riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 33 Packages Projects Releases Wiki Activity

Add TF whisper (#19378)

Browse Source 
* simplify loop

* add featur extractor

* add model

* start conversion

* add dropout

* initial commit of test files

* copnversion for all models

* update processor for correct padding

* update feature extraction

* update integration test logits match

* fmnt: off for the logits

* on the fly mel bank

* small nit

* update test

* update tokenizer

* nit feature extraction

* update

* update tokenizer test

* adds logit processor and update tokenizer to get supress tokens

* style

* clean convert

* revert to original modeling tf utils

* Update

* update

* nit

* clean convert file

* update tests and nits

* quality

* slow generation test

* ffn_dim to allow customization

* update readme

* add to toctreee

* start fixing integration tests

* update tests and code

* fix feature extractor

* fix config tests common

* update code to fix tests

* fix feature exctractor

* nit feature extraction

* update test for new feature extractor

* style

* add absrtact

* large logits wioth custom decoder input ids

* wraap around is otrch available

* fix feature extractor

* correct logits for whisper small.en

* nit

* fix encoder_attentino_mask

* some fixes

* remove unnecessary inputs

* nits

* add normalizer file

* update etst tokenization

* fix attention mask not defined

* fix generate

* remove uncoder attention mask useless

* update test modeling whisper

* update condfig to add second non supress tokens

* nits on feature exrtactor

* nit for test tokenizers

* update etsts

* update tests

* update tokenization test

* fixup

* invalidated hf token. Clean convert openai to whisper

* fix logit tests

* fixup

* Add model to README

* Fix doc tests

* clean merge

* revert toc_tree changes

* remove useless LogitProcessor

* Update whisper .mdx

* update config file doc

* update configuration docstring

* update test tokenization

* update test tokenization

* update tokenization whisper
Added copied from where needed

* update feature extraction

* nit test name

* style

* quality

* remove get suppress tokens and update non_speech tokens global variables

* Update src/transformers/models/whisper/feature_extraction_whisper.py

Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* clean modeling whisper and test
Removed the attention mask arguments that are deprecated

* fix large test

* Add multilingual audio test, and translate test

* style

* fix larg multilingual test

* nits

* add copied from for attention layer

* remove attention masks in doc

* add english normalizer

* Update docs/source/en/model_doc/whisper.mdx

Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* update tokenization test

* remove copied from in whisper attention : no bias in k_proj only

* wrap around dependencies in english normalizer

* style

* correct import generation logits

* for now, wrap feature extractor with torch

* remove torch depencies for feature extraction and style

* Update src/transformers/models/whisper/convert_openai_whisper_to_tfms.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/models/whisper/configuration_whisper.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update docs/source/en/model_doc/whisper.mdx

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* fixup

* nit

* update logitds

* style

* nit

* nits and fix final tests

* add `is_more_itertools_available` to utils

* quality

* add begin supress tokens, supress tokens to generate args and config

* clean supressTokensLogitProcessor in generation logits

* Nit naming

* add supressTokensAtBegin

* udpate tests, supress tokens to None or correct values

* nit and style

* update RAG to fit test and generate_logit

* add copy pasted statment on english normalizer

* add arguments to config_common_kwargs

* Update src/transformers/generation_utils.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/generation_logits_process.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* revert changes based on reviews

* update doc and nits

* Update src/transformers/models/whisper/configuration_whisper.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* more nits

* last nits

* update test configuration common

* add BART name in decoder attention mask documentation

* Update src/transformers/models/whisper/modeling_whisper.py

Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* style

* nit

* nit

* add english.json file to git

* nits on documentation

* nit

* nits

* last styling

* add main toctree file

* remove sentence piece dependency

* clean init file

* fix tokenizer that has no dependencies on sentencepiece

* update whisper init file, nit

* remove english.json file

* add get decoder prompt id

* All weights loading

* Remove hanging pdb

* Fixup and tidy up

* Use same copied from as PT model

* Remove whitespace changes

* Remove torch references

* Tie embeddings

* Remove logits processor input to generate

* Update logit values

* revert changes and add forced logit processor

* nit

* clean normalizer

* remove protected

* Add logit processors and update generation code & tests

* Some tidy up

* Update docstring

* update

* update based on review

* Update src/transformers/models/whisper/configuration_whisper.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/models/whisper/configuration_whisper.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update to reflect changes on the PT model branch

* Tidy up

* Remove extra whitespace

* Fix test - make input ids small enough we can append

* Include upstream changes on main

* PR comments - add batch tests, remove comments & defaults

* Fix model output imports

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/generation_tf_logits_process.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update tests/models/whisper/test_modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update docstring example

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Matt <Rocketknight1@users.noreply.github.com>

* Remove changes to adjust_logits_during_generation function

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Tidy up imports that don't require TF

* Update tests - skip and no more skip

* Update tests/generation/test_generation_tf_logits_process.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py

* Update src/transformers/models/whisper/modeling_tf_whisper.py

Co-authored-by: Matt <Rocketknight1@users.noreply.github.com>

* Add training flags

* Add (skipped) XLA generation tests

* Add embedding correctness test

* Add constant ids for generation tests

* Make logits finding a bit tidier

* Remove unused args

* xla generation enabled

* Don't skip XLA tests anymore

* Fix tests - add position ids to expected signature and update rag generation

* Undo method reorder

* Remove added whitespace

* Remove copy-paste gradient checkopint ref

* Remove

* Trigger CI - (issue with refs when pulling)

Co-authored-by: Arthur Zucker <arthur.zucker@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: NielsRogge <niels.rogge1@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Matt <Rocketknight1@users.noreply.github.com>
Co-authored-by: Joao Gante <joao@huggingface.co>
This commit is contained in:
amyeroberts 2022-10-10 14:48:17 +01:00 committed by GitHub
parent af69360bf9
commit e3f028f3af
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
20 changed files with 2755 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
README.md
View File

@ -55,13 +55,13 @@ limitations under the License.
<a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
</h3>
🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
These models can be applied on:
* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, text generation, in over 100 languages.
* 🖼️ Images, for tasks like image classification, object detection, and segmentation.
* 🗣️ Audio, for tasks like speech recognition and audio classification.
* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, text generation, in over 100 languages.
* 🖼️ Images, for tasks like image classification, object detection, and segmentation.
* 🗣️ Audio, for tasks like speech recognition and audio classification.
Transformer models can also perform tasks on **several modalities combined**, such as table question answering, optical character recognition, information extraction from scanned documents, video classification, and visual question answering.

22
README_ko.md
View File

@ -59,7 +59,7 @@ limitations under the License.
🤗 Transformers는 이러한 사전학습 모델을 빠르게 다운로드해 특정 텍스트에 사용하고, 원하는 데이터로 fine-tuning해 커뮤니티나 우리의 [모델 허브](https://huggingface.co/models)에 공유할 수 있도록 API를 제공합니다. 또한, 모델 구조를 정의하는 각 파이썬 모듈은 완전히 독립적이여서 연구 실험을 위해 손쉽게 수정할 수 있습니다.
🤗 Transformers는 가장 유명한 3개의 딥러닝 라이브러리를 지원합니다. 이들은 서로 완벽히 연동됩니다 — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/), [TensorFlow](https://www.tensorflow.org/). 간단하게 이 라이브러리 중 하나로 모델을 학습하고, 또 다른 라이브러리로 추론을 위해 모델을 불러올 수 있습니다.
🤗 Transformers는 가장 유명한 3개의 딥러닝 라이브러리를 지원합니다. 이들은 서로 완벽히 연동됩니다 — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/), [TensorFlow](https://www.tensorflow.org/). 간단하게 이 라이브러리 중 하나로 모델을 학습하고, 또 다른 라이브러리로 추론을 위해 모델을 불러올 수 있습니다.
## 온라인 데모
@ -74,7 +74,7 @@ limitations under the License.
- [DistilBERT를 이용한 질문 답변](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
- [T5로 번역하기](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
**[Transformer와 글쓰기](https://transformer.huggingface.co)** 는 이 저장소의 텍스트 생성 능력에 관한 Hugging Face 팀의 공식 데모입니다.
**[Transformer와 글쓰기](https://transformer.huggingface.co)** 는 이 저장소의 텍스트 생성 능력에 관한 Hugging Face 팀의 공식 데모입니다.
## Hugging Face 팀의 커스텀 지원을 원한다면
@ -258,7 +258,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
@ -297,7 +297,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
@ -307,9 +307,9 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
@ -325,11 +325,11 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
@ -345,7 +345,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
@ -355,7 +355,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
1. 새로운 모델을 올리고 싶나요? 우리가 **상세한 가이드와 템플릿** 으로 새로운 모델을 올리도록 도와드릴게요. 가이드와 템플릿은 이 저장소의 [`templates`](./templates) 폴더에서 확인하실 수 있습니다. [컨트리뷰션 가이드라인](./CONTRIBUTING.md)을 꼭 확인해주시고, PR을 올리기 전에 메인테이너에게 연락하거나 이슈를 오픈해 피드백을 받으시길 바랍니다.
1. 새로운 모델을 올리고 싶나요? 우리가 **상세한 가이드와 템플릿** 으로 새로운 모델을 올리도록 도와드릴게요. 가이드와 템플릿은 이 저장소의 [`templates`](./templates) 폴더에서 확인하실 수 있습니다. [컨트리뷰션 가이드라인](./CONTRIBUTING.md)을 꼭 확인해주시고, PR을 올리기 전에 메인테이너에게 연락하거나 이슈를 오픈해 피드백을 받으시길 바랍니다.
각 모델이 Flax, PyTorch, TensorFlow으로 구현되었는지 또는 🤗 Tokenizers 라이브러리가 지원하는 토크나이저를 사용하는지 확인하려면, [이 표](https://huggingface.co/docs/transformers/index#supported-frameworks)를 확인하세요.

