transformers-haystack 0.1.0__tar.gz → 0.3.0__tar.gz

transformers_haystack-0.3.0/CHANGELOG.md

@@ -0,0 +1,28 @@
+# Changelog
+
+## [integrations/transformers-v0.2.0] - 2026-07-06
+
+### 📚 Documentation
+
+- Replace old haystack core imports with haystack_integrations paths (#3545)
+
+### 🧪 Testing
+
+- Improve del_hf_env_vars fixture (#3428)
+- Trust test modules under Haystack 3.0's deserialization allowlist (#3537)
+- Make Tool/Agent serialization assertions version-agnostic for Haystack 2.x/3.x (#3533)
+- Force Transformers and Sentence Transformers integration tests to run on CPU (#3550)
+
+### 🧹 Chores
+
+- Improve consistency of integrations folder structure (#3430)
+- Support sync streaming callbacks in async contexts for Haystack 2.x/3.x compatibility (#3534)
+
+
+## [integrations/transformers-v0.1.0] - 2026-06-08
+
+### 🚀 Features
+
+- Move Transformers components from Haystack (#3409)
+
+<!-- generated by git-cliff -->

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: transformers-haystack
-Version: 0.1.0
+Version: 0.3.0
 Summary: Haystack integration for transformers
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/transformers#readme
 Project-URL: Issues, https://github.com/deepset-ai/haystack-core-integrations/issues

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/pyproject.toml

@@ -66,7 +66,7 @@ integration = 'pytest -m "integration" {args:tests}'
 all = 'pytest {args:tests}'
 unit-cov-retry = 'pytest --cov=haystack_integrations --reruns 3 --reruns-delay 30 -x -m "not integration" {args:tests}'
 integration-cov-append-retry = 'pytest --cov=haystack_integrations --cov-append --reruns 3 --reruns-delay 30 -x -m "integration" {args:tests}'
-types = """mypy -p haystack_integrations.components.common.transformers \
+types = """mypy -p haystack_integrations.common.transformers \
 -p haystack_integrations.components.classifiers.transformers \
 -p haystack_integrations.components.extractors.transformers \
 -p haystack_integrations.components.generators.transformers \
@@ -140,6 +140,7 @@ ignore = [
     "PLR0912",
     "PLR0913",
     "PLR0915",
+    "PLR0917",
     # Allow `Any` type - used legitimately for dynamic types and SDK boundaries
     "ANN401",
 ]

{transformers_haystack-0.1.0/src/haystack_integrations/components → transformers_haystack-0.3.0/src/haystack_integrations}/common/transformers/utils.py

@@ -4,11 +4,12 @@
 
 import asyncio
 import copy
+import inspect
 from typing import Any
 
 import torch
 from haystack import logging
-from haystack.dataclasses import AsyncStreamingCallbackT, ComponentInfo, StreamingChunk, SyncStreamingCallbackT
+from haystack.dataclasses import ComponentInfo, StreamingCallbackT, StreamingChunk, SyncStreamingCallbackT
 from haystack.utils.auth import Secret
 from haystack.utils.device import ComponentDevice
 from huggingface_hub import model_info
@@ -198,7 +199,7 @@ class _AsyncHFTokenStreamingHandler(TextStreamer):
     Async streaming handler for TransformersChatGenerator.
 
     Note: This is a helper class for TransformersChatGenerator enabling
-    async streaming of generated text via Haystack Callable[StreamingChunk, Awaitable[None]] callbacks.
+    async streaming of generated text via Haystack StreamingCallbackT callbacks.
 
     Do not use this class directly.
     """
@@ -206,7 +207,7 @@ class _AsyncHFTokenStreamingHandler(TextStreamer):
     def __init__(
         self,
         tokenizer: PreTrainedTokenizerBase,
-        stream_handler: AsyncStreamingCallbackT,
+        stream_handler: StreamingCallbackT,
         stop_words: list[str] | None = None,
         component_info: ComponentInfo | None = None,
     ) -> None:
@@ -228,7 +229,10 @@ class _AsyncHFTokenStreamingHandler(TextStreamer):
         while True:
             try:
                 chunk = await self._queue.get()
-                await self.token_handler(chunk)
+                # sync callbacks are allowed in async contexts with Haystack >= 3.0, so only await async ones
+                callback_result = self.token_handler(chunk)
+                if inspect.isawaitable(callback_result):
+                    await callback_result
                 self._queue.task_done()
             except asyncio.CancelledError:
                 break

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/classifiers/transformers/zero_shot_document_classifier.py

@@ -9,7 +9,7 @@ from haystack import Document, component, default_from_dict, default_to_dict
 from haystack.utils import ComponentDevice, Secret
 from haystack.utils.hf import deserialize_hf_model_kwargs, serialize_hf_model_kwargs
 
-from haystack_integrations.components.common.transformers.utils import _resolve_hf_pipeline_kwargs
+from haystack_integrations.common.transformers.utils import _resolve_hf_pipeline_kwargs
 from transformers import Pipeline as HfPipeline
 from transformers import pipeline
 

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/extractors/transformers/named_entity_extractor.py

@@ -10,7 +10,7 @@ from haystack.utils.auth import Secret
 from haystack.utils.device import ComponentDevice
 from haystack.utils.hf import deserialize_hf_model_kwargs, serialize_hf_model_kwargs
 
