[Renderer] Move InputPreprocessor into Renderer (1/2) (#34510)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
@@ -54,6 +54,7 @@ class MockModelConfig:
|
||||
media_io_kwargs: dict[str, dict[str, Any]] = field(default_factory=dict)
|
||||
skip_tokenizer_init = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
def get_diff_sampling_param(self):
|
||||
return self.diff_sampling_param or {}
|
||||
@@ -67,7 +68,7 @@ class MockVllmConfig:
|
||||
def _build_renderer(model_config: MockModelConfig):
|
||||
_, tokenizer_name, _, kwargs = tokenizer_args_from_config(model_config)
|
||||
|
||||
return HfRenderer(
|
||||
return HfRenderer.from_config(
|
||||
MockVllmConfig(model_config),
|
||||
tokenizer_kwargs={**kwargs, "tokenizer_name": tokenizer_name},
|
||||
)
|
||||
|
||||
@@ -53,6 +53,7 @@ class MockModelConfig:
|
||||
media_io_kwargs: dict[str, dict[str, Any]] = field(default_factory=dict)
|
||||
skip_tokenizer_init = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
def get_diff_sampling_param(self):
|
||||
return self.diff_sampling_param or {}
|
||||
@@ -78,7 +79,7 @@ def _build_serving_completion(engine: AsyncLLM) -> OpenAIServingCompletion:
|
||||
def _build_renderer(model_config: MockModelConfig):
|
||||
_, tokenizer_name, _, kwargs = tokenizer_args_from_config(model_config)
|
||||
|
||||
return HfRenderer(
|
||||
return HfRenderer.from_config(
|
||||
MockVllmConfig(model_config),
|
||||
tokenizer_kwargs={**kwargs, "tokenizer_name": tokenizer_name},
|
||||
)
|
||||
|
||||
@@ -52,6 +52,7 @@ class MockModelConfig:
|
||||
generation_config: str = "auto"
|
||||
skip_tokenizer_init: bool = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
def get_diff_sampling_param(self):
|
||||
return self.diff_sampling_param or {}
|
||||
@@ -95,7 +96,7 @@ def register_mock_resolver():
|
||||
def _build_renderer(model_config: MockModelConfig):
|
||||
_, tokenizer_name, _, kwargs = tokenizer_args_from_config(model_config)
|
||||
|
||||
return HfRenderer(
|
||||
return HfRenderer.from_config(
|
||||
MockVllmConfig(model_config),
|
||||
tokenizer_kwargs={**kwargs, "tokenizer_name": tokenizer_name},
|
||||
)
|
||||
|
||||
@@ -529,6 +529,7 @@ class MockModelConfig:
|
||||
media_io_kwargs: dict[str, dict[str, Any]] = field(default_factory=dict)
|
||||
skip_tokenizer_init: bool = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
def get_diff_sampling_param(self):
|
||||
return self.diff_sampling_param or {}
|
||||
@@ -542,7 +543,7 @@ class MockVllmConfig:
|
||||
def _build_renderer(model_config: MockModelConfig):
|
||||
_, tokenizer_name, _, kwargs = tokenizer_args_from_config(model_config)
|
||||
|
||||
return HfRenderer(
|
||||
return HfRenderer.from_config(
|
||||
MockVllmConfig(model_config),
|
||||
tokenizer_kwargs={**kwargs, "tokenizer_name": tokenizer_name},
|
||||
)
|
||||
@@ -756,9 +757,8 @@ async def test_serving_chat_mistral_token_ids_prompt_is_validated():
|
||||
mock_tokenizer = MagicMock(spec=MistralTokenizer)
|
||||
mock_renderer = MistralRenderer(
|
||||
MockVllmConfig(mock_engine.model_config),
|
||||
tokenizer_kwargs={},
|
||||
tokenizer=mock_tokenizer,
|
||||
)
|
||||
mock_renderer._tokenizer = mock_tokenizer
|
||||
# Force the Mistral chat template renderer to return token IDs.
|
||||
# Choose a prompt length that is < max_model_len, but large enough that
|
||||
# adding max_tokens should exceed the model context window.
|
||||
@@ -798,9 +798,8 @@ async def test_serving_chat_mistral_token_ids_prompt_too_long_is_rejected():
|
||||
mock_tokenizer = MagicMock(spec=MistralTokenizer)
|
||||
mock_renderer = MistralRenderer(
|
||||
MockVllmConfig(mock_engine.model_config),
|
||||
tokenizer_kwargs={},
|
||||
tokenizer=mock_tokenizer,
|
||||
)
