+# -----------------------------
+skip_if_no_hf = pytest.mark.skipif(
+    "HF_TOKEN" not in os.environ, reason="Hugging Face token not available"
+)
+
+@pytest.fixture(scope="module")
+@skip_if_no_hf
+@pytest.mark.requires("integration")
+def detector_real():
+    """Runs only if Hugging Face token is available."""
+    return HallucinationDetector(model_name="facebook/bart-large-mnli")
+
+
+@skip_if_no_hf
+@pytest.mark.requires("integration")
+def test_extract_claims_integration(detector_real):
+    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    claims = detector_real.extract_claims(text)
+    # Check structure and basic logic
+    assert isinstance(claims, list)
+    assert len(claims) == 2
+    # Ensure at least one claim matches expected
+    assert any("Barack Obama was the 44th President" in c for c in claims)
+
+
+@skip_if_no_hf
+@pytest.mark.requires("integration")
+def test_compute_hallucination_rate_integration(detector_real):
+    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    evidence = [
+        "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+        "Barack Obama was born in Hawaii, not Kenya."
+    ]
+    result = detector_real.compute_hallucination_rate(text, evidence)
+
+    # Validate structure
+    for key in ["total_claims", "unsupported_claims", "hallucination_rate"]:
+        assert key in result
+
+    total = result["total_claims"]
+    unsupported = result["unsupported_claims"]
+    hallucination_rate = result["hallucination_rate"]
+
+    # Check logical consistency
+    assert total == 2
+    assert 0 <= unsupported <= total  # Dynamic check
+    assert abs(hallucination_rate - unsupported / total) < 1e-6  # Matches formula
+    assert 0 <= hallucination_rate <= 1
+
+
+# -----------------------------
+# Unit Tests (Mocked)
+# -----------------------------
+@pytest.fixture(scope="module")
+def detector_mock():
+    """Mock pipeline to make unit tests deterministic."""
+    with patch("langchain.evaluation.hallucination.detector.pipeline") as mock_pipeline:
+        # Mock NLI behavior
+        mock_pipeline.return_value = lambda text: [
+            {"label": "ENTAILMENT", "score": 0.9} if "President" in text else {"label": "CONTRADICTION", "score": 0.9}
+        ]
+        detector = HallucinationDetector(model_name="any")  # Model not loaded
+        yield detector
+
+
+def test_extract_claims_mock(detector_mock):
+    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    claims = detector_mock.extract_claims(text)
+    assert isinstance(claims, list)
+    assert len(claims) == 2
+
+
+def test_verify_claim_supported_mock(detector_mock):
+    claim = "Barack Obama was the 44th President of the United States"
+    evidence = "Barack Obama served as the 44th President of the United States from 2009 to 2017."
+    assert detector_mock.verify_claim(claim, evidence) is True
+
+
+def test_verify_claim_unsupported_mock(detector_mock):
+    claim = "Barack Obama was born in Kenya"
+    evidence = "Barack Obama was born in Hawaii, not Kenya."
+    assert detector_mock.verify_claim(claim, evidence) is False
+
+
+def test_compute_hallucination_rate_mock(detector_mock):
+    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    evidence = [
+        "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+        "Barack Obama was born in Hawaii, not Kenya."
+    ]
+    result = detector_mock.compute_hallucination_rate(text, evidence)
+    # Validate structure and logical consistency
+    for key in ["total_claims", "unsupported_claims", "hallucination_rate"]:
+        assert key in result
+    assert result["total_claims"] == 2
+    assert 0 <= result["unsupported_claims"] <= 2
+    assert abs(result["hallucination_rate"] - result["unsupported_claims"] / result["total_claims"]) < 1e-6

From df5cb4f36974d5aed94f3738977e75e10c078509 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 21:48:40 +0530
Subject: [PATCH 2/9] pyproject changed

