haystack-experimental 0.14.1__tar.gz → 0.14.3__tar.gz

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: haystack-experimental
-Version: 0.14.1
+Version: 0.14.3
 Summary: Experimental components and features for the Haystack LLM framework.
 Project-URL: CI: GitHub, https://github.com/deepset-ai/haystack-experimental/actions
 Project-URL: GitHub: issues, https://github.com/deepset-ai/haystack-experimental/issues

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/agents/agent.py

@@ -5,6 +5,7 @@
 # pylint: disable=wrong-import-order,wrong-import-position,ungrouped-imports
 # ruff: noqa: I001
 
+import inspect
 from dataclasses import dataclass
 from typing import Any, Optional, Union
 
@@ -171,6 +172,7 @@ class Agent(HaystackAgent):
         requires_async: bool,
         *,
         system_prompt: Optional[str] = None,
+        generation_kwargs: Optional[dict[str, Any]] = None,
         tools: Optional[Union[ToolsType, list[str]]] = None,
         **kwargs: dict[str, Any],
     ) -> _ExecutionContext:
@@ -185,14 +187,28 @@ class Agent(HaystackAgent):
             When passing tool names, tools are selected from the Agent's originally configured tools.
         :param kwargs: Additional data to pass to the State used by the Agent.
         """
-        exe_context = super(Agent, self)._initialize_fresh_execution(
-            messages=messages,
-            streaming_callback=streaming_callback,
-            requires_async=requires_async,
-            system_prompt=system_prompt,
-            tools=tools,
-            **kwargs,
-        )
+        # The PR https://github.com/deepset-ai/haystack/pull/9616 added the generation_kwargs parameter to
+        # _initialize_fresh_execution. This change has been released in Haystack 2.20.0.
+        # To maintain compatibility with Haystack 2.19 we check the number of parameters and call accordingly.
+        if inspect.signature(super(Agent, self)._initialize_fresh_execution).parameters.get("generation_kwargs"):
+            exe_context = super(Agent, self)._initialize_fresh_execution(
+                messages=messages,
+                streaming_callback=streaming_callback,
+                requires_async=requires_async,
+                system_prompt=system_prompt,
+                generation_kwargs=generation_kwargs,
+                tools=tools,
+                **kwargs,
+            )
+        else:
+            exe_context = super(Agent, self)._initialize_fresh_execution(
+                messages=messages,
+                streaming_callback=streaming_callback,
+                requires_async=requires_async,
+                system_prompt=system_prompt,
+                tools=tools,
+                **kwargs,
+            )
         # NOTE: 1st difference with parent method to add this to tool_invoker_inputs
         if self._tool_invoker:
             exe_context.tool_invoker_inputs["enable_streaming_callback_passthrough"] = (
@@ -212,6 +228,7 @@ class Agent(HaystackAgent):
         streaming_callback: Optional[StreamingCallbackT],
         requires_async: bool,
         *,
+        generation_kwargs: Optional[dict[str, Any]] = None,
         tools: Optional[Union[ToolsType, list[str]]] = None,
     ) -> _ExecutionContext:
         """
@@ -220,12 +237,26 @@ class Agent(HaystackAgent):
         :param snapshot: An AgentSnapshot containing the state of a previously saved agent execution.
         :param streaming_callback: Optional callback for streaming responses.
         :param requires_async: Whether the agent run requires asynchronous execution.
+        :param generation_kwargs: Additional keyword arguments for chat generator. These parameters will
+            override the parameters passed during component initialization.
         :param tools: Optional list of Tool objects, a Toolset, or list of tool names to use for this run.
             When passing tool names, tools are selected from the Agent's originally configured tools.
         """
-        exe_context = super(Agent, self)._initialize_from_snapshot(
-            snapshot=snapshot, streaming_callback=streaming_callback, requires_async=requires_async, tools=tools
-        )
+        # The PR https://github.com/deepset-ai/haystack/pull/9616 added the generation_kwargs parameter to
+        # _initialize_from_snapshot. This change has been released in Haystack 2.20.0.
+        # To maintain compatibility with Haystack 2.19 we check the number of parameters and call accordingly.
+        if inspect.signature(super(Agent, self)._initialize_from_snapshot).parameters.get("generation_kwargs"):
+            exe_context = super(Agent, self)._initialize_from_snapshot(
+                snapshot=snapshot,
+                streaming_callback=streaming_callback,
+                requires_async=requires_async,
+                generation_kwargs=generation_kwargs,
+                tools=tools,
+            )
+        else:
+            exe_context = super(Agent, self)._initialize_from_snapshot(
+                snapshot=snapshot, streaming_callback=streaming_callback, requires_async=requires_async, tools=tools
+            )
         # NOTE: 1st difference with parent method to add this to tool_invoker_inputs
         if self._tool_invoker:
             exe_context.tool_invoker_inputs["enable_streaming_callback_passthrough"] = (
@@ -247,6 +278,7 @@ class Agent(HaystackAgent):
         messages: list[ChatMessage],
         streaming_callback: Optional[StreamingCallbackT] = None,
         *,
+        generation_kwargs: Optional[dict[str, Any]] = None,
         break_point: Optional[AgentBreakpoint] = None,
         snapshot: Optional[AgentSnapshot] = None,  # type: ignore[override]
         system_prompt: Optional[str] = None,
@@ -259,6 +291,8 @@ class Agent(HaystackAgent):
         :param messages: List of Haystack ChatMessage objects to process.
         :param streaming_callback: A callback that will be invoked when a response is streamed from the LLM.
             The same callback can be configured to emit tool results when a tool is called.
+        :param generation_kwargs: Additional keyword arguments for LLM. These parameters will
+            override the parameters passed during component initialization.
         :param break_point: An AgentBreakpoint, can be a Breakpoint for the "chat_generator" or a ToolBreakpoint
             for "tool_invoker".
         :param snapshot: A dictionary containing a snapshot of a previously saved agent execution. The snapshot contains
@@ -285,11 +319,21 @@ class Agent(HaystackAgent):
             "snapshot": snapshot,
             **kwargs,
         }
-        self._runtime_checks(break_point=break_point, snapshot=snapshot)
+        # The PR https://github.com/deepset-ai/haystack/pull/9987 removed the unused snapshot parameter from
+        # _runtime_checks. This change will be released in Haystack 2.20.0.
+        # To maintain compatibility with Haystack 2.19 we check the number of parameters and call accordingly.
+        if len(inspect.signature(self._runtime_checks).parameters) == 2:
+            self._runtime_checks(break_point, snapshot)  # type: ignore[call-arg]  # pylint: disable=too-many-function-args
+        else:
+            self._runtime_checks(break_point)  # type: ignore[call-arg]  # pylint: disable=no-value-for-parameter
 
