[Refactor] Simplify dummy data generation (#35025)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -266,11 +266,14 @@ class InputProcessingContext:
|
||||
if isinstance(tokenizer, MistralTokenizer):
|
||||
tokenizer = tokenizer.transformers_tokenizer
|
||||
|
||||
merged_kwargs = self.get_merged_mm_kwargs(kwargs)
|
||||
merged_kwargs.pop("tokenizer", None)
|
||||
|
||||
return cached_processor_from_config(
|
||||
self.model_config,
|
||||
processor_cls=typ,
|
||||
tokenizer=tokenizer,
|
||||
**kwargs,
|
||||
**merged_kwargs,
|
||||
)
|
||||
|
||||
def init_processor(
|
||||
@@ -283,12 +286,7 @@ class InputProcessingContext:
|
||||
Initialize a HuggingFace-like processor class, merging the
|
||||
keyword arguments with those in the model's configuration.
|
||||
"""
|
||||
mm_config = self.model_config.get_multimodal_config()
|
||||
base_kwargs = mm_config.mm_processor_kwargs
|
||||
if base_kwargs is None:
|
||||
base_kwargs = {}
|
||||
|
||||
merged_kwargs = {**base_kwargs, **kwargs}
|
||||
merged_kwargs = self.get_merged_mm_kwargs(kwargs)
|
||||
|
||||
return typ(**merged_kwargs)
|
||||
|
||||
|
||||
@@ -62,8 +62,7 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
|
||||
self,
|
||||
seq_len: int,
|
||||
mm_counts: Mapping[str, int],
|
||||
mm_options: Mapping[str, BaseDummyOptions] | None = None,
|
||||
mm_processor_kwargs: Mapping[str, object] | None = None,
|
||||
mm_options: Mapping[str, BaseDummyOptions],
|
||||
) -> MultiModalDataDict:
|
||||
"""
|
||||
Build the multimodal input which, after processing, results in
|
||||
@@ -83,8 +82,7 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
|
||||
self,
|
||||
seq_len: int,
|
||||
mm_counts: Mapping[str, int],
|
||||
mm_options: Mapping[str, BaseDummyOptions] | None = None,
|
||||
mm_processor_kwargs: Mapping[str, object] | None = None,
|
||||
mm_options: Mapping[str, BaseDummyOptions],
|
||||
) -> ProcessorInputs:
|
||||
"""
|
||||
Build the input which, after processing, results in
|
||||
@@ -94,16 +92,9 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
|
||||
seq_len: Sequence length
|
||||
mm_counts: Count of items per modality
|
||||
mm_options: Configurable options per modality (optional)
|
||||
mm_processor_kwargs: Additional keyword arguments
|
||||
for hf_processor (optional)
|
||||
"""
|
||||
dummy_text = self.get_dummy_text(mm_counts)
|
||||
dummy_mm_data = self.get_dummy_mm_data(
|
||||
seq_len,
|
||||
mm_counts,
|
||||
mm_options,
|
||||
mm_processor_kwargs=mm_processor_kwargs,
|
||||
)
|
||||
dummy_mm_data = self.get_dummy_mm_data(seq_len, mm_counts, mm_options)
|
||||
dummy_mm_items = self.info.parse_mm_data(dummy_mm_data, validate=False)
|
||||
|
||||
tokenization_kwargs = {"truncation": False}
|
||||
@@ -111,7 +102,6 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
|
||||
return ProcessorInputs(
|
||||
prompt=dummy_text,
|
||||
mm_items=dummy_mm_items,
|
||||
hf_processor_mm_kwargs=mm_processor_kwargs or {},
|
||||
tokenization_kwargs=tokenization_kwargs,
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user