[Bugfix] Fix profiling dummy data for Pixtral (#18677)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-05-25 14:05:30 +00:00
committed by GitHub
parent 3a886bd58c
commit 57fd13a707
5 changed files with 151 additions and 168 deletions
+10 -5
View File
@@ -3,7 +3,7 @@
from abc import ABC
from collections.abc import Mapping
from dataclasses import dataclass, field
from typing import Generic, NamedTuple, Optional, TypeVar, cast
from typing import Generic, NamedTuple, Optional, TypeVar, Union, cast
import numpy as np
import numpy.typing as npt
@@ -27,7 +27,7 @@ class ProcessorInputs:
Represents the keyword arguments to
{meth}`vllm.multimodal.processing.BaseMultiModalProcessor.apply`.
"""
prompt_text: str
prompt: Union[str, list[int]]
mm_data: MultiModalDataDict
hf_processor_mm_kwargs: Mapping[str, object] = field(default_factory=dict)
@@ -75,7 +75,12 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
"in an upcoming release.")
seq_len = self.info.ctx.model_config.max_model_len
return self.get_dummy_processor_inputs(seq_len, mm_counts).prompt_text
prompt = self.get_dummy_processor_inputs(seq_len, mm_counts).prompt
if not isinstance(prompt, str):
prompt = self.info.get_tokenizer().decode(prompt)
return prompt
# TODO: @abstractmethod after transition
def get_dummy_mm_data(
@@ -101,7 +106,7 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
dummy_text = self.get_dummy_text(mm_counts)
dummy_mm_data = self.get_dummy_mm_data(seq_len, mm_counts)
return ProcessorInputs(prompt_text=dummy_text, mm_data=dummy_mm_data)
return ProcessorInputs(prompt=dummy_text, mm_data=dummy_mm_data)
def _get_dummy_audios(
self,
@@ -177,7 +182,7 @@ class MultiModalProfiler(Generic[_I]):
seq_len, mm_counts)
return self.processor.apply(
prompt=processor_inputs.prompt_text,
prompt=processor_inputs.prompt,
mm_data=processor_inputs.mm_data,
hf_processor_mm_kwargs=processor_inputs.hf_processor_mm_kwargs,
)