         if snapshot:
             exe_context = self._initialize_from_snapshot(
-                snapshot=snapshot, streaming_callback=streaming_callback, requires_async=False, tools=tools
+                snapshot=snapshot,
+                streaming_callback=streaming_callback,
+                requires_async=False,
+                generation_kwargs=generation_kwargs,
+                tools=tools,
             )
         else:
             exe_context = self._initialize_fresh_execution(
@@ -297,6 +341,7 @@ class Agent(HaystackAgent):
                 streaming_callback=streaming_callback,
                 requires_async=False,
                 system_prompt=system_prompt,
+                generation_kwargs=generation_kwargs,
                 tools=tools,
                 **kwargs,
             )
@@ -431,6 +476,7 @@ class Agent(HaystackAgent):
         messages: list[ChatMessage],
         streaming_callback: Optional[StreamingCallbackT] = None,
         *,
+        generation_kwargs: Optional[dict[str, Any]] = None,
         break_point: Optional[AgentBreakpoint] = None,
         snapshot: Optional[AgentSnapshot] = None,  # type: ignore[override]
         system_prompt: Optional[str] = None,
@@ -447,6 +493,8 @@ class Agent(HaystackAgent):
         :param messages: List of Haystack ChatMessage objects to process.
         :param streaming_callback: An asynchronous callback that will be invoked when a response is streamed from the
             LLM. The same callback can be configured to emit tool results when a tool is called.
+        :param generation_kwargs: Additional keyword arguments for LLM. These parameters will
+            override the parameters passed during component initialization.
         :param break_point: An AgentBreakpoint, can be a Breakpoint for the "chat_generator" or a ToolBreakpoint
             for "tool_invoker".
         :param snapshot: A dictionary containing a snapshot of a previously saved agent execution. The snapshot contains
@@ -472,11 +520,21 @@ class Agent(HaystackAgent):
             "snapshot": snapshot,
             **kwargs,
         }
-        self._runtime_checks(break_point=break_point, snapshot=snapshot)
+        # The PR https://github.com/deepset-ai/haystack/pull/9987 removed the unused snapshot parameter from
+        # _runtime_checks. This change will be released in Haystack 2.20.0.
+        # To maintain compatibility with Haystack 2.19 we check the number of parameters and call accordingly.
+        if len(inspect.signature(self._runtime_checks).parameters) == 2:
+            self._runtime_checks(break_point, snapshot)  # type: ignore[call-arg]  # pylint: disable=too-many-function-args
+        else:
+            self._runtime_checks(break_point)  # type: ignore[call-arg]  # pylint: disable=no-value-for-parameter
 