20
README_zh-hans.md
View File

@ -245,7 +245,7 @@ conda install -c huggingface transformers
1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (来自 Google Research) 伴随论文 [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) 由 Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed 发布。
1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (来自 Alexa) 伴随论文 [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) 由 Adrian de Wynter and Daniel J. Perry 发布。
1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (来自 Google Research) 伴随论文 [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) 由 Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel 发布。
1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (来自 Inria/Facebook/Sorbonne) 伴随论文 [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) 由 Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot 发布。
@ -282,7 +282,7 @@ conda install -c huggingface transformers
1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。
1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。
1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (来自 EleutherAI) 随仓库 [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) 发布。作者为 Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy 发布。
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (来自 ABEJA) 由 Shinya Otani, Takayoshi Makabe, Anuj Arora, Kyo Hattori。
1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (来自 OpenAI) 伴随论文 [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) 由 Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever** 发布。
1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (来自 EleutherAI) 伴随论文 [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) 由 Ben Wang and Aran Komatsuzaki 发布。
@ -314,14 +314,14 @@ conda install -c huggingface transformers
1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (来自 Apple) 伴随论文 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 由 Sachin Mehta and Mohammad Rastegari 发布。
1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (来自 Microsoft Research) 伴随论文 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 由 Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 发布。
1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。
1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (来自 Meta AI) 伴随论文 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 由 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 发布。
1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (来自 Google AI) 伴随论文 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 由 Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 发布。
1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (来自 Google) 伴随论文 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 由 Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 发布。
1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。
1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。
1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (来自 Deepmind) 伴随论文 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 由 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 发布。
1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (来自 VinAI Research) 伴随论文 [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) 由 Dat Quoc Nguyen and Anh Tuan Nguyen 发布。
1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (来自 UCLA NLP) 伴随论文 [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) 由 Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang 发布。
@ -331,9 +331,9 @@ conda install -c huggingface transformers
1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (来自 Facebook) 伴随论文 [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) 由 Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela 发布。
1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (来自 Google Research) 伴随论文 [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) 由 Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang 发布。
1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (来自 Google Research) 伴随论文 [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) 由 Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya 发布。
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (来自 Google Research) 伴随论文 [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) 由 Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder 发布。
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (来自 Facebook), 伴随论文 [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) 由 Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov 发布。
1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (来自 ZhuiyiTechnology), 伴随论文 [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 由 Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 发布。
1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
@ -349,11 +349,11 @@ conda install -c huggingface transformers
1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (来自 Google AI) 伴随论文 [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) 由 Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu 发布。
1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (来自 Google AI) 伴随论文 [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) 由 Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos 发布。
1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (来自 Microsoft Research) 伴随论文 [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) 由 Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou 发布。
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (来自 Google/CMU) 伴随论文 [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) 由 Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov 发布。
1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (来自 Microsoft) 伴随论文 [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) 由 Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei 发布。
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (来自 Microsoft Research) 伴随论文 [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) 由 Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang 发布。
1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (来自 Microsoft Research) 伴随论文 [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) 由 Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu 发布。
1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (来自 Tsinghua University and Nankai University) 伴随论文 [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) 由 Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu 发布。
@ -369,7 +369,7 @@ conda install -c huggingface transformers
1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (来自 OpenAI) 伴随论文 [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) 由 Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever 发布。
1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (来自 Microsoft Research) 伴随论文 [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) 由 Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling 发布。
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (来自 Facebook) 伴随论文 [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) 由 Guillaume Lample and Alexis Conneau 发布。
1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (来自 Microsoft Research) 伴随论文 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 由 Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 发布。
1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (来自 Facebook AI), 伴随论文 [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) 由 Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov 发布。

20
README_zh-hant.md
View File

@ -257,7 +257,7 @@ conda install -c huggingface transformers
1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
@ -294,7 +294,7 @@ conda install -c huggingface transformers
1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released with the paper [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
@ -316,7 +316,7 @@ conda install -c huggingface transformers
1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
1. **[MarkupLM](https://huggingface.co/docs/transformers/main/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov
1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov
1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
@ -343,9 +343,9 @@ conda install -c huggingface transformers
1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
@ -361,17 +361,17 @@ conda install -c huggingface transformers
1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released with the paper [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Time Series Transformer](https://huggingface.co/docs/transformers/main/model_doc/time_series_transformer)** (from HuggingFace).
1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft) released with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
1. **[ViTMSN](https://huggingface.co/docs/transformers/main/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
@ -381,7 +381,7 @@ conda install -c huggingface transformers
1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.

2
docs/source/en/index.mdx
View File

@ -330,7 +330,7 @@ Flax), PyTorch, and/or TensorFlow.
| Wav2Vec2 | ✅ | ❌ | ✅ | ✅ | ✅ |
| Wav2Vec2-Conformer | ❌ | ❌ | ✅ | ❌ | ❌ |
| WavLM | ❌ | ❌ | ✅ | ❌ | ❌ |
| Whisper | ✅ | ❌ | ✅ | ❌ | ❌ |
| Whisper | ✅ | ❌ | ✅ | ✅ | ❌ |
| X-CLIP | ❌ | ❌ | ✅ | ❌ | ❌ |
| XGLM | ✅ | ✅ | ✅ | ✅ | ✅ |
| XLM | ✅ | ❌ | ✅ | ✅ | ❌ |

15
docs/source/en/model_doc/whisper.mdx
View File

@ -23,11 +23,11 @@ The abstract from the paper is the following:
Tips:
- The model usually performs well without requiring any finetuning.
- The model usually performs well without requiring any finetuning.
- The architecture follows a classic encoder-decoder architecture, which means that it relies on the [`~generation_utils.GenerationMixin.generate`] function for inference.
- One can use [`WhisperProcessor`] to prepare audio for the model, and decode the predicted ID's back into text.
This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ).
This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ). The Tensorflow version of this model was contributed by [amyeroberts](https://huggingface.co/amyeroberts).
The original code can be found [here](https://github.com/openai/whisper).
@ -66,3 +66,14 @@ The original code can be found [here](https://github.com/openai/whisper).
[[autodoc]] WhisperForConditionalGeneration
- forward
## TFWhisperModel
[[autodoc]] TFWhisperModel
- call
## TFWhisperForConditionalGeneration
[[autodoc]] TFWhisperForConditionalGeneration
- call

14
src/transformers/__init__.py
View File

@ -2754,6 +2754,14 @@ else:
"TFWav2Vec2PreTrainedModel",
]
)
_import_structure["models.whisper"].extend(
[
"TF_WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
"TFWhisperForConditionalGeneration",
"TFWhisperModel",
"TFWhisperPreTrainedModel",
]
)
_import_structure["models.xglm"].extend(
[
"TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST",
@ -5303,6 +5311,12 @@ if TYPE_CHECKING:
TFWav2Vec2Model,
TFWav2Vec2PreTrainedModel,
)
from .models.whisper import (
TF_WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST,
TFWhisperForConditionalGeneration,
TFWhisperModel,
TFWhisperPreTrainedModel,
)
from .models.xglm import (
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST,
TFXGLMForCausalLM,