-from haystack_integrations.components.common.transformers.utils import _resolve_hf_pipeline_kwargs
+from haystack_integrations.common.transformers.utils import _resolve_hf_pipeline_kwargs
 from transformers import AutoModelForTokenClassification, AutoTokenizer, pipeline
 from transformers import Pipeline as HfPipeline
 

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/generators/transformers/chat/chat_generator.py

@@ -31,7 +31,7 @@ from huggingface_hub import model_info
 from packaging.version import Version
 
 import transformers
-from haystack_integrations.components.common.transformers.utils import (
+from haystack_integrations.common.transformers.utils import (
     _AsyncHFTokenStreamingHandler,
     _HFTokenStreamingHandler,
     _StopWordsCriteria,
@@ -353,7 +353,9 @@ class TransformersChatGenerator:
 
         :param messages: A list of ChatMessage objects representing the input messages. If a string is provided,
             it is converted to a list containing a ChatMessage with user role.
-        :param generation_kwargs: Additional keyword arguments for text generation.
+        :param generation_kwargs: Additional keyword arguments for text generation. These are merged per key with
+            the `generation_kwargs` passed at initialization: keys provided here take precedence, keys set only at
+            initialization are kept.
         :param streaming_callback: An optional callable for handling streaming responses.
         :param tools: A list of Tool and/or Toolset objects, or a single Toolset for which the model can prepare calls.
             If set, it will override the `tools` parameter provided during initialization.
@@ -472,7 +474,9 @@ class TransformersChatGenerator:
         and return values but can be used with `await` in an async code.
 
         :param messages: A list of ChatMessage objects representing the input messages.
-        :param generation_kwargs: Additional keyword arguments for text generation.
+        :param generation_kwargs: Additional keyword arguments for text generation. These are merged per key with
+            the `generation_kwargs` passed at initialization: keys provided here take precedence, keys set only at
+            initialization are kept.
         :param streaming_callback: An optional callable for handling streaming responses.
         :param tools: A list of Tool and/or Toolset objects, or a single Toolset for which the model can prepare calls.
             If set, it will override the `tools` parameter provided during initialization.

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/readers/transformers/extractive_reader.py

@@ -14,7 +14,7 @@ from haystack.utils import ComponentDevice, Device, DeviceMap, Secret
 from haystack.utils.hf import deserialize_hf_model_kwargs, serialize_hf_model_kwargs
 from tokenizers import Encoding
 
-from haystack_integrations.components.common.transformers.utils import _resolve_hf_device_map
+from haystack_integrations.common.transformers.utils import _resolve_hf_device_map
 from transformers import AutoModelForQuestionAnswering, AutoTokenizer
 
 logger = logging.getLogger(__name__)

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/routers/transformers/text_router.py

@@ -8,7 +8,7 @@ from haystack import component, default_from_dict, default_to_dict
 from haystack.utils import ComponentDevice, Secret
 from haystack.utils.hf import deserialize_hf_model_kwargs, serialize_hf_model_kwargs
 
-from haystack_integrations.components.common.transformers.utils import _resolve_hf_pipeline_kwargs
+from haystack_integrations.common.transformers.utils import _resolve_hf_pipeline_kwargs
 from transformers import AutoConfig, Pipeline, pipeline
 
 

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/src/haystack_integrations/components/routers/transformers/zero_shot_text_router.py

@@ -8,7 +8,7 @@ from haystack import component, default_from_dict, default_to_dict
 from haystack.utils import ComponentDevice, Secret
 from haystack.utils.hf import deserialize_hf_model_kwargs, serialize_hf_model_kwargs
 
-from haystack_integrations.components.common.transformers.utils import _resolve_hf_pipeline_kwargs
+from haystack_integrations.common.transformers.utils import _resolve_hf_pipeline_kwargs
 from transformers import Pipeline as HfPipeline
 from transformers import pipeline
 
@@ -24,7 +24,9 @@ class TransformersZeroShotTextRouter:
 
     ```python
     from haystack import Document
-    from haystack.components.embedders import SentenceTransformersTextEmbedder, SentenceTransformersDocumentEmbedder
+    # Requires: pip install sentence-transformers-haystack
+    from haystack_integrations.components.embedders.sentence_transformers import SentenceTransformersTextEmbedder
+    from haystack_integrations.components.embedders.sentence_transformers import SentenceTransformersDocumentEmbedder
     from haystack.components.retrievers import InMemoryEmbeddingRetriever
     from haystack.core.pipeline import Pipeline
     from haystack.document_stores.in_memory import InMemoryDocumentStore
@@ -154,7 +156,11 @@ class TransformersZeroShotTextRouter:
             Dictionary with serialized data.
         """
         serialization_dict = default_to_dict(
-            self, labels=self.labels, huggingface_pipeline_kwargs=self.huggingface_pipeline_kwargs, token=self.token
+            self,
+            labels=self.labels,
+            multi_label=self.multi_label,
+            huggingface_pipeline_kwargs=self.huggingface_pipeline_kwargs,
+            token=self.token,
         )
 
         huggingface_pipeline_kwargs = serialization_dict["init_parameters"]["huggingface_pipeline_kwargs"]