         if snapshot:
             exe_context = self._initialize_from_snapshot(
-                snapshot=snapshot, streaming_callback=streaming_callback, requires_async=True, tools=tools
+                snapshot=snapshot,
+                streaming_callback=streaming_callback,
+                requires_async=True,
+                generation_kwargs=generation_kwargs,
+                tools=tools,
             )
         else:
             exe_context = self._initialize_fresh_execution(
@@ -484,6 +542,7 @@ class Agent(HaystackAgent):
                 streaming_callback=streaming_callback,
                 requires_async=True,
                 system_prompt=system_prompt,
+                generation_kwargs=generation_kwargs,
                 tools=tools,
                 **kwargs,
             )

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/preprocessors/__init__.py

@@ -9,10 +9,12 @@ from lazy_imports import LazyImporter
 
 _import_structure = {
     "embedding_based_document_splitter": ["EmbeddingBasedDocumentSplitter"],
+    "md_header_level_inferrer": ["MarkdownHeaderLevelInferrer"],
 }
 
 if TYPE_CHECKING:
     from .embedding_based_document_splitter import EmbeddingBasedDocumentSplitter
+    from .md_header_level_inferrer import MarkdownHeaderLevelInferrer
 
 else:
     sys.modules[__name__] = LazyImporter(name=__name__, module_file=__file__, import_structure=_import_structure)

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/preprocessors/md_header_level_inferrer.py

@@ -24,7 +24,7 @@ class MarkdownHeaderLevelInferrer:
     from haystack_experimental.components.preprocessors import MarkdownHeaderLevelInferrer
 
     # Create a document with uniform header levels
-    text = "## Title\nSome content\n## Section\nMore content\n## Subsection\nFinal content"
+    text = "## Title\n## Subheader\nSection\n## Subheader\nMore Content"
     doc = Document(content=text)
 
     # Initialize the inferrer and process the document
@@ -33,7 +33,7 @@ class MarkdownHeaderLevelInferrer:
 
     # The headers are now normalized with proper hierarchy
     print(result["documents"][0].content)
-    > # Title\nSome content\n## Section\nMore content\n### Subsection\nFinal content
+    > # Title\n## Subheader\nSection\n## Subheader\nMore Content
     ```
     """
 

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/query/query_expander.py

@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import json
-from typing import Any, Dict, List, Optional
+from typing import Any, Optional
 
 from haystack import default_from_dict, default_to_dict, logging
 from haystack.components.builders.prompt_builder import PromptBuilder
@@ -91,7 +91,7 @@ class QueryExpander:
         prompt_template: Optional[str] = None,
         n_expansions: int = 4,
         include_original_query: bool = True,
-    ):
+    ) -> None:
         """
         Initialize the QueryExpander component.
 
