SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[vlm] adjust max length for special tokens (#37342)

Browse Source 
* update

* apply suggestion

* fix tests for main branch

* remove unused logger

* add special tokens in tests

* nit

* fix more tests

* fix test

* pg also
This commit is contained in:
Raushan Turganbay
2025-04-16 20:49:20 +02:00
committed by GitHub
parent c94c59fc47
commit 32eca7197a
39 changed files with 414 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/transformers/models/aria/modular_aria.py
View File

@@ -946,6 +946,8 @@ class AriaProcessor(ProcessorMixin):
size_conversion = {490: 128, 980: 256}
self.size_conversion = {int(k): v for k, v in size_conversion.items()}
self.image_token = tokenizer.image_token
self.image_token_id = tokenizer.image_token_id
if tokenizer is not None and tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.unk_token
@@ -986,10 +988,12 @@ class AriaProcessor(ProcessorMixin):
tokenizer_init_kwargs=self.tokenizer.init_kwargs,
**kwargs,
)
if isinstance(text, str):
text = [text]
elif not isinstance(text, list) and not isinstance(text[0], str):
raise ValueError("Invalid input text. Please provide a string, or a list of strings")
if images is not None:
image_inputs = self.image_processor(
images,
@@ -1007,12 +1011,11 @@ class AriaProcessor(ProcessorMixin):
image_inputs = {}
prompt_strings = text
text_inputs = self.tokenizer(
prompt_strings,
**output_kwargs["text_kwargs"],
)
return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None)
text_inputs = self.tokenizer(prompt_strings, **output_kwargs["text_kwargs"])
self._check_special_mm_tokens(prompt_strings, text_inputs, modalities=["image"])
return BatchFeature(data={**text_inputs, **image_inputs})
return BatchFeature(data={**text_inputs, **image_inputs}, tensor_type=return_tensors)
def batch_decode(self, *args, **kwargs):
"""

13
src/transformers/models/aria/processing_aria.py
View File

@@ -72,6 +72,8 @@ class AriaProcessor(ProcessorMixin):
size_conversion = {490: 128, 980: 256}
self.size_conversion = {int(k): v for k, v in size_conversion.items()}
self.image_token = tokenizer.image_token
self.image_token_id = tokenizer.image_token_id
if tokenizer is not None and tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.unk_token
@@ -112,10 +114,12 @@ class AriaProcessor(ProcessorMixin):
tokenizer_init_kwargs=self.tokenizer.init_kwargs,
**kwargs,
)
if isinstance(text, str):
text = [text]
elif not isinstance(text, list) and not isinstance(text[0], str):
raise ValueError("Invalid input text. Please provide a string, or a list of strings")
if images is not None:
image_inputs = self.image_processor(
images,
@@ -133,12 +137,11 @@ class AriaProcessor(ProcessorMixin):
image_inputs = {}
prompt_strings = text
text_inputs = self.tokenizer(
prompt_strings,
**output_kwargs["text_kwargs"],
)
return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None)
text_inputs = self.tokenizer(prompt_strings, **output_kwargs["text_kwargs"])
self._check_special_mm_tokens(prompt_strings, text_inputs, modalities=["image"])
return BatchFeature(data={**text_inputs, **image_inputs})
return BatchFeature(data={**text_inputs, **image_inputs}, tensor_type=return_tensors)
def batch_decode(self, *args, **kwargs):
"""