transformers_haystack-0.3.0/tests/conftest.py

@@ -0,0 +1,39 @@
+# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
+#
+# SPDX-License-Identifier: Apache-2.0
+
+import os
+
+import pytest
+from haystack.document_stores.in_memory import InMemoryDocumentStore
+
+
+@pytest.fixture()
+def del_hf_env_vars_if_empty(monkeypatch):
+    """
+    Delete Hugging Face environment variables for tests if empty.
+
+    Prevents passing empty tokens to Hugging Face, which would cause API calls to fail.
+    This is particularly relevant for PRs opened from forks, where secrets are not available
+    and empty environment variables might be set instead of being removed.
+
+    See https://github.com/deepset-ai/haystack/issues/8811 for more details.
+    """
+    for var in ("HF_API_TOKEN", "HF_TOKEN"):
+        if not os.environ.get(var, "").strip():
+            monkeypatch.delenv(var, raising=False)
+
+
+@pytest.fixture()
+def in_memory_doc_store():
+    return InMemoryDocumentStore()
+
+
+@pytest.fixture(autouse=True)
+def allow_deserialization_of_test_modules(monkeypatch):
+    """
+    haystack-ai >= 3.0 refuses to deserialize classes and callables from modules outside its
+    trusted-module allowlist. Tools and callbacks defined in the test modules live outside that
+    allowlist, so trust them explicitly; haystack-ai 2.x ignores this environment variable.
+    """
+    monkeypatch.setenv("HAYSTACK_DESERIALIZATION_ALLOWLIST", "tests,test_*")

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_chat_generator.py

@@ -206,20 +206,10 @@ class TestTransformersChatGenerator:
         assert init_params["streaming_callback"] is None
         assert init_params["chat_template"] == "irrelevant"
         assert init_params["enable_thinking"] is True
-        assert init_params["tools"] == [
-            {
-                "type": "haystack.tools.tool.Tool",
-                "data": {
-                    "inputs_from_state": None,
-                    "name": "weather",
-                    "outputs_to_state": None,
-                    "outputs_to_string": None,
-                    "description": "useful to determine the weather in a given location",
-                    "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]},
-                    "function": "tests.test_chat_generator.get_weather",
-                },
-            }
-        ]
+
+        # deserializing the serialized component must reproduce the original tools
+        loaded = TransformersChatGenerator.from_dict(result)
+        assert loaded.tools == tools
 
     def test_from_dict(self, model_info_mock, tools):
         generator = TransformersChatGenerator(
@@ -256,7 +246,7 @@ class TestTransformersChatGenerator:
         }
 
     @patch("haystack_integrations.components.generators.transformers.chat.chat_generator.pipeline")
-    def test_warm_up(self, pipeline_mock, del_hf_env_vars):
+    def test_warm_up(self, pipeline_mock, del_hf_env_vars_if_empty):
         generator = TransformersChatGenerator(
             model="mistralai/Mistral-7B-Instruct-v0.2", task="text-generation", device=ComponentDevice.from_str("cpu")
         )
@@ -266,11 +256,14 @@ class TestTransformersChatGenerator:
         generator.warm_up()
 
         pipeline_mock.assert_called_once_with(
-            model="mistralai/Mistral-7B-Instruct-v0.2", task="text-generation", token=None, device="cpu"
+            model="mistralai/Mistral-7B-Instruct-v0.2",
+            task="text-generation",
+            token=generator.token.resolve_value(),
+            device="cpu",
         )
 
     @patch("haystack_integrations.components.generators.transformers.chat.chat_generator.pipeline")
-    def test_warm_up_with_tools(self, pipeline_mock, del_hf_env_vars):
+    def test_warm_up_with_tools(self, pipeline_mock, del_hf_env_vars_if_empty):
         """Test that warm_up() calls warm_up on tools and is idempotent."""
 
         # Create a mock tool that tracks if warm_up() was called
@@ -324,7 +317,7 @@ class TestTransformersChatGenerator:
         pipeline_mock.assert_called_once()
 
     @patch("haystack_integrations.components.generators.transformers.chat.chat_generator.pipeline")
-    def test_warm_up_with_no_tools(self, pipeline_mock, del_hf_env_vars):
+    def test_warm_up_with_no_tools(self, pipeline_mock, del_hf_env_vars_if_empty):
         """Test that warm_up() works when no tools are provided."""
 
         generator = TransformersChatGenerator(
@@ -349,7 +342,7 @@ class TestTransformersChatGenerator:
         pipeline_mock.assert_called_once()
 
     @patch("haystack_integrations.components.generators.transformers.chat.chat_generator.pipeline")
-    def test_warm_up_with_multiple_tools(self, pipeline_mock, del_hf_env_vars):
+    def test_warm_up_with_multiple_tools(self, pipeline_mock, del_hf_env_vars_if_empty):
         """Test that warm_up() works with multiple tools."""
 