---
 libs/langchain/pyproject.toml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/libs/langchain/pyproject.toml b/libs/langchain/pyproject.toml
index 3e4db79137df1..2f189a3dd9b28 100644
--- a/libs/langchain/pyproject.toml
+++ b/libs/langchain/pyproject.toml
@@ -75,6 +75,8 @@ test = [
     "packaging>=24.2.0,<26.0.0",
     "numpy>=1.26.4; python_version<'3.13'",
     "numpy>=2.1.0; python_version>='3.13'",
+    "transformers>=4.35.0,<5.0.0",
+    "torch>=2.1.0,<3.0.0",
 ]
 test_integration = [
     "vcrpy>=7.0.0,<8.0.0",

From c1d99ea6fd5c420660d932447f721c1df769e8af Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 21:55:19 +0530
Subject: [PATCH 3/9] commented code has been removed

---
 .../evaluation/hallucination/test_detector.py | 102 ------------------
 1 file changed, 102 deletions(-)

diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
index 9b42e83c7bf43..c23ef146257ac 100644
--- a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
+++ b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
@@ -1,105 +1,3 @@
-# import os
-# import pytest
-# from unittest.mock import patch
-
-# from langchain.evaluation.hallucination.detector import HallucinationDetector
-
-# # -----------------------------
-# # Integration Tests (Real HF model)
-# # -----------------------------
-# skip_if_no_hf = pytest.mark.skipif(
-#     "HF_TOKEN" not in os.environ, reason="Hugging Face token not available"
-# )
-
-# @pytest.fixture(scope="module")
-# @skip_if_no_hf
-# @pytest.mark.requires("integration")
-# def detector_real():
-#     # Only runs locally if HF token is available
-#     return HallucinationDetector(model_name="facebook/bart-large-mnli")
-
-
-# @skip_if_no_hf
-# @pytest.mark.requires("integration")
-# def test_extract_claims_integration(detector_real):
-#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
-#     claims = detector_real.extract_claims(text)
-#     assert isinstance(claims, list)
-#     assert len(claims) == 2
-#     assert "Barack Obama was the 44th President of the United States" in claims
-
-
-# @skip_if_no_hf
-# @pytest.mark.requires("integration")
-# def test_compute_hallucination_rate_integration(detector_real):
-#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
-#     evidence = [
-#         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
-#         "Barack Obama was born in Hawaii, not Kenya."
-#     ]
-#     result = detector_real.compute_hallucination_rate(text, evidence)
-#     unsupported = result["unsupported_claims"]
-#     total = result["total_claims"]
-#     hallucination_rate = result["hallucination_rate"]
-
-#     assert "total_claims" in result
-#     assert "unsupported_claims" in result
-#     assert "hallucination_rate" in result
-#     assert result["total_claims"] == 2
-#     assert unsupported in [1, 2]  # Accepts both possible outputs
-#     assert 0 <= hallucination_rate <= 1  # Just check it’s a valid rate
-
-
-# # -----------------------------
-# # Unit Tests (Mocked)
-# # -----------------------------
-# # Unit test fixture
-# @pytest.fixture(scope="module")
-# def detector_mock():
-#     with patch("langchain.evaluation.hallucination.detector.pipeline") as mock_pipeline:
-#         # mock NLI results
-#         mock_pipeline.return_value = lambda text: [
-#             {"label": "ENTAILMENT", "score": 0.9} if "President" in text else {"label": "CONTRADICTION", "score": 0.9}
-#         ]
-#         # Now constructor won't load HF model
-#         detector = HallucinationDetector(model_name="any")
-#         yield detector
-
-
-# def test_extract_claims_mock(detector_mock):
-#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
-#     claims = detector_mock.extract_claims(text)
-#     assert isinstance(claims, list)
-#     assert len(claims) == 2
-
-
-# def test_verify_claim_supported_mock(detector_mock):
-#     claim = "Barack Obama was the 44th President of the United States"
-#     evidence = "Barack Obama served as the 44th President of the United States from 2009 to 2017."
-#     result = detector_mock.verify_claim(claim, evidence)
-#     assert result is True
-
-
-# def test_verify_claim_unsupported_mock(detector_mock):
-#     claim = "Barack Obama was born in Kenya"
-#     evidence = "Barack Obama was born in Hawaii, not Kenya."
-#     result = detector_mock.verify_claim(claim, evidence)
-#     assert result is False
-
-
-# def test_compute_hallucination_rate_mock(detector_mock):
-#     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
-#     evidence = [
-#         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
-#         "Barack Obama was born in Hawaii, not Kenya."
-#     ]
-#     result = detector_mock.compute_hallucination_rate(text, evidence)
-#     assert "total_claims" in result
-#     assert "unsupported_claims" in result
-#     assert "hallucination_rate" in result
-
-
-
 import os
 import pytest
 from unittest.mock import patch