81
src/transformers/generation_tf_logits_process.py
View File

@ -504,3 +504,84 @@ class TFForcedEOSTokenLogitsProcessor(TFLogitsProcessor):
axis=-1,
)
return scores
class TFSuppressTokensAtBeginLogitsProcessor(TFLogitsProcessor):
r"""
[`TFSuppressTokensAtBeginLogitsProcessor`] suppresses a list of tokens as soon as the `generate` function starts
generating using `begin_index` tokens. This should ensure that the tokens defined by `begin_suppress_tokens` at not
sampled at the begining of the generation.
"""
def __init__(self, begin_suppress_tokens, begin_index):
self.begin_suppress_tokens = list(begin_suppress_tokens)
self.begin_index = begin_index
def __call__(self, input_ids: tf.Tensor, scores: tf.Tensor, cur_len: int) -> tf.Tensor:
scores = tf.cond(
tf.equal(cur_len, self.begin_index),
lambda: tf.tensor_scatter_nd_update(
scores,
indices=[[i, token] for i in range(scores.shape[0]) for token in self.begin_suppress_tokens],
updates=[-float("inf") for _ in range(scores.shape[0] * len(self.begin_suppress_tokens))],
),
lambda: scores,
)
return scores
class TFSuppressTokensLogitsProcessor(TFLogitsProcessor):
r"""This processor can be used to suppress a list of tokens. The processor will set their log probs to `-inf` so that they
are not sampled."""
def __init__(self, suppress_tokens):
self.suppress_tokens = list(suppress_tokens)
def __call__(self, input_ids: tf.Tensor, scores: tf.Tensor, cur_len: int) -> tf.Tensor:
scores = tf.tensor_scatter_nd_update(
scores,
indices=[[i, token] for i in range(scores.shape[0]) for token in self.suppress_tokens],
updates=[-float("inf") for _ in range(scores.shape[0] * len(self.suppress_tokens))],
)
return scores
class TFForceTokensLogitsProcessor(TFLogitsProcessor):
r"""This processor can be used to force a list of tokens. The processor will set their log probs to `0` and all
other tokens to `-inf` so that they are sampled at their corresponding index."""
def __init__(self, force_token_map):
force_token_map = dict(force_token_map)
# Converts the dictionary of format {index: token} containing the tokens to be forced to an array, where the
# index of the array corresponds to the index of the token to be forced, for XLA compatibility.
# Indexes without forced tokens will have an negative value.
force_token_array = np.ones((max(force_token_map.keys()) + 1), dtype=np.int32) * -1
for index, token in force_token_map.items():
force_token_array[index] = token
self.force_token_array = tf.convert_to_tensor(force_token_array, dtype=tf.int32)
def __call__(self, input_ids: tf.Tensor, scores: tf.Tensor, cur_len: int) -> tf.Tensor:
def _force_token(generation_idx):
batch_size = scores.shape[0]
current_token = self.force_token_array[generation_idx]
new_scores = tf.ones_like(scores, dtype=scores.dtype) * -float("inf")
indices = tf.stack((tf.range(batch_size), tf.tile([current_token], [batch_size])), axis=1)
updates = tf.zeros((batch_size,), dtype=scores.dtype)
new_scores = tf.tensor_scatter_nd_update(new_scores, indices, updates)
return new_scores
scores = tf.cond(
tf.greater_equal(cur_len, tf.shape(self.force_token_array)[0]),
# If the current length is geq than the length of force_token_array, the processor does nothing.
lambda: tf.identity(scores),
# Otherwise, it may force a certain token.
lambda: tf.cond(
tf.greater_equal(self.force_token_array[cur_len], 0),
# Only valid (positive) tokens are forced
lambda: _force_token(cur_len),
# Otherwise, the processor does nothing.
lambda: scores,
),
)
return scores

62
src/transformers/generation_tf_utils.py
View File

@ -26,11 +26,14 @@ from tensorflow.compiler.tf2xla.python.xla import dynamic_update_slice
from .generation_tf_logits_process import (
TFForcedBOSTokenLogitsProcessor,
TFForcedEOSTokenLogitsProcessor,
TFForceTokensLogitsProcessor,
TFLogitsProcessorList,
TFMinLengthLogitsProcessor,
TFNoBadWordsLogitsProcessor,
TFNoRepeatNGramLogitsProcessor,
TFRepetitionPenaltyLogitsProcessor,
TFSuppressTokensAtBeginLogitsProcessor,
TFSuppressTokensLogitsProcessor,
TFTemperatureLogitsWarper,
TFTopKLogitsWarper,
TFTopPLogitsWarper,
@ -401,6 +404,9 @@ class TFGenerationMixin:
return_dict_in_generate=None,
forced_bos_token_id=None,
forced_eos_token_id=None,
suppress_tokens: Optional[List[int]] = None,
begin_suppress_tokens: Optional[List[int]] = None,
forced_decoder_ids: Optional[List[int]] = None,
**model_kwargs,
) -> Union[TFGreedySearchOutput, TFSampleOutput, TFBeamSearchOutput, TFBeamSampleOutput, tf.Tensor]:
r"""
@ -494,6 +500,14 @@ class TFGenerationMixin:
the target language token.
forced_eos_token_id (`int`, *optional*):
The id of the token to force as the last generated token when `max_length` is reached.
suppress_tokens (`List[int]`, *optional*, defaults to `model.config.suppress_tokens`):
A list of tokens that will be supressed at generation. The `SupressTokens` logit processor will set
their log probs to `-inf` so that they are not sampled.
begin_suppress_tokens (`List[int]`, *optional*, defaults to `model.config.begin_suppress_tokens`):
A list of tokens that will be supressed at the begining of the generation. The `SupressBeginTokens`
logit processor will set their log probs to `-inf` so that they are not sampled.
forced_decoder_ids (`List[int]`, *optional*, defaults to `model.config.forced_decoder_ids`):
A list of tokens that will be forced as beginning tokens, before sampling.
model_specific_kwargs:
Additional model specific kwargs will be forwarded to the `forward` function of the model.
@ -609,6 +623,9 @@ class TFGenerationMixin:
return_dict_in_generate=return_dict_in_generate,
forced_bos_token_id=forced_bos_token_id,
forced_eos_token_id=forced_eos_token_id,
suppress_tokens=suppress_tokens,
begin_suppress_tokens=begin_suppress_tokens,
forced_decoder_ids=forced_decoder_ids,
**model_kwargs,
)
@ -648,6 +665,12 @@ class TFGenerationMixin:
forced_eos_token_id = (
forced_eos_token_id if forced_eos_token_id is not None else self.config.forced_eos_token_id
)
suppress_tokens = suppress_tokens if suppress_tokens is not None else self.config.suppress_tokens
begin_suppress_tokens = (
begin_suppress_tokens if begin_suppress_tokens is not None else self.config.begin_suppress_tokens
)
if forced_decoder_ids is None and hasattr(self.config, "forced_decoder_ids"):
forced_decoder_ids = self.config.forced_decoder_ids
output_scores = output_scores if output_scores is not None else self.config.output_scores
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
@ -1368,6 +1391,9 @@ class TFGenerationMixin:
return_dict_in_generate=None,
forced_bos_token_id=None,
forced_eos_token_id=None,
suppress_tokens=None,
begin_suppress_tokens=None,
forced_decoder_ids=None,
**model_kwargs,
) -> Union[TFGreedySearchOutput, TFSampleOutput, TFBeamSearchOutput, TFBeamSampleOutput, tf.Tensor]:
r"""
@ -1461,6 +1487,15 @@ class TFGenerationMixin:
the target language token.
forced_eos_token_id (`int`, *optional*):
The id of the token to force as the last generated token when `max_length` is reached.
suppress_tokens (`List[int]`, *optional*, defaults to `model.config.suppress_tokens`):
A list of tokens that will be supressed at generation. The `SupressTokens` logit processor will set
their log probs to `-inf` so that they are not sampled.
begin_suppress_tokens (`List[int]`, *optional*, defaults to `model.config.begin_suppress_tokens`):
A list of tokens that will be supressed at the begining of the generation. The `SupressBeginTokens`
logit processor will set their log probs to `-inf` so that they are not sampled.
forced_decoder_ids (`List[int]`, *optional*, defaults to `model.config.forced_decoder_ids`):
A list of tokens that will be forced as beginning tokens.
model_kwargs:
Additional model specific kwargs will be forwarded to the `call` function of the model.
@ -1695,12 +1730,16 @@ class TFGenerationMixin:
logits_processor = self._get_logits_processor(
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
input_ids_seq_length=input_ids_seq_length,
bad_words_ids=bad_words_ids,
min_length=min_length,
max_length=max_length,
eos_token_id=eos_token_id,
forced_bos_token_id=forced_bos_token_id,
forced_eos_token_id=forced_eos_token_id,
suppress_tokens=suppress_tokens,
begin_suppress_tokens=begin_suppress_tokens,
forced_decoder_ids=forced_decoder_ids,
)
# 9. go into different generation modes
@ -1994,7 +2033,7 @@ class TFGenerationMixin:
def _initialize_past(past, num_padding_values, batch_axis):
"""initialize past with zeros -- the structure depends on `batch_axis`"""
if batch_axis == 0:
padding_values = tf.scatter_nd(indices=[[2, 1]], updates=[num_padding_values], shape=(4, 2))
padding_values = tf.constant([[0, 0], [0, 0], [0, num_padding_values], [0, 0]], dtype=tf.int32)
new_past = ()
for past_layer in past:
new_past_layer = list(past_layer)
@ -2099,12 +2138,16 @@ class TFGenerationMixin:
self,
repetition_penalty: float,
no_repeat_ngram_size: int,
input_ids_seq_length: int,
bad_words_ids: List[List[int]],
min_length: int,
max_length: int,
eos_token_id: int,
forced_bos_token_id: int,
forced_eos_token_id: int,
suppress_tokens: Optional[List[int]] = None,
begin_suppress_tokens: Optional[List[int]] = None,
forced_decoder_ids: Optional[List[int]] = None,
) -> TFLogitsProcessorList:
"""
This class returns a [`TFLogitsProcessorList`] list object that contains all relevant [`TFLogitsProcessor`]
@ -2118,6 +2161,12 @@ class TFGenerationMixin:
)
bad_words_ids = bad_words_ids if bad_words_ids is not None else self.config.bad_words_ids
eos_token_id = eos_token_id if eos_token_id is not None else self.config.eos_token_id
suppress_tokens = suppress_tokens if suppress_tokens is not None else self.config.suppress_tokens
begin_suppress_tokens = (
begin_suppress_tokens if begin_suppress_tokens is not None else self.config.begin_suppress_tokens
)
if forced_decoder_ids is None and hasattr(self.config, "forced_decoder_ids"):
forced_decoder_ids = self.config.forced_decoder_ids
# instantiate processors list
if repetition_penalty is not None and repetition_penalty != 1.0:
@ -2132,7 +2181,16 @@ class TFGenerationMixin:
processors.append(TFForcedBOSTokenLogitsProcessor(forced_bos_token_id))
if forced_eos_token_id is not None:
processors.append(TFForcedEOSTokenLogitsProcessor(max_length, forced_eos_token_id))
if suppress_tokens is not None:
processors.append(TFSuppressTokensLogitsProcessor(suppress_tokens))
if begin_suppress_tokens is not None:
begin_index = input_ids_seq_length
begin_index = begin_index if (input_ids_seq_length > 1 or forced_bos_token_id is None) else begin_index + 1
if forced_decoder_ids is not None:
begin_index += forced_decoder_ids[-1][0] # generation starts after the last token that is forced
processors.append(TFSuppressTokensAtBeginLogitsProcessor(begin_suppress_tokens, begin_index))
if forced_decoder_ids is not None:
processors.append(TFForceTokensLogitsProcessor(forced_decoder_ids))
return processors
def greedy_search(

