[Misc] Break chat_utils into more maintainable modules. (#60972)
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
@@ -511,7 +511,7 @@ class TestConvertToolMessage:
|
||||
# A list of text blocks maps 1-to-1 onto ``text`` content
|
||||
# parts, mirroring the shape ``_convert_user_message`` emits
|
||||
# for multi-part user content. vLLM's
|
||||
# ``_parse_chat_message_content`` will collapse this back to
|
||||
# ``parse_chat_message_content`` will collapse this back to
|
||||
# ``"\n".join(...)`` for the ``ConversationMessage`` handed to
|
||||
# renderers (see the tool-role branch in
|
||||
# ``vllm/entrypoints/chat_utils.py``), so downstream chat
|
||||
@@ -533,7 +533,7 @@ class TestConvertToolMessage:
|
||||
"""Document blocks must not survive on the OpenAI-shaped tool
|
||||
message.
|
||||
|
||||
vLLM's shared ``_parse_chat_message_content_parts`` runs a
|
||||
vLLM's shared ``parse_chat_message_content_parts`` runs a
|
||||
Pydantic ``ValidatorIterator`` over the OpenAI content-part
|
||||
union, which only knows ``text`` / ``image_url`` / etc. and
|
||||
rejects ``{"type": "document", ...}`` blocks outright. We
|
||||
@@ -588,7 +588,7 @@ class TestConvertToolMessage:
|
||||
def test_converted_tool_message_content_survives_openai_validation(self):
|
||||
"""Regression for the reported crash.
|
||||
|
||||
``vllm/entrypoints/chat_utils.py:_parse_chat_message_content_parts``
|
||||
``vllm/renders/chat_message_content.py:parse_chat_message_content_parts``
|
||||
runs a Pydantic ``ValidatorIterator`` over
|
||||
``list[ChatCompletionContentPartParam]``, which is a
|
||||
discriminated union that only accepts ``text`` / ``image_url``
|
||||
|
||||
@@ -6,7 +6,7 @@ from unittest.mock import MagicMock
|
||||
|
||||
import pytest
|
||||
|
||||
from vllm.renderers.chat_utils import _parse_chat_message_content
|
||||
from vllm.renderers.chat_utils import parse_chat_message_content
|
||||
|
||||
HEADER = (
|
||||
"x-anthropic-billing-header: cc_version=2.1.220.300; "
|
||||
@@ -17,7 +17,7 @@ HEADER = (
|
||||
def _parse(message, content_format="openai"):
|
||||
tracker = MagicMock()
|
||||
tracker.create_parser.return_value.mm_placeholder_storage.return_value = {}
|
||||
return _parse_chat_message_content(
|
||||
return parse_chat_message_content(
|
||||
message, tracker, content_format, interleave_strings=False
|
||||
)
|
||||
|
||||
|
||||
@@ -29,7 +29,7 @@ from vllm.multimodal.cache import MultiModalProcessorOnlyCache
|
||||
from vllm.multimodal.inputs import MultiModalFeatureSpec, batched_tensors_equal
|
||||
from vllm.multimodal.processing import BaseMultiModalProcessor, InputProcessingContext
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.renderers.chat_utils import _get_embeds_data
|
||||
from vllm.renderers.chat_utils.item_tracker import _get_embeds_data
|
||||
from vllm.tokenizers import TokenizerLike, cached_tokenizer_from_config
|
||||
from vllm.utils.mistral import is_mistral_tokenizer
|
||||
|
||||
|
||||
@@ -16,24 +16,24 @@ from vllm.assets.video import VideoAsset
|
||||
from vllm.config import ModelConfig
|
||||
from vllm.exceptions import VLLMValidationError
|
||||
from vllm.inputs import MultiModalDataDict, MultiModalUUIDDict
|
||||
from vllm.multimodal.media import MEDIA_CONNECTOR_REGISTRY
|
||||
from vllm.multimodal.utils import (
|
||||
encode_audio_url,
|
||||
encode_image_url,
|
||||
encode_video_url,
|
||||
)
|
||||
from vllm.renderers.chat_utils import (
|
||||
MEDIA_CONNECTOR_REGISTRY,
|
||||
AsyncMultiModalItemTracker,
|
||||
ChatCompletionMessageParam,
|
||||
ConversationMessage,
|
||||
_load_embeds_dict,
|
||||
_parse_chat_message_content,
|
||||
_parse_metadata_array,
|
||||
_postprocess_messages,
|
||||
parse_chat_message_content,
|
||||
parse_chat_messages,
|
||||
parse_chat_messages_async,
|
||||
validate_chat_template,
|
||||
)
|
||||
from vllm.renderers.chat_utils.chat_message import _postprocess_messages
|
||||
from vllm.renderers.chat_utils.content_parser import _load_embeds_dict
|
||||
from vllm.renderers.chat_utils.item_tracker import _parse_metadata_array
|
||||
from vllm.utils.serial_utils import tensor2base64
|
||||
|
||||
KIMI_K2_5_MODEL_ID = "moonshotai/Kimi-K2.5"
|
||||
@@ -3205,7 +3205,7 @@ def _parse_one(message) -> ConversationMessage:
|
||||
parser = tracker.create_parser.return_value
|
||||
parser.model_config.enable_prompt_embeds = False
|
||||
parser.mm_placeholder_storage.return_value = {}
|
||||
result = _parse_chat_message_content(
|
||||
result = parse_chat_message_content(
|
||||
message, tracker, content_format="string", interleave_strings=False
|
||||
)
|
||||
assert len(result) == 1
|
||||
|
||||
@@ -18,15 +18,16 @@ from transformers import AutoTokenizer
|
||||
|
||||
from vllm.exceptions import VLLMValidationError
|
||||
from vllm.renderers import TokenizeParams
|
||||
from vllm.renderers.chat_utils import (
|
||||
_ENABLE_PROMPT_EMBEDS_ERROR,
|
||||
from vllm.renderers.chat_utils import parse_chat_messages, parse_chat_messages_async
|
||||
from vllm.renderers.chat_utils.chat_message import (
|
||||
_PROMPT_EMBEDS_MISSING_DATA_ERROR,
|
||||
_RESERVED_PLACEHOLDER_IN_TEXT_ERROR,
|
||||
MM_PARSER_MAP,
|
||||
)
|
||||
from vllm.renderers.chat_utils.content_parser import (
|
||||
_ENABLE_PROMPT_EMBEDS_ERROR,
|
||||
MODALITY_PLACEHOLDERS_MAP,
|
||||
PROMPT_EMBEDS_PLACEHOLDER_TOKEN,
|
||||
parse_chat_messages,
|
||||
parse_chat_messages_async,
|
||||
)
|
||||
from vllm.renderers.hf import (
|
||||
_PROMPT_EMBEDS_PLACEHOLDER_SPAN_MISMATCH_ERROR,
|
||||
|
||||
@@ -28,9 +28,9 @@ from unittest.mock import MagicMock, patch
|
||||
|
||||
import pytest
|
||||
|
||||
from vllm.exceptions import ChatTemplateResolutionError
|
||||
from vllm.multimodal.processing import BaseMultiModalProcessor
|
||||
from vllm.renderers.base import BaseRenderer
|
||||
from vllm.renderers.chat_utils import ChatTemplateResolutionError
|
||||
from vllm.renderers.params import ChatParams
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user