From 75ce98a3a8589812a26eb2d39cd342012803d544 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 22:12:56 +0530
Subject: [PATCH 4/9] changes

---
 libs/langchain/pyproject.toml | 69 ++++++++++++++++-------------------
 1 file changed, 31 insertions(+), 38 deletions(-)

diff --git a/libs/langchain/pyproject.toml b/libs/langchain/pyproject.toml
index 2f189a3dd9b28..2e97c53c46539 100644
--- a/libs/langchain/pyproject.toml
+++ b/libs/langchain/pyproject.toml
@@ -1,13 +1,13 @@
 [build-system]
-requires = ["pdm-backend"]
-build-backend = "pdm.backend"
+requires = ["hatchling"]
+build-backend = "hatchling.build"
 
 [project]
 authors = []
 license = { text = "MIT" }
 requires-python = ">=3.10.0,<4.0.0"
 dependencies = [
-    "langchain-core>=0.3.72,<1.0.0",
+    "langchain-core>=0.3.72,<2.0.0",
     "langchain-text-splitters>=0.3.9,<1.0.0",
     "langsmith>=0.1.17,<1.0.0",
     "pydantic>=2.7.4,<3.0.0",
@@ -16,31 +16,29 @@ dependencies = [
     "PyYAML>=5.3.0,<7.0.0",
     "async-timeout>=4.0.0,<5.0.0; python_version < \"3.11\"",
 ]
-name = "langchain"
+name = "langchain-classic"
 version = "0.3.27"
 description = "Building applications with LLMs through composability"
 readme = "README.md"
 
 [project.optional-dependencies]
-community = ["langchain-community"]
+# community = ["langchain-community"]
 anthropic = ["langchain-anthropic"]
 openai = ["langchain-openai"]
-# azure-ai = ["langchain-azure-ai"]
-# cohere = ["langchain-cohere"]
+#azure-ai = ["langchain-azure-ai"]
+#cohere = ["langchain-cohere"]
 google-vertexai = ["langchain-google-vertexai"]
 google-genai = ["langchain-google-genai"]
-# fireworks = ["langchain-fireworks"]
-# ollama = ["langchain-ollama"]
+fireworks = ["langchain-fireworks"]
+ollama = ["langchain-ollama"]
 together = ["langchain-together"]
-# mistralai = ["langchain-mistralai"]
-# huggingface = ["langchain-huggingface"]
-# groq = ["langchain-groq"]
-# aws = ["langchain-aws"]
-# deepseek = ["langchain-deepseek"]
-# xai = ["langchain-xai"]
-# perplexity = ["langchain-perplexity"]
-# hallucination = ["transformers", "torch"]
-
+mistralai = ["langchain-mistralai"]
+#huggingface = ["langchain-huggingface"]
+groq = ["langchain-groq"]
+aws = ["langchain-aws"]
+deepseek = ["langchain-deepseek"]
+xai = ["langchain-xai"]
+perplexity = ["langchain-perplexity"]
 