@@ -99,7 +99,7 @@ class QueryExpander:
             If None, a default OpenAIChatGenerator with gpt-4.1-mini model is used.
         :param prompt_template: Custom [PromptBuilder](https://docs.haystack.deepset.ai/docs/promptbuilder)
             template for query expansion. The template should instruct the LLM to return a JSON response with the
-            structure: {"queries": ["query1", "query2", "query3"]}. The template should include 'query' and
+            structure: `{"queries": ["query1", "query2", "query3"]}`. The template should include 'query' and
             'n_expansions' variables.
         :param n_expansions: Number of alternative queries to generate (default: 4).
         :param include_original_query: Whether to include the original query in the output.
@@ -134,7 +134,6 @@ class QueryExpander:
             self.chat_generator = chat_generator
 
         self._is_warmed_up = False
-        self._supports_warm_up = hasattr(self.chat_generator, "warm_up")
         self.prompt_template = prompt_template or DEFAULT_PROMPT_TEMPLATE
 
         # Check if required variables are present in the template
@@ -153,7 +152,7 @@ class QueryExpander:
             required_variables=["n_expansions", "query"],
         )
 
-    def to_dict(self) -> Dict[str, Any]:
+    def to_dict(self) -> dict[str, Any]:
         """
         Serializes the component to a dictionary.
 
@@ -168,7 +167,7 @@ class QueryExpander:
         )
 
     @classmethod
-    def from_dict(cls, data: Dict[str, Any]) -> "QueryExpander":
+    def from_dict(cls, data: dict[str, Any]) -> "QueryExpander":
         """
         Deserializes the component from a dictionary.
 
@@ -181,12 +180,8 @@ class QueryExpander:
 
         return default_from_dict(cls, data)
 
-    @component.output_types(queries=List[str])
-    def run(
-        self,
-        query: str,
-        n_expansions: Optional[int] = None,
-    ) -> Dict[str, List[str]]:
+    @component.output_types(queries=list[str])
+    def run(self, query: str, n_expansions: Optional[int] = None) -> dict[str, list[str]]:
         """
         Expand the input query into multiple semantically similar queries.
 
@@ -199,11 +194,10 @@ class QueryExpander:
             If include_original_query=True, the original query will be included in addition
             to the n_expansions alternative queries.
         :raises ValueError: If n_expansions is not positive (less than or equal to 0).
-        :raises RuntimeError: If the component is not warmed up and the chat generator does not support warm up.
         """
 
-        if not self._is_warmed_up and self._supports_warm_up:
-            raise RuntimeError("The component is not warmed up. Please call the `warm_up` method first.")
+        if not self._is_warmed_up:
+            self.warm_up()
 
         response = {"queries": [query] if self.include_original_query else []}
 
@@ -252,14 +246,15 @@ class QueryExpander:
 
     def warm_up(self):
         """
-        Warm up the underlying LLM if it supports it.
+        Warm up the LLM provider component.
         """
-        if not self._is_warmed_up and self._supports_warm_up:
-            self.chat_generator.warm_up()  # type: ignore[attr-defined]
+        if not self._is_warmed_up:
+            if hasattr(self.chat_generator, "warm_up"):
+                self.chat_generator.warm_up()
             self._is_warmed_up = True
 
     @staticmethod
-    def _parse_expanded_queries(generator_response: str) -> List[str]:
+    def _parse_expanded_queries(generator_response: str) -> list[str]:
         """
         Parse the generator response to extract individual expanded queries.
 

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/retrievers/multi_query_embedding_retriever.py

@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 from concurrent.futures import ThreadPoolExecutor
-from typing import Any, List, Optional
+from typing import Any, Optional
 
 from haystack import Document, component, default_from_dict, default_to_dict
 from haystack.components.embedders.types.protocol import TextEmbedder
@@ -74,13 +74,7 @@ class MultiQueryEmbeddingRetriever:
     ```
     """  # noqa E501
 
-    def __init__(
-        self,
-        *,
-        retriever: EmbeddingRetriever,
-        query_embedder: TextEmbedder,
-        max_workers: int = 3,
-    ):
+    def __init__(self, *, retriever: EmbeddingRetriever, query_embedder: TextEmbedder, max_workers: int = 3) -> None:
         """
         Initialize MultiQueryEmbeddingRetriever.
 
@@ -104,12 +98,8 @@ class MultiQueryEmbeddingRetriever:
                 self.retriever.warm_up()
             self._is_warmed_up = True
 
-    @component.output_types(documents=List[Document])
-    def run(
-        self,
-        queries: List[str],
-        retriever_kwargs: Optional[dict[str, Any]] = None,
-    ) -> dict[str, Any]:
+    @component.output_types(documents=list[Document])
+    def run(self, queries: list[str], retriever_kwargs: Optional[dict[str, Any]] = None) -> dict[str, list[Document]]:
         """
         Retrieve documents using multiple queries in parallel.
 
@@ -123,6 +113,9 @@ class MultiQueryEmbeddingRetriever:
         seen_contents = set()
         retriever_kwargs = retriever_kwargs or {}
 
+        if not self._is_warmed_up:
+            self.warm_up()
+
         with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
             queries_results = executor.map(lambda query: self._run_on_thread(query, retriever_kwargs), queries)
             for result in queries_results:
@@ -137,7 +130,7 @@ class MultiQueryEmbeddingRetriever:
         docs.sort(key=lambda x: x.score or 0.0, reverse=True)
         return {"documents": docs}
 
-    def _run_on_thread(self, query: str, retriever_kwargs: Optional[dict[str, Any]] = None) -> Optional[List[Document]]:
+    def _run_on_thread(self, query: str, retriever_kwargs: Optional[dict[str, Any]] = None) -> Optional[list[Document]]:
         """
         Process a single query on a separate thread.
 

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/components/retrievers/multi_query_text_retriever.py

@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 from concurrent.futures import ThreadPoolExecutor
-from typing import Any, List, Optional
+from typing import Any, Optional
 
 from haystack import Document, component, default_from_dict, default_to_dict
 from haystack.core.serialization import component_to_dict
@@ -57,11 +57,7 @@ class MultiQueryTextRetriever:
     ```
     """  # noqa E501
 
-    def __init__(
-        self,
-        retriever: TextRetriever,
-        max_workers: int = 3,
-    ):
+    def __init__(self, *, retriever: TextRetriever, max_workers: int = 3) -> None:
         """
         Initialize MultiQueryTextRetriever.
 
@@ -82,11 +78,7 @@ class MultiQueryTextRetriever:
             self._is_warmed_up = True
 
     @component.output_types(documents=list[Document])
-    def run(
-        self,
-        queries: List[str],
-        retriever_kwargs: Optional[dict[str, Any]] = None,
-    ) -> dict[str, Any]:
+    def run(self, queries: list[str], retriever_kwargs: Optional[dict[str, Any]] = None) -> dict[str, list[Document]]:
         """
         Retrieve documents using multiple queries in parallel.
 

{haystack_experimental-0.14.1 → haystack_experimental-0.14.3}/haystack_experimental/core/pipeline/breakpoint.py

@@ -2,12 +2,12 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-from copy import deepcopy
 from dataclasses import replace
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Optional
 
 from haystack import logging
+from haystack.core.pipeline.utils import _deepcopy_with_exceptions
 from haystack.dataclasses.breakpoints import AgentBreakpoint, PipelineSnapshot, PipelineState, ToolBreakpoint
 from haystack.utils.base_serialization import _serialize_value_with_schema
 from haystack.utils.misc import _get_output_dir
@@ -44,8 +44,10 @@ def _create_agent_snapshot(
     """
     return AgentSnapshot(
         component_inputs={
-            "chat_generator": _serialize_value_with_schema(deepcopy(component_inputs["chat_generator"])),
-            "tool_invoker": _serialize_value_with_schema(deepcopy(component_inputs["tool_invoker"])),
+            "chat_generator": _serialize_value_with_schema(
+                _deepcopy_with_exceptions(component_inputs["chat_generator"])
+            ),
+            "tool_invoker": _serialize_value_with_schema(_deepcopy_with_exceptions(component_inputs["tool_invoker"])),
         },
         component_visits=component_visits,
         break_point=agent_breakpoint,