3
src/transformers/models/auto/modeling_tf_auto.py
View File

@ -80,6 +80,7 @@ TF_MODEL_MAPPING_NAMES = OrderedDict(
("vit", "TFViTModel"),
("vit_mae", "TFViTMAEModel"),
("wav2vec2", "TFWav2Vec2Model"),
("whisper", "TFWhisperModel"),
("xglm", "TFXGLMModel"),
("xlm", "TFXLMModel"),
("xlm-roberta", "TFXLMRobertaModel"),
@ -145,6 +146,7 @@ TF_MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
("t5", "TFT5ForConditionalGeneration"),
("tapas", "TFTapasForMaskedLM"),
("transfo-xl", "TFTransfoXLLMHeadModel"),
("whisper", "TFWhisperForConditionalGeneration"),
("xlm", "TFXLMWithLMHeadModel"),
("xlm-roberta", "TFXLMRobertaForMaskedLM"),
("xlnet", "TFXLNetLMHeadModel"),
@ -253,6 +255,7 @@ TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
TF_MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES = OrderedDict(
[
("speech_to_text", "TFSpeech2TextForConditionalGeneration"),
("whisper", "TFWhisperForConditionalGeneration"),
]
)

1
src/transformers/models/rag/modeling_tf_rag.py
View File

@ -1262,6 +1262,7 @@ class TFRagTokenForGeneration(TFRagPreTrainedModel, TFCausalLanguageModelingLoss
eos_token_id=eos_token_id,
forced_bos_token_id=None,
forced_eos_token_id=None,
input_ids_seq_length=tf.shape(decoder_input_ids)[-1],
)
model_kwargs["attention_mask"] = context_attention_mask

27
src/transformers/models/whisper/__init__.py
View File

@ -17,7 +17,7 @@
# limitations under the License.
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
_import_structure = {
@ -41,6 +41,18 @@ else:
"WhisperPreTrainedModel",
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_import_structure["modeling_tf_whisper"] = [
"TF_WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
"TFWhisperForConditionalGeneration",
"TFWhisperModel",
"TFWhisperPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_whisper import WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP, WhisperConfig
@ -61,6 +73,19 @@ if TYPE_CHECKING:
WhisperPreTrainedModel,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_whisper import (
TF_WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST,
TFWhisperForConditionalGeneration,
TFWhisperModel,
TFWhisperPreTrainedModel,
)
else:
import sys

14
src/transformers/models/whisper/feature_extraction_whisper.py
View File

@ -218,7 +218,6 @@ class WhisperFeatureExtractor(SequenceFeatureExtractor):
return_attention_mask: Optional[bool] = None,
padding: Optional[str] = "max_length",
max_length: Optional[int] = None,
sampling_rate: Optional[int] = None,
**kwargs
) -> BatchFeature:
"""
@ -262,19 +261,6 @@ class WhisperFeatureExtractor(SequenceFeatureExtractor):
The value that is used to fill the padding values / vectors.
"""
if sampling_rate is not None:
if sampling_rate != self.sampling_rate:
raise ValueError(
f"The model corresponding to this feature extractor: {self} was trained using a sampling rate of"
f" {self.sampling_rate}. Please make sure that the provided `raw_speech` input was sampled with"
f" {self.sampling_rate} and not {sampling_rate}."
)
else:
logger.warning(
"It is strongly recommended to pass the `sampling_rate` argument to this function. "
"Failing to do so can result in silent errors that might be hard to debug."
)
is_batched = bool(
isinstance(raw_speech, (list, tuple))
and (isinstance(raw_speech[0], np.ndarray) or isinstance(raw_speech[0], (tuple, list)))

1401
src/transformers/models/whisper/modeling_tf_whisper.py Normal file
View File

File diff suppressed because it is too large Load Diff

24
src/transformers/utils/dummy_tf_objects.py
View File

@ -2394,6 +2394,30 @@ class TFWav2Vec2PreTrainedModel(metaclass=DummyObject):
requires_backends(self, ["tf"])
TF_WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST = None
class TFWhisperForConditionalGeneration(metaclass=DummyObject):
_backends = ["tf"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["tf"])
class TFWhisperModel(metaclass=DummyObject):
_backends = ["tf"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["tf"])
class TFWhisperPreTrainedModel(metaclass=DummyObject):
_backends = ["tf"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["tf"])
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST = None

83
tests/generation/test_generation_tf_logits_process.py
View File

@ -29,11 +29,14 @@ if is_tf_available():
from transformers.generation_tf_logits_process import (
TFForcedBOSTokenLogitsProcessor,
TFForcedEOSTokenLogitsProcessor,
TFForceTokensLogitsProcessor,
TFLogitsProcessorList,
TFMinLengthLogitsProcessor,
TFNoBadWordsLogitsProcessor,
TFNoRepeatNGramLogitsProcessor,
TFRepetitionPenaltyLogitsProcessor,
TFSuppressTokensAtBeginLogitsProcessor,
TFSuppressTokensLogitsProcessor,
TFTemperatureLogitsWarper,
TFTopKLogitsWarper,
TFTopPLogitsWarper,
@ -331,6 +334,86 @@ class TFLogitsProcessorTest(unittest.TestCase):
scores = logits_processor(input_ids, scores, cur_len)
self.assertFalse(tf.math.reduce_any(tf.math.is_inf((scores))))
@parameterized.expand([(False,), (True,)])
def test_suppress_tokens_at_begin_logits_processor(self, use_xla):
vocab_size = 20
batch_size = 4
begin_suppress_tokens = [1, 2, 3]
begin_index = 5
logits_processor = TFSuppressTokensAtBeginLogitsProcessor(
begin_suppress_tokens=begin_suppress_tokens, begin_index=begin_index
)
if use_xla:
logits_processor = tf.function(logits_processor, jit_compile=True)
# Check that no scores are suppressed if begin_index is not reached
cur_len = 4
input_ids = tf.convert_to_tensor([[11, 17, 15, 8], [14, 0, 19, 5], [13, 11, 18, 19], [11, 12, 16, 15]])
scores = self._get_uniform_logits(batch_size, vocab_size)
scores = logits_processor(input_ids, scores, cur_len)
self.assertFalse(tf.math.reduce_any(tf.math.is_inf((scores))))
# Check that scores are suppressed if begin_index is reached
cur_len = 5
input_ids = tf.convert_to_tensor([[5, 5, 5, 0, 17], [18, 1, 9, 14, 17], [18, 6, 8, 15, 19], [8, 12, 17, 1, 2]])
scores = self._get_uniform_logits(batch_size, vocab_size)
scores = logits_processor(input_ids, scores, cur_len)
self.assertTrue(tf.math.reduce_all(tf.math.is_inf(tf.gather(scores, begin_suppress_tokens, axis=1))))
@parameterized.expand([(False,), (True,)])
def test_suppress_tokens_logits_processor(self, use_xla):
vocab_size = 20
batch_size = 4
suppress_tokens = [1, 3, 5]
keep_tokens = [i for i in range(vocab_size) if i not in suppress_tokens]
logits_processor = TFSuppressTokensLogitsProcessor(suppress_tokens=suppress_tokens)
if use_xla:
logits_processor = tf.function(logits_processor, jit_compile=True)
# Check that suppress_tokens are suppressed and others are not
cur_len = 5
input_ids = tf.convert_to_tensor([[0, 10, 19, 6, 3], [17, 4, 8, 17, 2], [7, 1, 11, 6, 15], [5, 8, 13, 16, 0]])
scores = self._get_uniform_logits(batch_size, vocab_size)
scores = logits_processor(input_ids, scores, cur_len)
self.assertTrue(tf.math.reduce_all(tf.math.is_inf(tf.gather(scores, suppress_tokens, axis=1))))
self.assertFalse(tf.math.reduce_any(tf.math.is_inf(tf.gather(scores, keep_tokens, axis=1))))
@parameterized.expand([(False,), (True,)])
def test_force_tokens_logits_processor(self, use_xla):
vocab_size = 20
batch_size = 4
force_token_map = {1: 2, 3: 2}
logits_processor = TFForceTokensLogitsProcessor(force_token_map=force_token_map)
if use_xla:
logits_processor = tf.function(logits_processor, jit_compile=True)
# check that if the cur_len is contained in the force_token_map, the logits are the same
# for all tokens except the one the force_token_map points to
cur_len = 1
input_ids = tf.convert_to_tensor([[11], [7], [5], [15]])
ids_tensor((batch_size, cur_len), vocab_size=20)
scores = self._get_uniform_logits(batch_size, vocab_size)
scores = logits_processor(input_ids, scores, cur_len)
tf.debugging.assert_near(tf.gather(scores, [force_token_map[cur_len]], axis=1), 0.0)
non_forced_inds = [i for i in range(vocab_size) if i != force_token_map[cur_len]]
self.assertTrue(
tf.math.reduce_all(tf.math.is_inf(tf.gather(scores, [non_forced_inds], axis=1))),
)
# check that if the cur_len is not contained in the force_token_map, the logits are not modified
cur_len = 2
input_ids = tf.convert_to_tensor([[2, 19], [19, 15], [4, 9], [7, 6]])
scores = self._get_uniform_logits(batch_size, vocab_size)
scores = logits_processor(input_ids, scores, cur_len)
self.assertFalse(tf.math.reduce_any(tf.math.is_inf((scores))))
@parameterized.expand([(False,), (True,)])
def test_processor_list(self, use_xla):
# TODO (Joao): reintroduce TFNoRepeatNGramLogitsProcessor when it gets compatible with XLA

983
tests/models/whisper/test_modeling_tf_whisper.py Normal file
View File

@ -0,0 +1,983 @@
# coding=utf-8
# Copyright 2022 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Testing suite for the TensorFlow Whisper model. """
import inspect
import tempfile
import unittest
import numpy as np
from transformers import WhisperConfig, WhisperFeatureExtractor, WhisperProcessor
from transformers.testing_utils import is_tf_available, require_tf, require_tokenizers, slow
from transformers.utils import cached_property
from transformers.utils.import_utils import is_datasets_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor
if is_datasets_available():
import datasets
from datasets import load_dataset
if is_tf_available():
import tensorflow as tf
from transformers import TFWhisperForConditionalGeneration, TFWhisperModel, set_seed
from transformers.models.whisper.modeling_tf_whisper import TFWhisperDecoder, TFWhisperEncoder
def prepare_whisper_inputs_dict(
config,
input_features,
decoder_input_ids,
attention_mask=None,
decoder_attention_mask=None,
head_mask=None,
decoder_head_mask=None,
cross_attn_head_mask=None,
):
if decoder_attention_mask is None:
decoder_attention_mask = tf.where(decoder_input_ids != config.pad_token_id, 1, 0)
if head_mask is None:
head_mask = tf.ones((config.encoder_layers, config.encoder_attention_heads))
if decoder_head_mask is None:
decoder_head_mask = tf.ones((config.decoder_layers, config.decoder_attention_heads))
if cross_attn_head_mask is None:
cross_attn_head_mask = tf.ones((config.decoder_layers, config.decoder_attention_heads))
return {
"input_features": input_features,
"decoder_input_ids": decoder_input_ids,
"decoder_attention_mask": decoder_attention_mask,
"head_mask": head_mask,
"decoder_head_mask": decoder_head_mask,
"cross_attn_head_mask": cross_attn_head_mask,
}
@require_tf
class TFWhisperModelTester:
def __init__(
self,
parent,
batch_size=13,
seq_length=60,
is_training=True,
use_labels=False,
vocab_size=99,
hidden_size=16,
num_hidden_layers=2,
num_attention_heads=4,
input_channels=1,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=20,
max_source_positions=30,
max_target_positions=60,
bos_token_id=98,
eos_token_id=98,
pad_token_id=0,
num_mel_bins=80,
decoder_start_token_id=85,
num_conv_layers=1,
suppress_tokens=None,
begin_suppress_tokens=None,
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_labels = use_labels
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.input_channels = input_channels
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.num_mel_bins = num_mel_bins
self.max_position_embeddings = max_position_embeddings
self.max_source_positions = max_source_positions
self.max_target_positions = max_target_positions
self.eos_token_id = eos_token_id
self.pad_token_id = pad_token_id
self.bos_token_id = bos_token_id
self.decoder_start_token_id = decoder_start_token_id
self.num_conv_layers = num_conv_layers
self.suppress_tokens = suppress_tokens
self.begin_suppress_tokens = begin_suppress_tokens
def prepare_config_and_inputs(self):
input_features = floats_tensor([self.batch_size, self.num_mel_bins, self.seq_length], self.vocab_size)
decoder_input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
config = self.get_config()
inputs_dict = prepare_whisper_inputs_dict(
config,
attention_mask=None,
input_features=input_features,
decoder_input_ids=decoder_input_ids,
)
return config, inputs_dict
def get_config(self):
return WhisperConfig(
vocab_size=self.vocab_size,
d_model=self.hidden_size,
encoder_layers=self.num_hidden_layers,
decoder_layers=self.num_hidden_layers,
encoder_attention_heads=self.num_attention_heads,
decoder_attention_heads=self.num_attention_heads,
input_channels=self.input_channels,
dropout=self.hidden_dropout_prob,
attention_dropout=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
max_source_positions=self.max_source_positions,
max_target_positions=self.max_target_positions,
eos_token_id=self.eos_token_id,
bos_token_id=self.bos_token_id,
pad_token_id=self.pad_token_id,
decoder_ffn_dim=self.hidden_size,
encoder_ffn_dim=self.hidden_size,
decoder_start_token_id=self.decoder_start_token_id,
suppress_tokens=self.suppress_tokens,
begin_suppress_tokens=self.begin_suppress_tokens,
)
def prepare_config_and_inputs_for_common(self):
config, inputs_dict = self.prepare_config_and_inputs()
return config, inputs_dict
def get_subsampled_output_lengths(self, input_lengths):
"""
Computes the output length of the convolutional layers
"""
for i in range(self.num_conv_layers):
input_lengths = (input_lengths - 1) // 2 + 1
return input_lengths
def create_and_check_model_forward(self, config, inputs_dict):
model = TFWhisperModel(config=config)
input_features = inputs_dict["input_features"]
decoder_input_ids = inputs_dict["decoder_input_ids"]
# first forward pass
last_hidden_state = model(input_features, decoder_input_ids=decoder_input_ids).last_hidden_state
self.parent.assertTrue(last_hidden_state.shape, (13, 7, 16))
def create_and_check_decoder_model_past_large_inputs(self, config, inputs_dict):
model = TFWhisperModel(config=config).get_decoder()
# take a slice so we're shorter than the seqeuence length and can append later
input_ids = inputs_dict["decoder_input_ids"][:, :-10]
attention_mask = inputs_dict["decoder_attention_mask"][:, :-10]
# first forward pass
outputs = model(input_ids, attention_mask=attention_mask, use_cache=True)
output, past_key_values = outputs.to_tuple()
# create hypothetical multiple next token and extent to next_input_ids
next_token = ids_tensor((self.batch_size, 3), config.vocab_size)
next_tokens = tf.where(next_token <= 2, 2, next_token)
next_attn_mask = ids_tensor((self.batch_size, 3), 2)
# append to next input_ids and
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
next_attention_mask = tf.concat([attention_mask, next_attn_mask], axis=-1)
output_from_no_past = model(next_input_ids, attention_mask=next_attention_mask)["last_hidden_state"]
output_from_past = model(next_tokens, attention_mask=next_attention_mask, past_key_values=past_key_values)[
"last_hidden_state"
]
# select random slice
random_slice_idx = np.random.randint(0, output_from_past.shape[-1])
output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx]
output_from_past_slice = output_from_past[:, :, random_slice_idx]
self.parent.assertTrue(output_from_past_slice.shape[1] == next_tokens.shape[1])
# test that outputs are equal for slice
self.parent.assertTrue(np.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-2))
def check_encoder_decoder_model_standalone(self, config, inputs_dict):
model = TFWhisperModel(config=config)
outputs = model(**inputs_dict)
encoder_last_hidden_state = outputs.encoder_last_hidden_state
last_hidden_state = outputs.last_hidden_state
with tempfile.TemporaryDirectory() as tmpdirname:
encoder = model.get_encoder()
encoder.save_pretrained(tmpdirname)
encoder = TFWhisperEncoder.from_pretrained(tmpdirname)
encoder_last_hidden_state_2 = encoder(inputs_dict["input_features"])[0]
self.parent.assertTrue((encoder_last_hidden_state_2 - encoder_last_hidden_state).abs().max() < 1e-3)
with tempfile.TemporaryDirectory() as tmpdirname:
decoder = model.get_decoder()
decoder.save_pretrained(tmpdirname)
decoder = TFWhisperDecoder.from_pretrained(tmpdirname)
last_hidden_state_2 = decoder(
input_ids=inputs_dict["decoder_input_ids"],
attention_mask=inputs_dict["decoder_attention_mask"],
encoder_hidden_states=encoder_last_hidden_state,
)[0]
self.parent.assertTrue((last_hidden_state_2 - last_hidden_state).abs().max() < 1e-3)
@require_tf
class TFWhisperModelTest(TFModelTesterMixin, unittest.TestCase):
all_model_classes = (TFWhisperModel, TFWhisperForConditionalGeneration) if is_tf_available() else ()
all_generative_model_classes = (TFWhisperForConditionalGeneration,) if is_tf_available() else ()
is_encoder_decoder = True
fx_compatible = False
test_pruning = False
test_missing_keys = False
test_onnx = False
input_name = "input_features"
def setUp(self):
self.model_tester = TFWhisperModelTester(self)
self.config_tester = ConfigTester(self, config_class=WhisperConfig)
self.maxDiff = 3000
def test_config(self):
self.config_tester.run_common_tests()
def test_save_load_strict(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs()
for model_class in self.all_model_classes:
model = model_class(config)
model(model.dummy_inputs)
with tempfile.TemporaryDirectory() as tmpdirname:
model.save_pretrained(tmpdirname, saved_model=False)
model2, info = model_class.from_pretrained(tmpdirname, output_loading_info=True)
self.assertEqual(info["missing_keys"], [])
def test_model_forward(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model_forward(*config_and_inputs)
def test_decoder_model_past_with_large_inputs(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_decoder_model_past_large_inputs(*config_and_inputs)
def _get_input_ids_and_config(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict[self.input_name]
# cut to half length & take max batch_size 3
max_batch_size = 3
input_ids = input_ids[:max_batch_size, :, :]
# generate max 3 tokens
max_length = input_ids.shape[-1] + 3
if config.eos_token_id is not None and config.pad_token_id is None:
# hack to allow generate for models such as GPT2 as is done in `generate()`
config.pad_token_id = config.eos_token_id
return config, input_ids, None, max_length
# not implemented currently
def test_inputs_embeds(self):
pass
@unittest.skip("Training is not yet supported")
def test_training(self):
pass
def test_generate_with_head_masking(self):
pass
@unittest.skip("fp16 is not yet supported for TF models")
def test_generate_fp16(self):
config, input_dict = self.model_tester.prepare_config_and_inputs()
config.max_target_positions = 400
input_features = input_dict["input_features"]
model = TFWhisperForConditionalGeneration(config)
model.generate(input_features)
model.generate(input_features, num_beams=4, do_sample=True, early_stopping=False, num_return_sequences=3)
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
signature = inspect.signature(model.call)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = [
"input_features",
"decoder_input_ids",
"decoder_attention_mask",
]
expected_arg_names.extend(
["decoder_position_ids", "head_mask", "decoder_head_mask", "cross_attn_head_mask", "encoder_outputs"]
if "head_mask" and "decoder_head_mask" and "cross_attn_head_mask" in arg_names
else ["encoder_outputs"]
)
self.assertListEqual(arg_names[: len(expected_arg_names)], expected_arg_names)
def test_hidden_states_output(self):
def check_hidden_states_output(inputs_dict, config, model_class):
model = model_class(config)
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
hidden_states = outputs.encoder_hidden_states if config.is_encoder_decoder else outputs.hidden_states
expected_num_layers = getattr(
self.model_tester, "expected_num_hidden_layers", self.model_tester.num_hidden_layers + 1
)
self.assertEqual(len(hidden_states), expected_num_layers)
if hasattr(self.model_tester, "encoder_seq_length"):
seq_length = self.model_tester.encoder_seq_length
else:
seq_length = self.model_tester.seq_length
subsampled_seq_length = model._get_feat_extract_output_lengths(seq_length)
self.assertListEqual(
list(hidden_states[0].shape[-2:]),
[subsampled_seq_length, self.model_tester.hidden_size],
)
if config.is_encoder_decoder:
hidden_states = outputs.decoder_hidden_states
self.assertIsInstance(hidden_states, (list, tuple))
self.assertEqual(len(hidden_states), expected_num_layers)
decoder_seq_length = getattr(self.model_tester, "decoder_seq_length", seq_length)
self.assertListEqual(
list(hidden_states[0].shape[-2:]),
[decoder_seq_length, self.model_tester.hidden_size],
)
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
inputs_dict["output_hidden_states"] = True
check_hidden_states_output(inputs_dict, config, model_class)
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
config.output_hidden_states = True
check_hidden_states_output(inputs_dict, config, model_class)
def test_attention_outputs(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.return_dict = True
seq_len = getattr(self.model_tester, "seq_length", None)
decoder_seq_length = getattr(self.model_tester, "decoder_seq_length", seq_len)
encoder_seq_length = getattr(self.model_tester, "encoder_seq_length", seq_len)
encoder_key_length = getattr(self.model_tester, "key_length", encoder_seq_length)
decoder_key_length = getattr(self.model_tester, "decoder_key_length", encoder_key_length)
for model_class in self.all_model_classes:
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
subsampled_encoder_seq_length = model._get_feat_extract_output_lengths(encoder_seq_length)
subsampled_encoder_key_length = model._get_feat_extract_output_lengths(encoder_key_length)
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
attentions = outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions
self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
config.output_attentions = True
model = model_class(config)
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
attentions = outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions
self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
self.assertListEqual(
list(attentions[0].shape[-3:]),
[self.model_tester.num_attention_heads, subsampled_encoder_seq_length, subsampled_encoder_key_length],
)
out_len = len(outputs)
correct_outlen = 5
# loss is at first position
if "labels" in inputs_dict:
correct_outlen += 1 # loss is added to beginning
if "past_key_values" in outputs:
correct_outlen += 1 # past_key_values have been returned
self.assertEqual(out_len, correct_outlen)
# decoder attentions
decoder_attentions = outputs.decoder_attentions
self.assertIsInstance(decoder_attentions, (list, tuple))
self.assertEqual(len(decoder_attentions), self.model_tester.num_hidden_layers)
self.assertListEqual(
list(decoder_attentions[0].shape[-3:]),
[self.model_tester.num_attention_heads, decoder_seq_length, decoder_key_length],
)
# cross attentions
cross_attentions = outputs.cross_attentions
self.assertIsInstance(cross_attentions, (list, tuple))
self.assertEqual(len(cross_attentions), self.model_tester.num_hidden_layers)
self.assertListEqual(
list(cross_attentions[0].shape[-3:]),
[
self.model_tester.num_attention_heads,
decoder_seq_length,
subsampled_encoder_key_length,
],
)
# Check attention is always last and order is fine
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = True
model = model_class(config)
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
added_hidden_states = 2
self.assertEqual(out_len + added_hidden_states, len(outputs))
self_attentions = outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions
self.assertEqual(len(self_attentions), self.model_tester.num_hidden_layers)
self.assertListEqual(
list(self_attentions[0].shape[-3:]),
[self.model_tester.num_attention_heads, subsampled_encoder_seq_length, subsampled_encoder_key_length],
)
def test_generate_without_input_ids(self):
pass
@staticmethod
def _get_encoder_outputs(
model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
):
encoder = model.get_encoder()
encoder_outputs = encoder(
input_ids,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
num_interleave, dim=0
)
input_ids = input_ids[:, :, 0]
input_ids = tf.zeros_like(input_ids[:, :1], dtype=tf.int64) + tf.convert_to_tensor(
[model._get_decoder_start_token_id()]
)
attention_mask = None
return encoder_outputs, input_ids, attention_mask
def _check_outputs(self, output, input_ids, config, use_cache=False, num_return_sequences=1):
batch_size, mel, seq_length = input_ids.shape
subsampled_seq_length = self.model_tester.get_subsampled_output_lengths(seq_length)
num_sequences_in_output = batch_size * num_return_sequences
gen_len = (
output.sequences.shape[-1] - 1 if config.is_encoder_decoder else output.sequences.shape[-1] - seq_length
)
# scores
self._check_scores(num_sequences_in_output, output.scores, length=gen_len, config=config)
# Attentions
# encoder
self._check_encoder_attention_for_generate(
output.encoder_attentions, batch_size, config, subsampled_seq_length
)
# decoder
self._check_attentions_for_generate(
num_sequences_in_output,
output.decoder_attentions,
min_length=1,
max_length=output.sequences.shape[-1],
config=config,
use_cache=use_cache,
)
# Hidden States
# encoder
self._check_encoder_hidden_states_for_generate(
output.encoder_hidden_states, batch_size, config, subsampled_seq_length
)
# decoder
self._check_hidden_states_for_generate(
num_sequences_in_output,
output.decoder_hidden_states,
min_length=1,
max_length=output.sequences.shape[-1],
config=config,
use_cache=use_cache,
)
# overwritten from parent due to the inability to work when non-text inputs are not passed AND because the input is
# `input_features`
def test_lm_head_model_random_no_beam_search_generate(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_features = inputs_dict.get("input_features", None)
# iterate over all generative models
for model_class in self.all_generative_model_classes:
model = model_class(config)
if config.bos_token_id is None:
# if bos token id is not defined model needs input_features
with self.assertRaises(AssertionError):
model.generate(do_sample=True, max_length=5)
# num_return_sequences = 1
self._check_generated_ids(model.generate(input_features, do_sample=True))
with self.assertRaises(ValueError):
# generating multiple sequences when no beam search generation
# is not allowed as it would always generate the same sequences
model.generate(input_features, do_sample=False, num_return_sequences=2)
# num_return_sequences > 1, sample
self._check_generated_ids(model.generate(input_features, do_sample=True, num_return_sequences=2))
# check bad words tokens language generation
# create list of 1-seq bad token and list of 2-seq of bad tokens
bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
output_tokens = model.generate(
input_features, do_sample=True, bad_words_ids=bad_words_ids, num_return_sequences=2
)
# only count generated tokens
generated_ids = output_tokens[:, input_features.shape[-1] :]
self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))
# overwritten from parent due to the inability to work when non-text inputs are not passed AND because the input is
# `input_features`
def test_lm_head_model_random_beam_search_generate(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_features = inputs_dict.get("input_features", None)
for model_class in self.all_generative_model_classes:
model = model_class(config)
if config.bos_token_id is None:
# if bos token id is not defined model needs input_ids, num_return_sequences = 1
self._check_generated_ids(model.generate(input_features, do_sample=True, num_beams=2))
with self.assertRaises(ValueError):
# generating more sequences than having beams leads is not possible
model.generate(input_features, do_sample=False, num_return_sequences=3, num_beams=2)
# num_return_sequences > 1, sample
self._check_generated_ids(
model.generate(
input_features,
do_sample=True,
num_beams=2,
num_return_sequences=2,
)
)
# num_return_sequences > 1, greedy
self._check_generated_ids(
model.generate(input_features, do_sample=False, num_beams=2, num_return_sequences=2)
)
# check bad words tokens language generation
# create list of 1-seq bad token and list of 2-seq of bad tokens
bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
output_tokens = model.generate(
input_features, do_sample=False, bad_words_ids=bad_words_ids, num_beams=2, num_return_sequences=2
)
# only count generated tokens
generated_ids = output_tokens[:, input_features.shape[-1] :]
self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))
@require_tf
@require_tokenizers
class TFWhisperModelIntegrationTests(unittest.TestCase):
@cached_property
def default_processor(self):
return WhisperProcessor.from_pretrained("openai/whisper-base")
def _load_datasamples(self, num_samples):
ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
# automatic decoding with librispeech
speech_samples = ds.sort("id").select(range(num_samples))[:num_samples]["audio"]
return [x["array"] for x in speech_samples]
@slow
def test_tiny_logits_librispeech(self):
set_seed(0)
model = TFWhisperModel.from_pretrained("openai/whisper-tiny")
input_speech = self._load_datasamples(1)
feature_extractor = WhisperFeatureExtractor()
input_features = feature_extractor(input_speech, return_tensors="tf").input_features
logits = model(
input_features,
decoder_input_ids=tf.convert_to_tensor([[50258, 50259, 50359]]),
output_hidden_states=False,
output_attentions=False,
return_dict=False,
use_cache=False,
)
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
2.9892, -6.7607, 5.7348, 3.6096, 0.2152, -5.7321, 4.8855, -1.6407,
0.2823, -1.5718, 10.4269, 3.4427, 0.0219, -8.0612, 3.4784, 8.4246,
4.0575, -2.2864, 11.1084, 0.9963, 0.9884, -8.5154, -3.5469, -9.3713,
0.9786, 3.5435, 7.4850, -5.2579, -1.4366, 10.4841
]
)
# fmt: on
self.assertTrue(np.allclose(logits[0][0, 0, :30], EXPECTED_LOGITS, atol=1e-4))
# fmt: off
EXPECTED_GENERATION = tf.convert_to_tensor(
[
-1.4651, -2.6944, 2.7821, 2.3793, 4.0738, 0.0188, -3.3203, 1.9836,
0.0520, 0.7095, 1.1063, 0.2952, -3.6786, -0.5249, 0.3105, 4.7691,
1.1562, 1.3046, 0.5810, -0.3624, 1.7006, 1.3424, 0.9817, 2.1958,
1.8775, -5.7046, -0.7679, 4.0113, 2.6848, 2.8609
]
)
# fmt: on
head_logits = logits[0] @ tf.transpose(model.model.decoder.embed_tokens.weights[0])
self.assertTrue(np.allclose(head_logits[0, 0, :30], EXPECTED_GENERATION, atol=1e-4))
@slow
def test_small_en_logits_librispeech(self):
set_seed(0)
model = TFWhisperModel.from_pretrained("openai/whisper-small.en")
input_speech = self._load_datasamples(1)
feaure_extractor = WhisperFeatureExtractor()
input_features = feaure_extractor(input_speech, return_tensors="tf").input_features
logits = model(
input_features,
decoder_input_ids=tf.convert_to_tensor([[model.config.decoder_start_token_id]]),
output_hidden_states=False,
output_attentions=False,
use_cache=False,
)
logits = logits.last_hidden_state @ tf.transpose(model.model.decoder.embed_tokens.weights[0])
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
-3.6784, -7.7211, -9.5070, -11.9286, -7.6489, -9.7026, -5.6188,
-8.0104, -4.6238, -5.1833, -9.0485, -3.4079, -5.4874, -2.6935,
-6.3479, -7.3398, -6.9558, -7.6867, -7.4748, -8.3463, -9.9781,
-10.8389, -10.3105, -11.7201, -9.7261, -7.1590, -5.9272, -12.4509,
-11.1146, -8.1918
]
)
# fmt: on
self.assertTrue(np.allclose(logits[0, 0, :30], EXPECTED_LOGITS, atol=1e-4))
@slow
def test_large_logits_librispeech(self):
set_seed(0)
model = TFWhisperModel.from_pretrained("openai/whisper-large")
input_speech = self._load_datasamples(1)
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
processed_inputs = processor(audio=input_speech, text="This part of the speech", return_tensors="tf")
input_features = processed_inputs.input_features
labels = processed_inputs.labels
logits = model(
input_features,
decoder_input_ids=labels,
output_hidden_states=False,
output_attentions=False,
use_cache=False,
)
logits = logits.last_hidden_state @ tf.transpose(model.model.decoder.embed_tokens.weights[0])
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
2.1382, 0.9381, 4.4671, 3.5589, 2.4022, 3.8576, -0.6521, 2.5472,
1.8301, 1.9957, 2.3432, 1.4678, 0.5459, 2.2597, 1.5179, 2.5357,
1.1624, 0.6194, 1.0757, 1.8259, 2.4076, 1.6601, 2.3503, 1.3376,
1.9891, 1.8635, 3.8931, 5.3699, 4.4772, 3.9184
]
)
# fmt: on
self.assertTrue(np.allclose(logits[0, 0, :30], EXPECTED_LOGITS, atol=1e-4))
@slow
def test_tiny_en_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
model.config.decoder_start_token_id = 50257
input_speech = self._load_datasamples(1)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
generated_ids = model.generate(input_features, num_beams=5)
transcript = processor.tokenizer.batch_decode(generated_ids)[0]
EXPECTED_TRANSCRIPT = (
"<|startoftranscript|><|notimestamps|> Mr. Quilter is the apostle of the middle"
" classes, and we are glad to"
)
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_tiny_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")
input_speech = self._load_datasamples(1)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
generated_ids = model.generate(input_features, num_beams=5)
transcript = processor.tokenizer.decode(generated_ids[0])
EXPECTED_TRANSCRIPT = (
"<|startoftranscript|><|en|><|transcribe|><|notimestamps|> Mr. Quilter is the apostle of the middle"
" classes and we are glad"
)
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_tiny_xla_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")
input_speech = self._load_datasamples(1)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
xla_generate = tf.function(model.generate, jit_compile=True)
generated_ids = model.generate(input_features, num_beams=5)
generated_ids_xla = xla_generate(input_features, num_beams=5)
transcript = processor.tokenizer.decode(generated_ids[0])
transcript_xla = processor.tokenizer.decode(generated_ids_xla[0])
EXPECTED_TRANSCRIPT = (
"<|startoftranscript|><|en|><|transcribe|><|notimestamps|> Mr. Quilter is the apostle of the middle"
" classes and we are glad"
)
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
self.assertEqual(transcript_xla, EXPECTED_TRANSCRIPT)
@slow
def test_large_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
input_speech = self._load_datasamples(1)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language="en", task="transcribe")
generated_ids = model.generate(
input_features,
do_sample=False,
)
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
EXPECTED_TRANSCRIPT = " Mr. Quilter is the apostle of the middle classes and we are glad"
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_large_generation_multilingual(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
ds = load_dataset("common_voice", "ja", split="test", streaming=True)
ds = ds.cast_column("audio", datasets.Audio(sampling_rate=16_000))
input_speech = next(iter(ds))["audio"]["array"]
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language="ja", task="transcribe")
generated_ids = model.generate(input_features, do_sample=False)
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
EXPECTED_TRANSCRIPT = "木村さんに電話を貸してもらいました"
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language="en", task="transcribe")
generated_ids = model.generate(
input_features,
do_sample=False,
)
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
EXPECTED_TRANSCRIPT = " Kimura san ni denwa wo kaite moraimashita"
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language="ja", task="translate")
generated_ids = model.generate(input_features, do_sample=False)
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
EXPECTED_TRANSCRIPT = " I borrowed a phone from Kimura san"
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_large_batched_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
input_speech = self._load_datasamples(4)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
generated_ids = model.generate(input_features)
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
[50258, 50358, 50363, 2221, 13, 2326, 388, 391, 307, 264, 50244, 295, 264, 2808, 5359, 293, 321, 366, 5404, 281],
[50258, 50358, 50363, 6966, 307, 2221, 13, 2326, 388, 391, 311, 9060, 1570, 1880, 813, 702, 1871, 13, 50257, 50257],
[50258, 50358, 50363, 634, 5112, 505, 300, 412, 341, 42729, 3196, 295, 264, 1064, 11, 365, 5272, 293, 12904, 9256],
[50258, 50358, 50363, 634, 575, 12525, 22618, 1968, 6144, 35617, 20084, 1756, 311, 589, 307, 534, 10281, 934, 439, 11]
]
)
# fmt: on
self.assertTrue(np.allclose(generated_ids, EXPECTED_LOGITS))
# fmt: off
EXPECTED_TRANSCRIPT = [
' Mr. Quilter is the apostle of the middle classes, and we are glad to',
" Nor is Mr. Quilter's manner less interesting than his matter.",
" He tells us that at this festive season of the year, with Christmas and roast beef",
" He has grave doubts whether Sir Frederick Layton's work is really Greek after all,"
]
# fmt: on
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)
self.assertListEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_tiny_en_batched_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
input_speech = self._load_datasamples(4)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
generated_ids = model.generate(input_features)
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
[50257, 50362, 1770, 13, 2264, 346, 353, 318, 262, 46329, 286, 262, 3504, 6097, 11, 290, 356, 389, 9675, 284],
[50257, 50362, 5414, 318, 1770, 13, 2264, 346, 353, 338, 5642, 1342, 3499, 621, 465, 2300, 13, 50256, 50256, 50256],
[50257, 50362, 679, 4952, 514, 326, 379, 428, 43856, 1622, 286, 262, 614, 11, 351, 6786, 290, 32595, 12023, 28236],
[50257, 50362, 679, 468, 12296, 17188, 1771, 7361, 26113, 18881, 1122, 338, 670, 318, 1107, 8312, 706, 477, 290, 460]
]
)
# fmt: on
self.assertTrue(np.allclose(generated_ids, EXPECTED_LOGITS))
# fmt: off
EXPECTED_TRANSCRIPT = [
" Mr. Quilter is the apostle of the middle classes, and we are glad to",
" Nor is Mr. Quilter's manner less interesting than his matter.",
" He tells us that at this festive season of the year, with Christmas and roast beef looming",
" He has grave doubts whether Sir Frederick Layton's work is really Greek after all and can",
]
# fmt: on
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)
self.assertListEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_tiny_en_batched_xla_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
input_speech = self._load_datasamples(4)
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="tf").input_features
xla_generate = tf.function(model.generate, jit_compile=True)
generated_ids = model.generate(input_features)
generated_ids_xla = xla_generate(input_features)
# fmt: off
EXPECTED_LOGITS = tf.convert_to_tensor(
[
[50257, 50362, 1770, 13, 2264, 346, 353, 318, 262, 46329, 286, 262, 3504, 6097, 11, 290, 356, 389, 9675, 284],
[50257, 50362, 5414, 318, 1770, 13, 2264, 346, 353, 338, 5642, 1342, 3499, 621, 465, 2300, 13, 50256, 50256, 50256],
[50257, 50362, 679, 4952, 514, 326, 379, 428, 43856, 1622, 286, 262, 614, 11, 351, 6786, 290, 32595, 12023, 28236],
[50257, 50362, 679, 468, 12296, 17188, 1771, 7361, 26113, 18881, 1122, 338, 670, 318, 1107, 8312, 706, 477, 290, 460]
]
)
# fmt: on
self.assertTrue(np.allclose(generated_ids, EXPECTED_LOGITS))
self.assertTrue(np.allclose(generated_ids_xla, EXPECTED_LOGITS))
# fmt: off
EXPECTED_TRANSCRIPT = [
" Mr. Quilter is the apostle of the middle classes, and we are glad to",
" Nor is Mr. Quilter's manner less interesting than his matter.",
" He tells us that at this festive season of the year, with Christmas and roast beef looming",
" He has grave doubts whether Sir Frederick Layton's work is really Greek after all and can",
]
# fmt: on
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True)
transcript_xla = processor.batch_decode(generated_ids_xla, skip_special_tokens=True)
self.assertListEqual(transcript, EXPECTED_TRANSCRIPT)
self.assertListEqual(transcript_xla, EXPECTED_TRANSCRIPT)

28
tests/test_modeling_tf_common.py
View File

@ -736,6 +736,23 @@ class TFModelTesterMixin:
dtype="float32",
),
}
elif model_class.__name__ in ["TFWhisperModel", "TFWhisperForConditionalGeneration"]:
inputs = {
"decoder_input_ids": tf.keras.Input(
batch_shape=(2, max_input),
name="decoder_input_ids",
dtype="int32",
),
"input_features": tf.keras.Input(
batch_shape=(
2,
self.model_tester.num_mel_bins,
self.model_tester.seq_length,
),
name="input_features",
dtype="float32",
),
}
elif self.is_encoder_decoder:
inputs = {
"decoder_input_ids": tf.keras.Input(
@ -1223,8 +1240,17 @@ class TFModelTesterMixin:
# fetch the output for an input exclusively made of new members of the vocabulary
inputs_dict = copy.deepcopy(original_inputs_dict)
new_vocab_input_ids = ids_tensor(inputs_dict["input_ids"].shape, new_tokens_size)
ids_feat_name = None
if "input_ids" in inputs_dict:
ids_feat_name = "input_ids"
elif "decoder_input_ids" in inputs_dict:
ids_feat_name = "decoder_input_ids"
else:
assert False, "No input ids feature found in the inputs dict"
new_vocab_input_ids = ids_tensor(inputs_dict[ids_feat_name].shape, new_tokens_size)
new_vocab_input_ids += old_total_size
inputs_dict[ids_feat_name] = new_vocab_input_ids
if "input_ids" in inputs_dict:
inputs_dict["input_ids"] = new_vocab_input_ids
if "decoder_input_ids" in inputs_dict:

2
utils/check_repo.py
View File

@ -105,6 +105,8 @@ IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
"TFElectraMainLayer", # Building part of bigger (tested) model (should it be a TFPreTrainedModel ?)
"TFRobertaForMultipleChoice", # TODO: fix
"TrOCRDecoderWrapper", # Building part of bigger (tested) model.
"TFWhisperEncoder", # Building part of bigger (tested) model.
"TFWhisperDecoder", # Building part of bigger (tested) model.
"SeparableConv1D", # Building part of bigger (tested) model.
"FlaxBartForCausalLM", # Building part of bigger (tested) model.
"FlaxBertForCausalLM", # Building part of bigger (tested) model. Tested implicitly through FlaxRobertaForCausalLM.

1
utils/documentation_tests.txt
View File

@ -97,4 +97,5 @@ src/transformers/models/wav2vec2_conformer/modeling_wav2vec2_conformer.py
src/transformers/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.py
src/transformers/models/wavlm/modeling_wavlm.py
src/transformers/models/whisper/modeling_whisper.py
src/transformers/models/whisper/modeling_tf_whisper.py
src/transformers/models/yolos/modeling_yolos.py