         # Track warm_up calls
@@ -447,6 +440,19 @@ class TestTransformersChatGenerator:
         assert chat_message.is_from(ChatRole.ASSISTANT)
         assert chat_message.text == "Berlin is cool"
 
+    def test_run_with_generation_kwargs(self, model_info_mock, mock_pipeline_with_tokenizer, chat_messages):
+        generator = TransformersChatGenerator(
+            model="meta-llama/Llama-2-13b-chat-hf",
+            generation_kwargs={"max_new_tokens": 100, "temperature": 0.5},
+        )
+        generator.pipeline = mock_pipeline_with_tokenizer
+
+        generator.run(messages=chat_messages, generation_kwargs={"temperature": 0.9})
+
+        _, kwargs = generator.pipeline.call_args
+        assert kwargs["max_new_tokens"] == 100
+        assert kwargs["temperature"] == 0.9
+
     def test_run_with_streaming_callback(self, model_info_mock, mock_pipeline_with_tokenizer, chat_messages):
         # Define the streaming callback function
         def streaming_callback_fn(chunk: StreamingChunk): ...
@@ -507,11 +513,15 @@ class TestTransformersChatGenerator:
 
     @pytest.mark.integration
     @pytest.mark.flaky(reruns=3, reruns_delay=10)
-    def test_live_run(self, del_hf_env_vars):
+    def test_live_run(self, del_hf_env_vars_if_empty):
         """Test live run with default behavior (no thinking)."""
         messages = [ChatMessage.from_user("Please create a summary about the following topic: Climate change")]
 
-        llm = TransformersChatGenerator(model="Qwen/Qwen3-0.6B", generation_kwargs={"max_new_tokens": 50})
+        llm = TransformersChatGenerator(
+            model="Qwen/Qwen3-0.6B",
+            generation_kwargs={"max_new_tokens": 50},
+            device=ComponentDevice.from_str("cpu"),
+        )
 
         result = llm.run(messages)
 
@@ -521,12 +531,15 @@ class TestTransformersChatGenerator:
 
     @pytest.mark.integration
     @pytest.mark.flaky(reruns=3, reruns_delay=10)
-    def test_live_run_thinking(self, del_hf_env_vars):
+    def test_live_run_thinking(self, del_hf_env_vars_if_empty):
         """Test live run with enable_thinking=True."""
         messages = [ChatMessage.from_user("What is 2+2?")]
 
         llm = TransformersChatGenerator(
-            model="Qwen/Qwen3-0.6B", generation_kwargs={"max_new_tokens": 450}, enable_thinking=True
+            model="Qwen/Qwen3-0.6B",
+            generation_kwargs={"max_new_tokens": 450},
+            enable_thinking=True,
+            device=ComponentDevice.from_str("cpu"),
         )
 
         result = llm.run(messages)
@@ -705,6 +718,21 @@ class TestTransformersChatGeneratorAsync:
         assert chat_message.text == "Berlin is cool"
         generator.shutdown()
 
+    @pytest.mark.asyncio
+    async def test_run_async_with_generation_kwargs(self, model_info_mock, mock_pipeline_with_tokenizer, chat_messages):
+        generator = TransformersChatGenerator(
+            model="meta-llama/Llama-2-13b-chat-hf",
+            generation_kwargs={"max_new_tokens": 100, "temperature": 0.5},
+        )
+        generator.pipeline = mock_pipeline_with_tokenizer
+
+        await generator.run_async(messages=chat_messages, generation_kwargs={"temperature": 0.9})
+
+        _, kwargs = generator.pipeline.call_args
+        assert kwargs["max_new_tokens"] == 100
+        assert kwargs["temperature"] == 0.9
+        generator.shutdown()
+
     @pytest.mark.asyncio
     async def test_run_async_with_string_input(self, model_info_mock, mock_pipeline_with_tokenizer):
         generator = TransformersChatGenerator(model="meta-llama/Llama-2-13b-chat-hf")
@@ -799,30 +827,10 @@ class TestTransformersChatGeneratorAsync:
         generator.pipeline = mock_pipeline_with_tokenizer
         data = generator.to_dict()
 
-        expected_tools_data = {
-            "type": "haystack.tools.toolset.Toolset",
-            "data": {
-                "tools": [
-                    {
-                        "type": "haystack.tools.tool.Tool",
-                        "data": {
-                            "name": "weather",
-                            "description": "useful to determine the weather in a given location",
-                            "parameters": {
-                                "type": "object",
-                                "properties": {"city": {"type": "string"}},
-                                "required": ["city"],
-                            },
-                            "function": "tests.test_chat_generator.get_weather",
-                            "outputs_to_string": None,
-                            "inputs_from_state": None,
-                            "outputs_to_state": None,
-                        },
-                    }
-                ]
-            },
-        }
-        assert data["init_parameters"]["tools"] == expected_tools_data
+        # deserializing the serialized component must reproduce the original toolset
+        loaded = TransformersChatGenerator.from_dict(data)
+        assert isinstance(loaded.tools, Toolset)
+        assert list(loaded.tools) == list(toolset)
 
     @pytest.mark.asyncio
     async def test_run_async_with_streaming_callback(self, model_info_mock, mock_pipeline_with_tokenizer):
@@ -865,7 +873,7 @@ class TestTransformersChatGeneratorAsync:
     @pytest.mark.integration
     @pytest.mark.flaky(reruns=3, reruns_delay=10)
     @pytest.mark.asyncio
-    async def test_live_run_async_with_streaming(self, del_hf_env_vars):
+    async def test_live_run_async_with_streaming(self, del_hf_env_vars_if_empty):
         """Test async streaming with a live model."""
         streaming_chunks = []
 