 [project.urls]
 "Source Code" = "https://github.com/langchain-ai/langchain/tree/master/libs/langchain"
@@ -52,31 +50,29 @@ test = [
     "pytest>=8.0.0,<9.0.0",
     "pytest-cov>=4.0.0,<5.0.0",
     "pytest-dotenv>=0.5.2,<1.0.0",
-    "duckdb-engine>=0.9.2,<1.0.0",
     "pytest-watcher>=0.2.6,<1.0.0",
+    "pytest-asyncio>=0.23.2,<1.0.0",
+    "pytest-mock>=3.10.0,<4.0.0",
+    "pytest-socket>=0.6.0,<1.0.0",
+    "pytest-xdist<4.0.0,>=3.6.1",
+    "numpy>=1.26.4; python_version<'3.13'",
+    "numpy>=2.1.0; python_version>='3.13'",
+    "cffi<1.17.1; python_version < \"3.10\"",
+    "cffi; python_version >= \"3.10\"",
+    "duckdb-engine>=0.9.2,<1.0.0",
     "freezegun>=1.2.2,<2.0.0",
     "responses>=0.22.0,<1.0.0",
-    "pytest-asyncio>=0.23.2,<1.0.0",
     "lark>=1.1.5,<2.0.0",
     "pandas>=2.0.0,<3.0.0",
-    "pytest-mock>=3.10.0,<4.0.0",
-    "pytest-socket>=0.6.0,<1.0.0",
     "syrupy>=4.0.2,<5.0.0",
     "requests-mock>=1.11.0,<2.0.0",
-    "pytest-xdist<4.0.0,>=3.6.1",
     "blockbuster>=1.5.18,<1.6.0",
-    "cffi<1.17.1; python_version < \"3.10\"",
-    "cffi; python_version >= \"3.10\"",
+    "toml>=0.10.2,<1.0.0",
+    "packaging>=24.2.0,<26.0.0",
     "langchain-tests",
     "langchain-core",
     "langchain-text-splitters",
     "langchain-openai",
-    "toml>=0.10.2,<1.0.0",
-    "packaging>=24.2.0,<26.0.0",
-    "numpy>=1.26.4; python_version<'3.13'",
-    "numpy>=2.1.0; python_version>='3.13'",
-    "transformers>=4.35.0,<5.0.0",
-    "torch>=2.1.0,<3.0.0",
 ]
 test_integration = [
     "vcrpy>=7.0.0,<8.0.0",
@@ -87,8 +83,6 @@ test_integration = [
     "langchainhub>=0.1.16,<1.0.0",
     "langchain-core",
     "langchain-text-splitters",
-    "transformers>=4.35.0,<5.0.0",
-    "torch>=2.1.0,<3.0.0",
 ]
 lint = [
     "ruff>=0.13.1,<0.14.0",
@@ -97,17 +91,17 @@ lint = [
 ]
 typing = [
     "mypy>=1.15.0,<1.16.0",
+    "mypy-protobuf>=3.0.0,<4.0.0",
     "types-pyyaml>=6.0.12.2,<7.0.0.0",
     "types-requests>=2.28.11.5,<3.0.0.0",
     "types-toml>=0.10.8.1,<1.0.0.0",
     "types-redis>=4.3.21.6,<5.0.0.0",
     "types-pytz>=2023.3.0.0,<2024.0.0.0",
     "types-chardet>=5.0.4.6,<6.0.0.0",
-    "mypy-protobuf>=3.0.0,<4.0.0",
-    "langchain-core",
-    "langchain-text-splitters",
     "numpy>=1.26.4; python_version < '3.13'",
     "numpy>=2.1.0; python_version >= '3.13'",
+    "langchain-core",
+    "langchain-text-splitters",
 ]
 dev = [
     "jupyter>=1.0.0,<2.0.0",
@@ -125,7 +119,6 @@ langchain-text-splitters = { path = "../text-splitters", editable = true }
 langchain-openai = { path = "../partners/openai", editable = true }
 
 [tool.ruff]
-target-version = "py39"
 exclude = ["tests/integration_tests/examples/non-utf8-encoding.py"]
 
 [tool.mypy]
@@ -197,7 +190,7 @@ ignore-var-parameters = true  # ignore missing documentation for *args and **kwa
 "scripts/*.py" = [
     "INP001",   # Not a package
 ]
-"langchain/chains/constitutional_ai/principles.py" = [
+"langchain_classic/chains/constitutional_ai/principles.py" = [
     "E501", # Line too long
 ]
 "**/retrievers/*time_weighted_retriever.py" = [

