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

[CI] doc builder without custom image (#36862)

Browse Source 
* no image

* test

* revert jax version updates

* make fixup

* update autodoc path for model_addition_debugger

* shieldgemma2

* add missing pages to toctree
This commit is contained in:
Joao Gante
2025-03-21 09:10:27 +00:00
committed by GitHub
parent 97d2f9d8ae
commit 949cca4061
6 changed files with 29 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.github/workflows/build_pr_documentation.yml vendored
View File

@@ -15,4 +15,3 @@ jobs:
pr_number: ${{ github.event.number }}
package: transformers
languages: ar de en es fr hi it ko pt tr zh ja te
custom_container: huggingface/transformers-doc-builder

4
docs/source/en/_toctree.yml
View File

@@ -985,6 +985,8 @@
title: Qwen2VL
- local: model_doc/sam
title: Segment Anything
- local: model_doc/shieldgemma2
title: ShieldGemma2
- local: model_doc/siglip
title: SigLIP
- local: model_doc/siglip2
@@ -1044,6 +1046,8 @@
- sections:
- local: internal/modeling_utils
title: Custom Layers and Utilities
- local: internal/model_debugging_utils
title: Utilities for Model Debugging
- local: internal/pipelines_utils
title: Utilities for pipelines
- local: internal/tokenization_utils

4
docs/source/en/internal/model_debugging_utils.md
View File

@@ -66,6 +66,6 @@ with model_addition_debugger_context(model, "optional_path_to_your_output_file.j
```
[[autodoc]] utils.model_addition_debugger
[[autodoc]] model_addition_debugger
[[autodoc]] utils.model_addition_debugger_context
[[autodoc]] model_addition_debugger_context

4
setup.py
View File

@@ -121,8 +121,8 @@ _deps = [
"importlib_metadata",
"ipadic>=1.0.0,<2.0",
"isort>=5.5.4",
"jax>=0.4.27,<=0.4.38",
"jaxlib>=0.4.27,<=0.4.38",
"jax>=0.4.1,<=0.4.13",
"jaxlib>=0.4.1,<=0.4.13",
"jieba",
"jinja2>=3.1.0",
"kenlm",

4
src/transformers/dependency_versions_table.py
View File

@@ -28,8 +28,8 @@ deps = {
"importlib_metadata": "importlib_metadata",
"ipadic": "ipadic>=1.0.0,<2.0",
"isort": "isort>=5.5.4",
"jax": "jax>=0.4.27,<=0.4.38",
"jaxlib": "jaxlib>=0.4.27,<=0.4.38",
"jax": "jax>=0.4.1,<=0.4.13",
"jaxlib": "jaxlib>=0.4.1,<=0.4.13",
"jieba": "jieba",
"jinja2": "jinja2>=3.1.0",
"kenlm": "kenlm",

42
src/transformers/models/shieldgemma2/modeling_shieldgemma2.py
View File

@@ -25,7 +25,6 @@ from ...modeling_utils import PreTrainedModel
from ...utils import (
add_start_docstrings_to_model_forward,
logging,
replace_return_docstrings,
)
from ...utils.deprecation import deprecate_kwarg
from ..auto import AutoModelForImageTextToText
@@ -109,25 +108,6 @@ SHIELDGEMMA2_INPUTS_DOCSTRING = r"""
Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`,
this tensor is not affected by padding. It is used to update the cache in the correct position and to infer
the complete sequence length.
Returns:
A `ShieldGemma2ImageClassifierOutputWithNoAttention` instance continaing the logits and probabilities
associated with the model predicting the `Yes` or `No` token as the response to that prompt, captured in the
following properties.
* `logits` (`torch.Tensor` of shape `(batch_size, 2)`):
The first position along dim=1 is the logits for the `Yes` token and the second position along dim=1 is
the logits for the `No` token.
* `probabilities` (`torch.Tensor` of shape `(batch_size, 2)`):
The first position along dim=1 is the probability of predicting the `Yes` token and the second position
along dim=1 is the probability of predicting the `No` token.
ShieldGemma prompts are constructed such that predicting the `Yes` token means the content *does violate* the
policy as described. If you are only interested in the violative condition, use
`violated = outputs.probabilities[:, 1]` to extract that slice from the output tensors.
When used with the `ShieldGemma2Processor`, the `batch_size` will be equal to `len(images) * len(policies)`,
and the order within the batch will be img1_policy1, ... img1_policyN, ... imgM_policyN.
"""
@@ -172,9 +152,6 @@ class ShieldGemma2ForImageClassification(PreTrainedModel):
@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
@add_start_docstrings_to_model_forward(SHIELDGEMMA2_INPUTS_DOCSTRING)
@replace_return_docstrings(
output_type=ShieldGemma2ImageClassifierOutputWithNoAttention, config_class=_CONFIG_FOR_DOC
)
def forward(
self,
input_ids: torch.LongTensor = None,
@@ -193,9 +170,26 @@ class ShieldGemma2ForImageClassification(PreTrainedModel):
logits_to_keep: Union[int, torch.Tensor] = 0,
**lm_kwargs,
) -> ShieldGemma2ImageClassifierOutputWithNoAttention:
"""Predicts the binary probability that the image violates the speicfied policy.
"""Predicts the binary probability that the image violates the specified policy.
Returns:
A `ShieldGemma2ImageClassifierOutputWithNoAttention` instance containing the logits and probabilities
associated with the model predicting the `Yes` or `No` token as the response to that prompt, captured in the
following properties.
* `logits` (`torch.Tensor` of shape `(batch_size, 2)`):
The first position along dim=1 is the logits for the `Yes` token and the second position along dim=1 is
the logits for the `No` token.
* `probabilities` (`torch.Tensor` of shape `(batch_size, 2)`):
The first position along dim=1 is the probability of predicting the `Yes` token and the second position
along dim=1 is the probability of predicting the `No` token.
ShieldGemma prompts are constructed such that predicting the `Yes` token means the content *does violate* the
policy as described. If you are only interested in the violative condition, use
`violated = outputs.probabilities[:, 1]` to extract that slice from the output tensors.
When used with the `ShieldGemma2Processor`, the `batch_size` will be equal to `len(images) * len(policies)`,
and the order within the batch will be img1_policy1, ... img1_policyN, ... imgM_policyN.
"""
outputs = self.model(
input_ids=input_ids,