@@ -873,7 +881,10 @@ class TestTransformersChatGeneratorAsync:
             streaming_chunks.append(chunk)
 
         llm = TransformersChatGenerator(
-            model="Qwen/Qwen3-0.6B", generation_kwargs={"max_new_tokens": 50}, streaming_callback=streaming_callback
+            model="Qwen/Qwen3-0.6B",
+            generation_kwargs={"max_new_tokens": 50},
+            streaming_callback=streaming_callback,
+            device=ComponentDevice.from_str("cpu"),
         )
 
         response = await llm.run_async(

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_extractive_reader.py

@@ -640,7 +640,7 @@ def test_warm_up_use_hf_token(mocked_automodel, mocked_autotokenizer, initialize
     "haystack_integrations.components.readers.transformers.extractive_reader."
     "AutoModelForQuestionAnswering.from_pretrained"
 )
-def test_device_map_auto(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars):
+def test_device_map_auto(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars_if_empty):
     reader = TransformersExtractiveReader("deepset/roberta-base-squad2", model_kwargs={"device_map": "auto"})
     auto_device = ComponentDevice.resolve_device(None)
 
@@ -651,7 +651,9 @@ def test_device_map_auto(mocked_automodel, _mocked_autotokenizer, del_hf_env_var
     mocked_automodel.return_value = MockedModel()
     reader.warm_up()
 
-    mocked_automodel.assert_called_once_with("deepset/roberta-base-squad2", token=None, device_map="auto")
+    mocked_automodel.assert_called_once_with(
+        "deepset/roberta-base-squad2", token=reader.token.resolve_value(), device_map="auto"
+    )
     assert reader.device == ComponentDevice.from_multiple(DeviceMap.from_hf({"": auto_device.to_hf()}))
 
 
@@ -660,7 +662,7 @@ def test_device_map_auto(mocked_automodel, _mocked_autotokenizer, del_hf_env_var
     "haystack_integrations.components.readers.transformers.extractive_reader."
     "AutoModelForQuestionAnswering.from_pretrained"
 )
-def test_device_map_str(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars):
+def test_device_map_str(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars_if_empty):
     reader = TransformersExtractiveReader("deepset/roberta-base-squad2", model_kwargs={"device_map": "cpu:0"})
 
     class MockedModel:
@@ -670,7 +672,9 @@ def test_device_map_str(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars
     mocked_automodel.return_value = MockedModel()
     reader.warm_up()
 
-    mocked_automodel.assert_called_once_with("deepset/roberta-base-squad2", token=None, device_map="cpu:0")
+    mocked_automodel.assert_called_once_with(
+        "deepset/roberta-base-squad2", token=reader.token.resolve_value(), device_map="cpu:0"
+    )
     assert reader.device == ComponentDevice.from_multiple(DeviceMap.from_hf({"": "cpu:0"}))
 
 
@@ -679,7 +683,7 @@ def test_device_map_str(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars
     "haystack_integrations.components.readers.transformers.extractive_reader."
     "AutoModelForQuestionAnswering.from_pretrained"
 )
-def test_device_map_dict(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars):
+def test_device_map_dict(mocked_automodel, _mocked_autotokenizer, del_hf_env_vars_if_empty):
     reader = TransformersExtractiveReader(
         "deepset/roberta-base-squad2", model_kwargs={"device_map": {"layer_1": 1, "classifier": "cpu"}}
     )
@@ -692,7 +696,9 @@ def test_device_map_dict(mocked_automodel, _mocked_autotokenizer, del_hf_env_var
     reader.warm_up()
 
     mocked_automodel.assert_called_once_with(
-        "deepset/roberta-base-squad2", token=None, device_map={"layer_1": 1, "classifier": "cpu"}
+        "deepset/roberta-base-squad2",
+        token=reader.token.resolve_value(),
+        device_map={"layer_1": 1, "classifier": "cpu"},
     )
     assert reader.device == ComponentDevice.from_multiple(DeviceMap.from_hf({"layer_1": 1, "classifier": "cpu"}))
 
@@ -907,11 +913,9 @@ class TestDeduplication:
 
 
 @pytest.mark.integration
-def test_t5(del_hf_env_vars):
-    reader = TransformersExtractiveReader("sjrhuschlee/flan-t5-base-squad2")
-    answers = reader.run(example_queries[0], example_documents[0], top_k=2)[
-        "answers"
-    ]  # remove indices when batching support is reintroduced
+def test_t5(del_hf_env_vars_if_empty):
+    reader = TransformersExtractiveReader("sjrhuschlee/flan-t5-base-squad2", device=ComponentDevice.from_str("cpu"))
+    answers = reader.run(example_queries[0], example_documents[0], top_k=2)["answers"]
     assert answers[0].data == "Olaf Scholz"
     assert answers[0].score == pytest.approx(0.8085031509399414, abs=1e-5)
     assert answers[1].data == "Angela Merkel"
@@ -919,22 +923,12 @@ def test_t5(del_hf_env_vars):
     assert answers[2].data is None
     assert answers[2].score == pytest.approx(0.0378925803599941, abs=1e-5)
     assert len(answers) == 3
-    # Uncomment assertions below when batching is reintroduced
-    # assert answers[0][2].score == pytest.approx(0.051331606147570596)
-    # assert answers[1][0].data == "Jerry"
-    # assert answers[1][0].score == pytest.approx(0.7413333654403687)
-    # assert answers[1][1].data == "Olaf Scholz"
-    # assert answers[1][1].score == pytest.approx(0.7266613841056824)
-    # assert answers[1][2].data is None
-    # assert answers[1][2].score == pytest.approx(0.0707035798685709)
 