|
||||
mock_renderer._tokenizer = mock_tokenizer
|
||||
# prompt_token_ids length == max_model_len should be rejected for
|
||||
# completion-like requests (ChatCompletionRequest).
|
||||
mock_renderer.render_messages_async = AsyncMock(
|
||||
|
||||
@@ -38,6 +38,7 @@ class MockModelConfig:
|
||||
enable_prompt_embeds: bool = True
|
||||
skip_tokenizer_init: bool = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
|
||||
@dataclass
|
||||
@@ -78,15 +79,16 @@ def _build_renderer(
|
||||
|
||||
renderer = HfRenderer(
|
||||
MockVllmConfig(model_config),
|
||||
tokenizer_kwargs={**kwargs, "tokenizer_name": tokenizer_name},
|
||||
tokenizer=(
|
||||
None
|
||||
if model_config.skip_tokenizer_init
|
||||
else DummyTokenizer(
|
||||
truncation_side=truncation_side,
|
||||
max_chars_per_token=max_chars_per_token,
|
||||
)
|
||||
),
|
||||
)
|
||||
|
||||
if not model_config.skip_tokenizer_init:
|
||||
renderer._tokenizer = DummyTokenizer(
|
||||
truncation_side=truncation_side,
|
||||
max_chars_per_token=max_chars_per_token,
|
||||
)
|
||||
|
||||
return renderer
|
||||
|
||||
|
||||
@@ -277,7 +279,7 @@ class TestRenderPrompt:
|
||||
)
|
||||
|
||||
# Should not even attempt tokenization
|
||||
assert renderer._tokenizer._captured_encode_kwargs == {}
|
||||
assert renderer.tokenizer._captured_encode_kwargs == {}
|
||||
|
||||
def test_text_max_length_exceeded_nonobvious(self):
|
||||
renderer = _build_renderer(MockModelConfig(), max_chars_per_token=2)
|
||||
@@ -298,8 +300,8 @@ class TestRenderPrompt:
|
||||
)
|
||||
|
||||
# Should only tokenize the first max_total_tokens + 1 tokens
|
||||
assert renderer._tokenizer._captured_encode_kwargs["truncation"] is True
|
||||
assert renderer._tokenizer._captured_encode_kwargs["max_length"] == 101
|
||||
assert renderer.tokenizer._captured_encode_kwargs["truncation"] is True
|
||||
assert renderer.tokenizer._captured_encode_kwargs["max_length"] == 101
|
||||
|
||||
def test_token_max_length_exceeded(self):
|
||||
renderer = _build_renderer(MockModelConfig())
|
||||
|
||||
@@ -36,6 +36,7 @@ class MockModelConfig:
|
||||
enable_prompt_embeds: bool = True
|
||||
skip_tokenizer_init: bool = False
|
||||
is_encoder_decoder: bool = False
|
||||
is_multimodal_model: bool = False
|
||||
|
||||
|
||||
@dataclass
|
||||
@@ -57,9 +58,8 @@ async def test_async_mistral_tokenizer_does_not_block_event_loop():
|
||||
mock_tokenizer.apply_chat_template = mocked_apply_chat_template
|
||||
mock_renderer = MistralRenderer(
|
||||
MockVllmConfig(mock_model_config),
|
||||
tokenizer_kwargs={},
|
||||
tokenizer=mock_tokenizer,
|
||||
)
|
||||
mock_renderer._tokenizer = mock_tokenizer
|
||||
|
||||
task = mock_renderer.render_messages_async([], ChatParams())
|
||||
|
||||
|
||||
@@ -19,7 +19,7 @@ import pytest
|
||||
import pytest_asyncio
|
||||
|
||||
from vllm import SamplingParams
|
||||
from vllm.inputs import StreamingInput
|
||||
from vllm.engine.protocol import StreamingInput
|
||||
from vllm.outputs import RequestOutput
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sampling_params import RequestOutputKind
|
||||
|
||||
@@ -7,7 +7,7 @@ from unittest.mock import AsyncMock, MagicMock
|
||||
|
||||
import pytest
|
||||
|
||||
from vllm.inputs import StreamingInput
|
||||
from vllm.engine.protocol import StreamingInput
|
||||
from vllm.outputs import RequestOutput
|
||||
from vllm.sampling_params import RequestOutputKind, SamplingParams
|
||||
from vllm.v1.engine.async_llm import AsyncLLM
|
||||
|
||||
Reference in New Issue
Block a user