From 6814cc7b95fa94b87ee79ed765bea1d03b51b6f6 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 22:49:27 +0530
Subject: [PATCH 5/9] code fixes according to linter

---
 libs/langchain/pyproject.toml                 |  2 +
 .../evaluation/hallucination/test_detector.py | 44 ++++++++++++-------
 2 files changed, 30 insertions(+), 16 deletions(-)

diff --git a/libs/langchain/pyproject.toml b/libs/langchain/pyproject.toml
index 2e97c53c46539..362e413aeec3b 100644
--- a/libs/langchain/pyproject.toml
+++ b/libs/langchain/pyproject.toml
@@ -39,6 +39,7 @@ aws = ["langchain-aws"]
 deepseek = ["langchain-deepseek"]
 xai = ["langchain-xai"]
 perplexity = ["langchain-perplexity"]
+transformers = ["transformers>=4.0.0,<5.0.0"]
 
 [project.urls]
 "Source Code" = "https://github.com/langchain-ai/langchain/tree/master/libs/langchain"
@@ -83,6 +84,7 @@ test_integration = [
     "langchainhub>=0.1.16,<1.0.0",
     "langchain-core",
     "langchain-text-splitters",
+    "transformers>=4.0.0,<5.0.0",
 ]
 lint = [
     "ruff>=0.13.1,<0.14.0",
diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
index c23ef146257ac..cd7eed0e190f3 100644
--- a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
+++ b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
@@ -1,7 +1,9 @@
 import os
-import pytest
+from collections.abc import Generator
 from unittest.mock import patch
 
+import pytest
+
 from langchain.evaluation.hallucination.detector import HallucinationDetector
 
 # -----------------------------
@@ -11,17 +13,18 @@
     "HF_TOKEN" not in os.environ, reason="Hugging Face token not available"
 )
 
+
 @pytest.fixture(scope="module")
 @skip_if_no_hf
 @pytest.mark.requires("integration")
-def detector_real():
+def detector_real() -> HallucinationDetector:
     """Runs only if Hugging Face token is available."""
     return HallucinationDetector(model_name="facebook/bart-large-mnli")
 
 
 @skip_if_no_hf
 @pytest.mark.requires("integration")
-def test_extract_claims_integration(detector_real):
+def test_extract_claims_integration(detector_real: HallucinationDetector) -> None:
     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
     claims = detector_real.extract_claims(text)
     # Check structure and basic logic
@@ -33,11 +36,13 @@ def test_extract_claims_integration(detector_real):
 
 @skip_if_no_hf
 @pytest.mark.requires("integration")
-def test_compute_hallucination_rate_integration(detector_real):
+def test_compute_hallucination_rate_integration(
+    detector_real: HallucinationDetector,
+) -> None:
     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
     evidence = [
         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
-        "Barack Obama was born in Hawaii, not Kenya."
+        "Barack Obama was born in Hawaii, not Kenya.",
     ]
     result = detector_real.compute_hallucination_rate(text, evidence)
 
@@ -49,10 +54,9 @@ def test_compute_hallucination_rate_integration(detector_real):
     unsupported = result["unsupported_claims"]
     hallucination_rate = result["hallucination_rate"]
 
-    # Check logical consistency
     assert total == 2
-    assert 0 <= unsupported <= total  # Dynamic check
-    assert abs(hallucination_rate - unsupported / total) < 1e-6  # Matches formula
+    assert 0 <= unsupported <= total
+    assert abs(hallucination_rate - unsupported / total) < 1e-6
     assert 0 <= hallucination_rate <= 1
 