 
 @pytest.mark.integration
-def test_roberta(del_hf_env_vars):
-    reader = TransformersExtractiveReader("deepset/tinyroberta-squad2")
-    answers = reader.run(example_queries[0], example_documents[0], top_k=2)[
-        "answers"
-    ]  # remove indices when batching is reintroduced
+def test_roberta(del_hf_env_vars_if_empty):
+    reader = TransformersExtractiveReader("deepset/tinyroberta-squad2", device=ComponentDevice.from_str("cpu"))
+    answers = reader.run(example_queries[0], example_documents[0], top_k=2)["answers"]
     assert answers[0].data == "Olaf Scholz"
     assert answers[0].score == pytest.approx(0.8614975214004517)
     assert answers[1].data == "Angela Merkel"
@@ -942,16 +936,3 @@ def test_roberta(del_hf_env_vars):
     assert answers[2].data is None
     assert answers[2].score == pytest.approx(0.019673851661650588, abs=1e-5)
     assert len(answers) == 3
-    # uncomment assertions below when there is batching in v2
-    # assert answers[0][0].data == "Olaf Scholz"
-    # assert answers[0][0].score == pytest.approx(0.8614975214004517)
-    # assert answers[0][1].data == "Angela Merkel"
-    # assert answers[0][1].score == pytest.approx(0.857952892780304)
-    # assert answers[0][2].data is None
-    # assert answers[0][2].score == pytest.approx(0.0196738764278237)
-    # assert answers[1][0].data == "Jerry"
-    # assert answers[1][0].score == pytest.approx(0.7048940658569336)
-    # assert answers[1][1].data == "Olaf Scholz"
-    # assert answers[1][1].score == pytest.approx(0.6604189872741699)
-    # assert answers[1][2].data is None
-    # assert answers[1][2].score == pytest.approx(0.1002123719777046)

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_named_entity_extractor.py

@@ -101,7 +101,7 @@ def test_named_entity_extractor_serde():
         _ = TransformersNamedEntityExtractor.from_dict(serde_data)
 
 
-def test_to_dict_default(del_hf_env_vars):
+def test_to_dict_default(del_hf_env_vars_if_empty):
     component = TransformersNamedEntityExtractor(
         model="dslim/bert-base-NER",
         device=ComponentDevice.from_str("mps"),
@@ -144,7 +144,7 @@ def test_to_dict_with_parameters():
     }
 
 
-def test_named_entity_extractor_from_dict_no_default_parameters(del_hf_env_vars):
+def test_named_entity_extractor_from_dict_no_default_parameters(del_hf_env_vars_if_empty):
     data = {
         "type": COMPONENT_TYPE,
         "init_parameters": {"model": "dslim/bert-base-NER"},
@@ -226,16 +226,16 @@ def test_named_entity_extractor_run_fails_with_wrong_number_of_annotations():
 
 
 @pytest.mark.integration
-def test_ner_extractor_init(del_hf_env_vars):
-    extractor = TransformersNamedEntityExtractor(model="dslim/bert-base-NER")
+def test_ner_extractor_init(del_hf_env_vars_if_empty):
+    extractor = TransformersNamedEntityExtractor(model="dslim/bert-base-NER", device=ComponentDevice.from_str("cpu"))
     extractor.warm_up()
     assert extractor.initialized
 
 
 @pytest.mark.integration
 @pytest.mark.parametrize("batch_size", [1, 3])
-def test_ner_extractor(raw_texts, hf_annotations, batch_size, del_hf_env_vars):
-    extractor = TransformersNamedEntityExtractor(model="dslim/bert-base-NER")
+def test_ner_extractor(raw_texts, hf_annotations, batch_size, del_hf_env_vars_if_empty):
+    extractor = TransformersNamedEntityExtractor(model="dslim/bert-base-NER", device=ComponentDevice.from_str("cpu"))
     extractor.warm_up()
 
     _extract_and_check_predictions(extractor, raw_texts, hf_annotations, batch_size)
@@ -248,7 +248,7 @@ def test_ner_extractor(raw_texts, hf_annotations, batch_size, del_hf_env_vars):
     reason="Export an env var called HF_API_TOKEN or HF_TOKEN containing the Hugging Face token to run this test.",
 )
 def test_ner_extractor_private_models(raw_texts, hf_annotations, batch_size):
-    extractor = TransformersNamedEntityExtractor(model="deepset/bert-base-NER")
+    extractor = TransformersNamedEntityExtractor(model="deepset/bert-base-NER", device=ComponentDevice.from_str("cpu"))
     extractor.warm_up()
 
     _extract_and_check_predictions(extractor, raw_texts, hf_annotations, batch_size)
@@ -256,11 +256,11 @@ def test_ner_extractor_private_models(raw_texts, hf_annotations, batch_size):
 
 @pytest.mark.integration
 @pytest.mark.parametrize("batch_size", [1, 3])
-def test_ner_extractor_in_pipeline(raw_texts, hf_annotations, batch_size, del_hf_env_vars):
+def test_ner_extractor_in_pipeline(raw_texts, hf_annotations, batch_size, del_hf_env_vars_if_empty):
     pipeline = Pipeline()
     pipeline.add_component(
         name="ner_extractor",
-        instance=TransformersNamedEntityExtractor(model="dslim/bert-base-NER"),
+        instance=TransformersNamedEntityExtractor(model="dslim/bert-base-NER", device=ComponentDevice.from_str("cpu")),
     )
 
     outputs = pipeline.run(

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_text_router.py

@@ -54,7 +54,7 @@ class TestTransformersTextRouter:
         }
 
     @patch("haystack_integrations.components.routers.transformers.text_router.AutoConfig.from_pretrained")
-    def test_from_dict(self, mock_auto_config_from_pretrained, del_hf_env_vars):
+    def test_from_dict(self, mock_auto_config_from_pretrained, del_hf_env_vars_if_empty):
         mock_auto_config_from_pretrained.return_value = MagicMock(label2id={"en": 0, "de": 1})
         data = {
             "type": COMPONENT_TYPE,
@@ -79,11 +79,11 @@ class TestTransformersTextRouter:
             "model": "papluca/xlm-roberta-base-language-detection",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "text-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
     @patch("haystack_integrations.components.routers.transformers.text_router.AutoConfig.from_pretrained")
-    def test_from_dict_no_default_parameters(self, mock_auto_config_from_pretrained, del_hf_env_vars):
+    def test_from_dict_no_default_parameters(self, mock_auto_config_from_pretrained, del_hf_env_vars_if_empty):
         mock_auto_config_from_pretrained.return_value = MagicMock(label2id={"en": 0, "de": 1})
         data = {
             "type": COMPONENT_TYPE,
@@ -99,11 +99,11 @@ class TestTransformersTextRouter:
             "model": "papluca/xlm-roberta-base-language-detection",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "text-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
     @patch("haystack_integrations.components.routers.transformers.text_router.AutoConfig.from_pretrained")
-    def test_from_dict_with_cpu_device(self, mock_auto_config_from_pretrained, del_hf_env_vars):
+    def test_from_dict_with_cpu_device(self, mock_auto_config_from_pretrained, del_hf_env_vars_if_empty):
         mock_auto_config_from_pretrained.return_value = MagicMock(label2id={"en": 0, "de": 1})
         data = {
             "type": COMPONENT_TYPE,
@@ -128,7 +128,7 @@ class TestTransformersTextRouter:
             "model": "papluca/xlm-roberta-base-language-detection",
             "device": ComponentDevice.from_str("cpu").to_hf(),
             "task": "text-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
     @patch("haystack_integrations.components.routers.transformers.text_router.AutoConfig.from_pretrained")
@@ -172,8 +172,10 @@ class TestTransformersTextRouter:
         assert out == {"en": "What is the color of the sky?"}
 
     @pytest.mark.integration
-    def test_run(self, del_hf_env_vars):
-        router = TransformersTextRouter(model="papluca/xlm-roberta-base-language-detection")
+    def test_run(self, del_hf_env_vars_if_empty):
+        router = TransformersTextRouter(
+            model="papluca/xlm-roberta-base-language-detection", device=ComponentDevice.from_str("cpu")
+        )
         out = router.run("What is the color of the sky?")
         assert set(router.labels) == {
             "ar",
@@ -201,7 +203,11 @@ class TestTransformersTextRouter:
         assert out == {"en": "What is the color of the sky?"}
 
     @pytest.mark.integration
-    def test_wrong_labels(self, del_hf_env_vars):
-        router = TransformersTextRouter(model="papluca/xlm-roberta-base-language-detection", labels=["en", "de"])
+    def test_wrong_labels(self, del_hf_env_vars_if_empty):
+        router = TransformersTextRouter(
+            model="papluca/xlm-roberta-base-language-detection",
+            labels=["en", "de"],
+            device=ComponentDevice.from_str("cpu"),
+        )
         with pytest.raises(ValueError):
             router.warm_up()

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_utils.py

@@ -10,7 +10,7 @@ import torch
 from haystack.utils.device import ComponentDevice
 from transformers import AutoTokenizer, PreTrainedTokenizerFast
 
-from haystack_integrations.components.common.transformers.utils import (
+from haystack_integrations.common.transformers.utils import (
     _resolve_hf_device_map,
     _StopWordsCriteria,
 )

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_zero_shot_document_classifier.py

@@ -52,7 +52,7 @@ class TestTransformersZeroShotDocumentClassifier:
             },
         }
 
