chore: fix mypy issues due to transformers 4.51.1 (#9198)

Author: anakin87

haystack/components/embedders/hugging_face_api_document_embedder.py

@@ -249,7 +249,7 @@ def _embed_batch(self, texts_to_embed: List[str], batch_size: int) -> List[List[
                 logger.warning(msg)
                 normalize = None
 
-        all_embeddings = []
+        all_embeddings: List = []
         for i in tqdm(
             range(0, len(texts_to_embed), batch_size), disable=not self.progress_bar, desc="Calculating embeddings"
         ):

haystack/components/generators/chat/hugging_face_local.py

@@ -7,7 +7,7 @@
 import re
 import sys
 from concurrent.futures import ThreadPoolExecutor
-from typing import Any, Callable, Dict, List, Literal, Optional, Union
+from typing import Any, Callable, Dict, List, Literal, Optional, Union, cast
 
 from haystack import component, default_from_dict, default_to_dict, logging
 from haystack.dataclasses import ChatMessage, StreamingChunk, ToolCall, select_streaming_callback
@@ -31,7 +31,9 @@
 
 with LazyImport(message="Run 'pip install \"transformers[torch]\"'") as torch_and_transformers_import:
     from huggingface_hub import model_info
-    from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast, StoppingCriteriaList, pipeline
+    from transformers import StoppingCriteriaList, pipeline
+    from transformers.tokenization_utils import PreTrainedTokenizer
+    from transformers.tokenization_utils_fast import PreTrainedTokenizerFast
 
     from haystack.utils.hf import (  # pylint: disable=ungrouped-imports
         HFTokenStreamingHandler,
@@ -555,6 +557,9 @@ async def _run_streaming_async(  # pylint: disable=too-many-positional-arguments
             hf_messages, tokenize=False, chat_template=self.chat_template, add_generation_prompt=True
         )
 
+        # prepared_prompt is a string, but transformers has some type issues
+        prepared_prompt = cast(str, prepared_prompt)
+
         # Avoid some unnecessary warnings in the generation pipeline call
         generation_kwargs["pad_token_id"] = (
             generation_kwargs.get("pad_token_id", tokenizer.pad_token_id) or tokenizer.eos_token_id
@@ -607,6 +612,9 @@ async def _run_non_streaming_async(  # pylint: disable=too-many-positional-argum
             tools=[tc.tool_spec for tc in tools] if tools else None,
         )
 
+        # prepared_prompt is a string, but transformers has some type issues
+        prepared_prompt = cast(str, prepared_prompt)
+
         # Avoid some unnecessary warnings in the generation pipeline call
         generation_kwargs["pad_token_id"] = (
             generation_kwargs.get("pad_token_id", tokenizer.pad_token_id) or tokenizer.eos_token_id

haystack/utils/hf.py

@@ -283,7 +283,9 @@ def convert_message_to_hf_format(message: ChatMessage) -> Dict[str, Any]:
 
 
 with LazyImport(message="Run 'pip install \"transformers[torch]\"'") as transformers_import:
-    from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast, StoppingCriteria, TextStreamer
+    from transformers import StoppingCriteria, TextStreamer
+    from transformers.tokenization_utils import PreTrainedTokenizer
+    from transformers.tokenization_utils_fast import PreTrainedTokenizerFast
 
     torch_import.check()
     transformers_import.check()