 
@@ -60,41 +64,43 @@ def test_compute_hallucination_rate_integration(detector_real):
 # Unit Tests (Mocked)
 # -----------------------------
 @pytest.fixture(scope="module")
-def detector_mock():
+def detector_mock() -> Generator[HallucinationDetector, None, None]:
     """Mock pipeline to make unit tests deterministic."""
     with patch("langchain.evaluation.hallucination.detector.pipeline") as mock_pipeline:
         # Mock NLI behavior
         mock_pipeline.return_value = lambda text: [
-            {"label": "ENTAILMENT", "score": 0.9} if "President" in text else {"label": "CONTRADICTION", "score": 0.9}
+            {"label": "ENTAILMENT", "score": 0.9}
+            if "President" in text
+            else {"label": "CONTRADICTION", "score": 0.9}
         ]
         detector = HallucinationDetector(model_name="any")  # Model not loaded
         yield detector
 
 
-def test_extract_claims_mock(detector_mock):
+def test_extract_claims_mock(detector_mock: HallucinationDetector) -> None:
     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
     claims = detector_mock.extract_claims(text)
     assert isinstance(claims, list)
     assert len(claims) == 2
 
 
-def test_verify_claim_supported_mock(detector_mock):
+def test_verify_claim_supported_mock(detector_mock: HallucinationDetector) -> None:
     claim = "Barack Obama was the 44th President of the United States"
     evidence = "Barack Obama served as the 44th President of the United States from 2009 to 2017."
     assert detector_mock.verify_claim(claim, evidence) is True
 
 
-def test_verify_claim_unsupported_mock(detector_mock):
+def test_verify_claim_unsupported_mock(detector_mock: HallucinationDetector) -> None:
     claim = "Barack Obama was born in Kenya"
     evidence = "Barack Obama was born in Hawaii, not Kenya."
     assert detector_mock.verify_claim(claim, evidence) is False
 
 
-def test_compute_hallucination_rate_mock(detector_mock):
+def test_compute_hallucination_rate_mock(detector_mock: HallucinationDetector) -> None:
     text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
     evidence = [
         "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
-        "Barack Obama was born in Hawaii, not Kenya."
+        "Barack Obama was born in Hawaii, not Kenya.",
     ]
     result = detector_mock.compute_hallucination_rate(text, evidence)
     # Validate structure and logical consistency
@@ -102,4 +108,10 @@ def test_compute_hallucination_rate_mock(detector_mock):
         assert key in result
     assert result["total_claims"] == 2
     assert 0 <= result["unsupported_claims"] <= 2
-    assert abs(result["hallucination_rate"] - result["unsupported_claims"] / result["total_claims"]) < 1e-6
+    assert (
+        abs(
+            result["hallucination_rate"]
+            - result["unsupported_claims"] / result["total_claims"]
+        )
+        < 1e-6
+    )

From 29612a513520f17f54051ddb30677871b5e1266f Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 23:11:56 +0530
Subject: [PATCH 6/9] following pep8

---
 .../evaluation/hallucination/test_detector.py | 38 +++++++++++++++----
 1 file changed, 30 insertions(+), 8 deletions(-)

diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
index cd7eed0e190f3..b7f7044d8aeca 100644
--- a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
+++ b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
@@ -10,7 +10,8 @@
 # Integration Tests (Real HF model)
 # -----------------------------
 skip_if_no_hf = pytest.mark.skipif(
-    "HF_TOKEN" not in os.environ, reason="Hugging Face token not available"
+    "HF_TOKEN" not in os.environ,
+    reason="Hugging Face token not available"
 )
 
 
@@ -25,7 +26,10 @@ def detector_real() -> HallucinationDetector:
 @skip_if_no_hf
 @pytest.mark.requires("integration")
 def test_extract_claims_integration(detector_real: HallucinationDetector) -> None:
-    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    text = (
+    "Barack Obama was the 44th President of the United States. "
+    "He was born in Kenya."
+    )
     claims = detector_real.extract_claims(text)
     # Check structure and basic logic
     assert isinstance(claims, list)
@@ -39,9 +43,15 @@ def test_extract_claims_integration(detector_real: HallucinationDetector) -> Non
 def test_compute_hallucination_rate_integration(
     detector_real: HallucinationDetector,
 ) -> None:
-    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    text = (
+    "Barack Obama was the 44th President of the United States. "
+    "He was born in Kenya."
+    )
     evidence = [
-        "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+        (
+        "Barack Obama served as the 44th President of the United States "
+        "from 2009 to 2017."
+        ),
         "Barack Obama was born in Hawaii, not Kenya.",
     ]
     result = detector_real.compute_hallucination_rate(text, evidence)
@@ -78,7 +88,10 @@ def detector_mock() -> Generator[HallucinationDetector, None, None]:
 