-    def test_from_dict(self, del_hf_env_vars):
+    def test_from_dict(self, del_hf_env_vars_if_empty):
         data = {
             "type": COMPONENT_TYPE,
             "init_parameters": {
@@ -80,10 +80,10 @@ class TestTransformersZeroShotDocumentClassifier:
             "model": "cross-encoder/nli-deberta-v3-xsmall",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "zero-shot-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
-    def test_from_dict_no_default_parameters(self, del_hf_env_vars):
+    def test_from_dict_no_default_parameters(self, del_hf_env_vars_if_empty):
         data = {
             "type": COMPONENT_TYPE,
             "init_parameters": {"model": "cross-encoder/nli-deberta-v3-xsmall", "labels": ["positive", "negative"]},
@@ -98,7 +98,7 @@ class TestTransformersZeroShotDocumentClassifier:
             "model": "cross-encoder/nli-deberta-v3-xsmall",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "zero-shot-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
     @patch("haystack_integrations.components.classifiers.transformers.zero_shot_document_classifier.pipeline")
@@ -165,9 +165,11 @@ class TestTransformersZeroShotDocumentClassifier:
             component.run(documents=documents)
 
     @pytest.mark.integration
-    def test_run(self, del_hf_env_vars):
+    def test_run(self, del_hf_env_vars_if_empty):
         component = TransformersZeroShotDocumentClassifier(
-            model="cross-encoder/nli-deberta-v3-xsmall", labels=["positive", "negative"]
+            model="cross-encoder/nli-deberta-v3-xsmall",
+            labels=["positive", "negative"],
+            device=ComponentDevice.from_str("cpu"),
         )
         positive_document = Document(content="That's good. I like it. " * 1000)
         negative_document = Document(content="That's bad. I don't like it.")

{transformers_haystack-0.1.0 → transformers_haystack-0.3.0}/tests/test_zero_shot_text_router.py

@@ -22,6 +22,7 @@ class TestTransformersZeroShotTextRouter:
             "type": COMPONENT_TYPE,
             "init_parameters": {
                 "labels": ["query", "passage"],
+                "multi_label": False,
                 "token": {"env_vars": ["HF_API_TOKEN", "HF_TOKEN"], "strict": False, "type": "env_var"},
                 "huggingface_pipeline_kwargs": {
                     "model": "MoritzLaurer/deberta-v3-base-zeroshot-v1.1-all-33",
@@ -31,7 +32,15 @@ class TestTransformersZeroShotTextRouter:
             },
         }
 
-    def test_from_dict(self, del_hf_env_vars):
+    def test_multi_label_survives_a_serialization_round_trip(self, del_hf_env_vars_if_empty):
+        """`multi_label` changes how the pipeline normalizes scores, so losing it changes routing decisions."""
+        router = TransformersZeroShotTextRouter(labels=["query", "passage"], multi_label=True)
+
+        restored = TransformersZeroShotTextRouter.from_dict(router.to_dict())
+
+        assert restored.multi_label is True
+
+    def test_from_dict(self, del_hf_env_vars_if_empty):
         data = {
             "type": COMPONENT_TYPE,
             "init_parameters": {
@@ -55,10 +64,10 @@ class TestTransformersZeroShotTextRouter:
             "model": "MoritzLaurer/deberta-v3-base-zeroshot-v1.1-all-33",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "zero-shot-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
-    def test_from_dict_no_default_parameters(self, del_hf_env_vars):
+    def test_from_dict_no_default_parameters(self, del_hf_env_vars_if_empty):
         data = {
             "type": COMPONENT_TYPE,
             "init_parameters": {"labels": ["query", "passage"]},
@@ -73,7 +82,7 @@ class TestTransformersZeroShotTextRouter:
             "model": "MoritzLaurer/deberta-v3-base-zeroshot-v1.1-all-33",
             "device": ComponentDevice.resolve_device(None).to_hf(),
             "task": "zero-shot-classification",
-            "token": None,
+            "token": component.token.resolve_value(),
         }
 
     @patch("haystack_integrations.components.routers.transformers.zero_shot_text_router.pipeline")
@@ -110,8 +119,8 @@ class TestTransformersZeroShotTextRouter:
         assert out == {"query": "What is the color of the sky?"}
 
     @pytest.mark.integration
-    def test_run(self, del_hf_env_vars):
-        router = TransformersZeroShotTextRouter(labels=["query", "passage"])
+    def test_run(self, del_hf_env_vars_if_empty):
+        router = TransformersZeroShotTextRouter(labels=["query", "passage"], device=ComponentDevice.from_str("cpu"))
         out = router.run("What is the color of the sky?")
         assert router.pipeline is not None
         assert out == {"query": "What is the color of the sky?"}

transformers_haystack-0.1.0/tests/conftest.py

@@ -1,26 +0,0 @@
-# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
-#
-# SPDX-License-Identifier: Apache-2.0
-
-import pytest
-from haystack.document_stores.in_memory import InMemoryDocumentStore
-
-
-@pytest.fixture()
-def del_hf_env_vars(monkeypatch):
-    """
-    Delete Hugging Face environment variables for tests.
-
-    Prevents passing empty tokens to Hugging Face, which would cause API calls to fail.
-    This is particularly relevant for PRs opened from forks, where secrets are not available
-    and empty environment variables might be set instead of being removed.
-
-    See https://github.com/deepset-ai/haystack/issues/8811 for more details.
-    """
-    monkeypatch.delenv("HF_API_TOKEN", raising=False)
-    monkeypatch.delenv("HF_TOKEN", raising=False)
-
-
-@pytest.fixture()
-def in_memory_doc_store():
-    return InMemoryDocumentStore()