 
 def test_extract_claims_mock(detector_mock: HallucinationDetector) -> None:
-    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    text = (
+        "Barack Obama was the 44th President of the United States. "
+        "He was born in Kenya."
+    )
     claims = detector_mock.extract_claims(text)
     assert isinstance(claims, list)
     assert len(claims) == 2
@@ -86,7 +99,10 @@ def test_extract_claims_mock(detector_mock: HallucinationDetector) -> None:
 
 def test_verify_claim_supported_mock(detector_mock: HallucinationDetector) -> None:
     claim = "Barack Obama was the 44th President of the United States"
-    evidence = "Barack Obama served as the 44th President of the United States from 2009 to 2017."
+    evidence = (
+        "Barack Obama served as the 44th President of the United States "
+        "from 2009 to 2017."
+    )
     assert detector_mock.verify_claim(claim, evidence) is True
 
 
@@ -97,9 +113,15 @@ def test_verify_claim_unsupported_mock(detector_mock: HallucinationDetector) ->
 
 
 def test_compute_hallucination_rate_mock(detector_mock: HallucinationDetector) -> None:
-    text = "Barack Obama was the 44th President of the United States. He was born in Kenya."
+    text = (
+        "Barack Obama was the 44th President of the United States. "
+        "He was born in Kenya."
+    )
     evidence = [
-        "Barack Obama served as the 44th President of the United States from 2009 to 2017.",
+        (
+        "Barack Obama served as the 44th President of the United States "
+        "from 2009 to 2017.",
+        ),
         "Barack Obama was born in Hawaii, not Kenya.",
     ]
     result = detector_mock.compute_hallucination_rate(text, evidence)

From 5109f3249a769fa15fa1a4ec40d94a1b9bc876c0 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 23:48:08 +0530
Subject: [PATCH 7/9] transformer issues get fixed

---
 .../evaluation/hallucination/detector.py      | 35 ++++++++++++++-----
 .../evaluation/hallucination/test_detector.py |  2 +-
 2 files changed, 28 insertions(+), 9 deletions(-)

diff --git a/libs/langchain/langchain/evaluation/hallucination/detector.py b/libs/langchain/langchain/evaluation/hallucination/detector.py
index 6db27161886d4..f772060be6def 100644
--- a/libs/langchain/langchain/evaluation/hallucination/detector.py
+++ b/libs/langchain/langchain/evaluation/hallucination/detector.py
@@ -1,18 +1,31 @@
-from typing import List, Dict
-from transformers import pipeline
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from transformers import pipeline as PipelineType
+
+# Lazy import for runtime
+try:
+    from transformers import pipeline
+except ImportError:
+    pipeline = None
+
 
 class HallucinationDetector:
-    """
-    Simple Hallucination Detector using NLI models (e.g., facebook/bart-large-mnli).
+    """Simple Hallucination Detector using NLI models (e.g., facebook/bart-large-mnli).
     - Extract claims (basic sentence split)
     - Verify claims against evidence docs using NLI
     - Compute hallucination rate
     """
 
     def __init__(self, model_name: str = "facebook/bart-large-mnli"):
+        if pipeline is None:
+            raise ImportError(
+                "The 'transformers' package is required for HallucinationDetector. "
+                "Install it with `pip install transformers`."
+            )
         self.nli_pipeline = pipeline("text-classification", model=model_name)
 
-    def extract_claims(self, text: str) -> List[str]:
+    def extract_claims(self, text: str) -> list[str]:
         """Naive sentence-based claim extraction"""
         return [c.strip() for c in text.split(".") if c.strip()]
 
@@ -21,14 +34,20 @@ def verify_claim(self, claim: str, evidence: str) -> bool:
         result = self.nli_pipeline(f"{claim} </s></s> {evidence}")
         return result[0]["label"].lower() == "entailment"
 
-    def verify_claim_multi(self, claim: str, evidence_docs: List[str]) -> bool:
+    def verify_claim_multi(self, claim: str, evidence_docs: list[str]) -> bool:
         """A claim is supported if any evidence doc entails it"""
         return any(self.verify_claim(claim, e) for e in evidence_docs)
 
-    def compute_hallucination_rate(self, text: str, evidence_docs: List[str]) -> Dict[str, float]:
+    def compute_hallucination_rate(
+        self, text: str, evidence_docs: list[str]
+    ) -> dict[str, float]:
         claims = self.extract_claims(text)
         if not claims:
-            return {"total_claims": 0, "unsupported_claims": 0, "hallucination_rate": 0.0}
+            return {
+                "total_claims": 0,
+                "unsupported_claims": 0,
+                "hallucination_rate": 0.0,
+            }
 
         unsupported = sum(not self.verify_claim_multi(c, evidence_docs) for c in claims)
         return {
diff --git a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
index b7f7044d8aeca..63a658a1f31db 100644
--- a/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
+++ b/libs/langchain/tests/integration_tests/evaluation/hallucination/test_detector.py
@@ -120,7 +120,7 @@ def test_compute_hallucination_rate_mock(detector_mock: HallucinationDetector) -
     evidence = [
         (
         "Barack Obama served as the 44th President of the United States "
-        "from 2009 to 2017.",
+        "from 2009 to 2017."
         ),
         "Barack Obama was born in Hawaii, not Kenya.",
     ]

From 7a0bed5a25e721fda346ec2792f8fd746cbe87f2 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Mon, 6 Oct 2025 23:58:11 +0530
Subject: [PATCH 8/9] handling pipeline issue

---
 libs/langchain/langchain/evaluation/hallucination/detector.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/libs/langchain/langchain/evaluation/hallucination/detector.py b/libs/langchain/langchain/evaluation/hallucination/detector.py
index f772060be6def..b90da31cc2a8b 100644
--- a/libs/langchain/langchain/evaluation/hallucination/detector.py
+++ b/libs/langchain/langchain/evaluation/hallucination/detector.py
@@ -3,6 +3,8 @@
 if TYPE_CHECKING:
     from transformers import pipeline as PipelineType
 
+
+pipeline: "PipelineType" | None = None # type: ignore
 # Lazy import for runtime
 try:
     from transformers import pipeline

From 9adee9e5a7c88013c21dc6f3b12b593aa6df1362 Mon Sep 17 00:00:00 2001
From: Vipul <vipulpandeyofficial@gmail.com>
Date: Tue, 7 Oct 2025 00:05:46 +0530
Subject: [PATCH 9/9] added typelinter standard pipeline so won't failed during
 ci

---
 .../evaluation/hallucination/detector.py          | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

diff --git a/libs/langchain/langchain/evaluation/hallucination/detector.py b/libs/langchain/langchain/evaluation/hallucination/detector.py
index b90da31cc2a8b..19265299c7f40 100644
--- a/libs/langchain/langchain/evaluation/hallucination/detector.py
+++ b/libs/langchain/langchain/evaluation/hallucination/detector.py
@@ -1,15 +1,12 @@
-from typing import TYPE_CHECKING
+from typing import Callable, Optional, Any
 
-if TYPE_CHECKING:
-    from transformers import pipeline as PipelineType
-
-
-pipeline: "PipelineType" | None = None # type: ignore
-# Lazy import for runtime
+# Lazy import for optional transformers dependency
+pipeline: Optional[Callable[..., Any]] = None
 try:
-    from transformers import pipeline
+    from transformers import pipeline as _pipeline
+    pipeline = _pipeline
 except ImportError:
-    pipeline = None
+    pass
 
 
 class HallucinationDetector: