1 Release published by 1 person

• v4.47.0 v4.47.0: PaliGemma-2, I-JEPA, OLMo-2, LayerSkip, Tensor Parallel

  published Dec 5, 2024

49 Pull requests merged by 36 people

• docs: clarify initializer_range parameter description in Idefics3VisionConfig

  #35215 merged Dec 11, 2024

• Fix seamless TTS generate

  #34968 merged Dec 11, 2024

• Fix CI

  #35208 merged Dec 11, 2024

• Cleanup: continue the init refactor

  #35170 merged Dec 11, 2024

• Add TimmWrapper

  #34564 merged Dec 11, 2024

• [PEFT] Better Trainer error when prompt learning with loading best model at the end

  #35087 merged Dec 11, 2024

• 🧹 Remove deprecated RotaryEmbedding parts in the Attention layers

  #34858 merged Dec 11, 2024

• BLIP: enable device map

  #34850 merged Dec 11, 2024

• [i18n-<languageCode>] Translating agents.md to Chinese

  #35139 merged Dec 10, 2024

• Update data collator docstrings to accurately reference Nvidia tensor core compute capability version

  #35188 merged Dec 10, 2024

• [docs] Fix FlashAttention link

  #35171 merged Dec 10, 2024

• [i18n-<languageCode>] Translating Benchmarks.md to Chinese

  #35137 merged Dec 10, 2024

• Only import torch.distributed if it is available

  #35133 merged Dec 10, 2024

• Multiple typo fixes in NLP, Audio docs

  #35181 merged Dec 10, 2024

• [i18n-ar] Translated file : docs/source/ar/community.md into Arabic

  #33027 merged Dec 10, 2024

• Fixing GGUF support for StableLm

  #35060 merged Dec 10, 2024

• Fix DBRX LayerNorm init method

  #35177 merged Dec 10, 2024

• Remove unnecessary masked_fill in deberta models

  #35182 merged Dec 10, 2024

• Support BatchNorm in Hubert pos_conv_emb as in fairseq

  #34389 merged Dec 10, 2024

• Fix file path for shard_num 1 with mllama converter

  #35053 merged Dec 10, 2024

• Assisted decoding multi-gpu

  #35116 merged Dec 10, 2024

• Fix num_items_in_batch not being an integer

  #35115 merged Dec 10, 2024

• [CI] Fix bnb quantization tests with accelerate>=1.2.0

  #35172 merged Dec 9, 2024

• Fixed typo of 'avilable' in prompts.py

  #35145 merged Dec 9, 2024

• Super tiny fix logging message

  #35132 merged Dec 9, 2024

• Cleanup: continue the init refactor

  #35167 merged Dec 9, 2024

• Fix typo in EETQ Tests

  #35160 merged Dec 9, 2024

• Option to set 'non_blocking' for to(device) in BatchEncoding and BatchFeature

  #34883 merged Dec 9, 2024

• Corrected typo in agent system prompts

  #35143 merged Dec 9, 2024

• [I-JEPA] Update docs

  #35148 merged Dec 9, 2024

• Fix GA loss bugs and add unit test

  #35121 merged Dec 9, 2024

• Update I-JEPA checkpoints path

  #35120 merged Dec 6, 2024

• Add feature dim attributes to BitLinear for easier PEFT integration

  #34946 merged Dec 6, 2024

• Add Aria

  #34157 merged Dec 6, 2024

• Fix private forked repo. CI

  #35114 merged Dec 6, 2024

• [docs] top_p, top_k, temperature docstrings

  #35065 merged Dec 5, 2024

• [docs] Update Python version in translations

  #35096 merged Dec 5, 2024

• Fix signatures for processing kwargs

  #35105 merged Dec 5, 2024

• Adaptive dynamic number of speculative tokens

  #34156 merged Dec 5, 2024

• Fix flaky Hub CI (test_trainer.py)

  #35062 merged Dec 5, 2024

• [trainer] fix the GA model_accepts_loss_kwargs

  #34915 merged Dec 5, 2024

• BLIP: this is correct now

  #35081 merged Dec 5, 2024

• Add I-JEPA

  #33125 merged Dec 5, 2024

• Deprecate quanto and switch to optimum-quanto

  #35001 merged Dec 5, 2024

• Fix tie_word_embeddings handling for GGUF models

  #35085 merged Dec 5, 2024

• Update Mistral conversion script

  #34829 merged Dec 5, 2024

• [tokenizers] bump to 0.21

  #34972 merged Dec 5, 2024

• [Whisper] Fix whisper tokenizer

  #34537 merged Dec 5, 2024

• Informative

  #35059 merged Dec 5, 2024

48 Pull requests opened by 30 people

• [Idefics3] Move image features to same device as input embeds

  #35100 opened Dec 5, 2024

• Let `EarlyStoppingCallback` not require `load_best_model_at_end`

  #35101 opened Dec 5, 2024

• Add check for if num_items_in_batch is not None

  #35102 opened Dec 5, 2024

• Support Python 3.10+ Union style in chat template type hints parsing

  #35103 opened Dec 5, 2024

• Add dinov2 with registers attempt 2

  #35104 opened Dec 5, 2024

• Fix the structure of images in PixtralProcessor

  #35107 opened Dec 5, 2024

• Keep `image_sizes` in output of `PixtralProcessor`

  #35110 opened Dec 6, 2024

• [WIP] Add flex attention for Qwen2VL

  #35112 opened Dec 6, 2024

• Use `rsfE` with `pytest`

  #35119 opened Dec 6, 2024

• [WIP] Pixtral: vectorize patch embeddings

  #35122 opened Dec 6, 2024

• Add weight norm rename in _load_state_dict_into_model

  #35123 opened Dec 6, 2024

• Add common test for `torch.export` and fix some vision models

  #35124 opened Dec 6, 2024

• logic was inverted for passing loss_kwargs to forward pass

  #35128 opened Dec 6, 2024

• Add compute_loss_func to Seq2SeqTrainer

  #35136 opened Dec 7, 2024

• addressing the issue #34611 to make FlaxDinov2 compatible with any batch size

  #35138 opened Dec 7, 2024

• Samhq model addition

  #35147 opened Dec 8, 2024

• [`Mamba2`] Fix caching, slow path, and multi-gpu

  #35154 opened Dec 8, 2024

• Adding FlexAttention Support for Qwen2 models

  #35155 opened Dec 8, 2024

• Add reminder config to issue template and print DS version in env

  #35156 opened Dec 9, 2024

• don't use no_sync when deepspeed doesn't support it for certain zero stages

  #35157 opened Dec 9, 2024

• Add ModernBERT to Transformers

  #35158 opened Dec 9, 2024

• Use modules_to_not_convert in AQLM

  #35161 opened Dec 9, 2024

• Init cache on meta device

  #35164 opened Dec 9, 2024

• Ascend NPU support SDPA

  #35165 opened Dec 9, 2024

• Add dtype check in check_quantized_param for bnb4

  #35183 opened Dec 10, 2024

• Add compile test for fast image processor

  #35184 opened Dec 10, 2024

• make LlamaModel._update_causal_mask torch compilable

  #35187 opened Dec 10, 2024

• Fix f-string to show `ACCELERATE_MIN_VERSION` on error

  #35189 opened Dec 10, 2024

• Add Flax activation functions and corresponding tests

  #35191 opened Dec 10, 2024

• [i18n-ar] Translated file : docs/source/ar/tasks/sequence_classification.md into Arabic

  #35192 opened Dec 10, 2024

• [i18n-ar] Translated file : docs/source/ar/tasks/token_classification.md into Arabic

  #35193 opened Dec 10, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/translation.md` into Arabic

  #35194 opened Dec 11, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/summarization.md` into Arabic

  #35195 opened Dec 11, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/question_answering.md` into Arabic

  #35196 opened Dec 11, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/language_modeling.md` into Arabic

  #35197 opened Dec 11, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/masked_language_modeling.md` into Arabic

  #35198 opened Dec 11, 2024

• [i18n-ar] Translated file: `docs/source/ar/tasks/multiple_choice.md` into Arabic

  #35199 opened Dec 11, 2024

• PaliGemma: Make sure to add <eos> to suffix if <image> is present in `text`

  #35201 opened Dec 11, 2024

• 🔴 Video processors as a separate class

  #35206 opened Dec 11, 2024

• Scale loss before backward

  #35207 opened Dec 11, 2024

• Cleanup: continue the init refactor

  #35209 opened Dec 11, 2024

• Trigger GitHub CI with a comment on PR

  #35211 opened Dec 11, 2024

• Fix FSDP no longer working

  #35212 opened Dec 11, 2024

• Add retry hf hub decorator

  #35213 opened Dec 11, 2024

• Added RAdamScheduleFree support | updated codecarbon to run test_trainer.py properly

  #35214 opened Dec 11, 2024

• Fix type hints for apply_chat_template

  #35216 opened Dec 11, 2024

• Fix loading with only state dict and low_cpu_mem_usage = True

  #35217 opened Dec 11, 2024

• added cached tokenizer

  #35218 opened Dec 11, 2024

39 Issues closed by 18 people

• How to load local transformers?

  #35118 closed Dec 12, 2024

• `SeamlessM4TForTextToSpeech.generate` not working if `generation_config` is passed

  #34811 closed Dec 11, 2024

• top-p sampling gives different results even after fixing all random seeds

  #34693 closed Dec 11, 2024

• [Idefics3] processing_idefics3 - IndexError: list index out of range for multiple image input

  #34727 closed Dec 11, 2024

• Video-Llava model's generation error due to causal mask shape mismatch

  #34696 closed Dec 11, 2024

• Compile of the Generate function raises ValueError

  #34767 closed Dec 11, 2024

• Error loading model with device_map="auto" for AutoModelForVisualQuestionAnswering in visual-question-answering pipeline

  #34681 closed Dec 11, 2024

• ValueError: Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' 'truncation=True' to have batched tensors with the same length

  #33999 closed Dec 11, 2024

• [i18n-<languageCode>] Translating Benchmarks.md to Chinese

  #35134 closed Dec 11, 2024

• DataCollator documentation references the wrong version of Nvidia GPUs for accelerated training on tensor cores

  #35174 closed Dec 10, 2024

• No module named 'torch._C._distributed_c10d'; 'torch._C' is not a package on darwin

  #35129 closed Dec 10, 2024

• Did you mean masked_fill_ in modeling_deberta.py?

  #35162 closed Dec 10, 2024

• Add support for HuBERT batch norm instead of weight norm in pos_conv_emb

  #34229 closed Dec 10, 2024

• Incorrect hardcoded consolidated.pth path for Llama 3.2 11B Vision+Instruct Model

  #35049 closed Dec 10, 2024

• Running AG and SD when assistant and target models are on different devices

  #35099 closed Dec 10, 2024

• Exception raised with trainer + `accelerate launch` FSDP + large gradient accumulation steps + small dataset

  #33413 closed Dec 10, 2024

• Documentation for HuBERT is Incomplete

  #33536 closed Dec 10, 2024

• CausalLM loss function throws runtime error in multi-gpu setup

  #35086 closed Dec 10, 2024

• Why can't `inputs_embeds` be used during the first generation in a multimodal model?

  #35131 closed Dec 10, 2024

• XLA FSDP V2 + TPU + T5 Family Models doesn't work

  #35142 closed Dec 9, 2024

• Typo in prompts.py: "avilable" should be "available"

  #35144 closed Dec 9, 2024

• Typo in agent system prompts

  #35109 closed Dec 9, 2024

• [2023-12-04 11:52:08,378] [INFO] [autotuner.py:1110:run_after_tuning] No optimal DeepSpeed configuration found by autotuning.

  #27830 closed Dec 8, 2024

• DDP error with load_best_model_at_end enabled

  #30702 closed Dec 8, 2024

• Saving model in safetensors format through Trainer fails for Gemma 2 due to shared tensors

  #33807 closed Dec 8, 2024

• T5Attention forward pass failing when not using KV cache

  #34448 closed Dec 8, 2024

• Flash attention build running forever on colab

  #34466 closed Dec 8, 2024

• Falcon model training on multiple GPUs

  #34492 closed Dec 8, 2024

• num_quantizer in EncodecConfig should accept variable codebook size

  #34521 closed Dec 8, 2024

• Deprecated `shard_checkpoint`'s replacement `save_torch_state_dict` does not save tied embeddings

  #35080 closed Dec 7, 2024

• [Query] Preprocessing Tutorial

  #35111 closed Dec 6, 2024

• who are you

  #35117 closed Dec 6, 2024

• Some Whisper beam search output (sequences_scores, etc.) is lost in _stack_split_outputs

  #32373 closed Dec 6, 2024

• Assert error in convert_llava_onevision_weights_to_hf.py

  #34467 closed Dec 6, 2024

• The document of generation seems to wrongly describe the default value of top_p, top_k and temperature

  #35045 closed Dec 5, 2024

• Mismatched keyword argument names of llama make GA fix invalid

  #34577 closed Dec 5, 2024

• WhisperTokenizer decode is offsetting timestamps incorrectly

  #34472 closed Dec 5, 2024

• Manually setting `device_map` causes a RuntimeError.

  #34456 closed Dec 5, 2024

• When using `save_pretrained` to save a model loaded with `from_pretrained`, its size becomes twice as large.

  #35072 closed Dec 5, 2024

32 Issues opened by 27 people

• Numpy is not available

  #35221 opened Dec 11, 2024

• Models numpy version and version installed after running `pip install transformers[sentencepiece]`

  #35220 opened Dec 11, 2024

• Shape mismatch in RoPE embeddings gpt_neox model when rotary_ndims is odd

  #35219 opened Dec 11, 2024

• run_mlm_flax on tpu v5-pods

  #35205 opened Dec 11, 2024

• logged loss is not correct with gradient accumulation

  #35204 opened Dec 11, 2024

• gradient calculation is not correct with gradient accumulation in Pretrain

  #35203 opened Dec 11, 2024

• Improve tensor parallel memory usage

  #35202 opened Dec 11, 2024

• PaliGemma2 Processor returns wrong labels array when <image> token is present in `text`

  #35200 opened Dec 11, 2024

• How to convert my Mask2Former model (ResNet-50 backbone) to Hugging Face transformer

  #35186 opened Dec 10, 2024

• QuantizedCache first token processing is counterintuitive / worse than in papers

  #35185 opened Dec 10, 2024

• Adding Mamba2ForTokenClassification to Mamba2

  #35180 opened Dec 10, 2024

• More rich documentation on pipelines

  #35179 opened Dec 10, 2024

• Incorrect file structure in convert_mask2former_original_pytorch_checkpoint_to_pytorch.py?

  #35178 opened Dec 10, 2024

• Detokenization discrepancy with Llama3.1

  #35175 opened Dec 9, 2024

• LlavaForConditionalGeneration._merge_input_ids_with_image_features throws error

  #35169 opened Dec 9, 2024

• DynamicCache does not support variable lengths, except for FA2

  #35168 opened Dec 9, 2024

• Mimi model gives different outputs when using batch encode vs single encode

  #35166 opened Dec 9, 2024

• Calling Trainer.create_model_card() with an empty dataset list causes an IndexError

  #35163 opened Dec 9, 2024

• [tests] run one test but got 2 test results

  #35159 opened Dec 9, 2024

• Impossible to change attention implementation

  #35153 opened Dec 8, 2024

• how to load the weight of decoder.embed_tokens.weight seperately from the shared weight?

  #35152 opened Dec 8, 2024

• Qwen2vl float16 inference bug in naive attention

  #35151 opened Dec 8, 2024

• Cuda OOM

  #35150 opened Dec 8, 2024

• RuntimeError: shape '[1, 3098, 6, 5, 128]' is invalid for input of size 12689408

  #35146 opened Dec 7, 2024

• resizing token embeddings causes output embedding to be reinitialized in `post_init` when `tie_word_embedding` is False

  #35141 opened Dec 7, 2024

• FileNotFoundError: Couldn't find a module script at /home/tooko/transformers-course/glue/glue.py.

  #35140 opened Dec 7, 2024

• [i18n-<languageCode>] Translating docs to Chinese Translate agents.md into Chinese

  #35135 opened Dec 7, 2024

• Special token ids are not longer typed properly in 4.47.0

  #35126 opened Dec 6, 2024

• (sort of) a bug with token offsets: some special tokens have (0, 0) offsets regardless of their position in the document

  #35125 opened Dec 6, 2024

• Training config that worked with transformers v4.4.6.3 results in OOM error with v4.47.0 (using SFTTrainer)

  #35108 opened Dec 5, 2024

• Error during training: "Expected dtype float for end but got dtype c10::BFloat16"

  #35106 opened Dec 5, 2024

• TextIteratorStreamer unable to create generator_kwargs

  #35098 opened Dec 5, 2024

102 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Add LightGlue model

  #31718 commented on Dec 9, 2024 • 54 new comments

• Efficient Inference Kernel for SpQR

  #34976 commented on Dec 11, 2024 • 23 new comments

• [Whisper] 🚨 Fix whisper decoding 🚨

  #34135 commented on Dec 11, 2024 • 18 new comments

• Enhanced Installation Section in README.md

  #35094 commented on Dec 5, 2024 • 12 new comments

• Add Zamba2

  #34517 commented on Dec 7, 2024 • 12 new comments

• [WIP] Refactoring of ImageProcessorFast

  #35069 commented on Dec 11, 2024 • 12 new comments

• Run model as compressed/uncompressed mode

  #34719 commented on Dec 11, 2024 • 10 new comments

• HIGGS Quantization Support

  #34997 commented on Dec 9, 2024 • 10 new comments

• [GGUF] Refactor and decouple gguf checkpoint loading logic

  #34385 commented on Dec 10, 2024 • 8 new comments

• Add dithering to the `Speech2TextFeatureExtractor` API.

  #34638 commented on Dec 10, 2024 • 6 new comments

• Universal Speculative Decoding `CandidateGenerator`

  #35029 commented on Dec 5, 2024 • 5 new comments

• Aggeregate test summary files in CircleCI workflow runs

  #34989 commented on Dec 10, 2024 • 4 new comments

• add bnb support for Ascend NPU

  #31512 commented on Dec 11, 2024 • 4 new comments

• Add diffllama

  #34083 commented on Dec 10, 2024 • 4 new comments

• Fix : Falcon processor doesn't account for a layout difference of qkv between transformers and GGUF

  #35088 commented on Dec 10, 2024 • 3 new comments

• Enable different torch dtype in sub models

  #34873 commented on Dec 10, 2024 • 3 new comments

• Fix case of nested tensors in BatchMixFeature

  #35063 commented on Dec 9, 2024 • 2 new comments

• [whisper] added dropping of attention weights after DTW calculations related to word timestamps if these weights are not requested in the output

  #33732 commented on Dec 9, 2024 • 2 new comments

• Add ColPali to 🤗 transformers

  #33736 commented on Dec 10, 2024 • 2 new comments

• Add support for Apple's Depth-Pro

  #34583 commented on Dec 6, 2024 • 2 new comments

• Output dicts support in text generation pipeline

  #35092 commented on Dec 7, 2024 • 2 new comments

• Add TextNet

  #34979 commented on Dec 8, 2024 • 1 new comment

• enable StaticCache for assisted generation

  #34797 commented on Dec 11, 2024 • 0 new comments

• FEAT : Adding VPTQ quantization method to HFQuantizer

  #34770 commented on Dec 6, 2024 • 0 new comments

• Update config validation

  #34726 commented on Dec 9, 2024 • 0 new comments

• Add GOT-OCR 2.0 to Transformers

  #34721 commented on Dec 5, 2024 • 0 new comments

• Add TimesFM Time Series Forecasting Model

  #34082 commented on Dec 10, 2024 • 0 new comments

• change bnb tests

  #34713 commented on Dec 9, 2024 • 0 new comments

• Past Keys Output now working with output router logits

  #34707 commented on Dec 6, 2024 • 0 new comments

• VLMs: major clean up 🧼

  #34502 commented on Dec 10, 2024 • 0 new comments

• LLaVA-NeXT: add new model checkpoints

  #34195 commented on Dec 11, 2024 • 0 new comments

• Modular phi

  #34361 commented on Dec 11, 2024 • 0 new comments

• [FEAT] Compatibility with dduf format from diffusers

  #35093 commented on Dec 11, 2024 • 0 new comments

• [Clean-up] Planned removal of the `max_size` argument

  #35090 commented on Dec 6, 2024 • 0 new comments

• Fix : model used to test ggml conversion of Falcon-7b is incorrect

  #35083 commented on Dec 10, 2024 • 0 new comments

• [setup] migrate setup script to `pyproject.toml` (reland #22539)

  #35077 commented on Dec 6, 2024 • 0 new comments

• Use AMD CI workflow defined in hf-workflows

  #35058 commented on Dec 6, 2024 • 0 new comments

• Add: num_additional_image_tokens to models

  #35052 commented on Dec 9, 2024 • 0 new comments

• Enable gptqmodel

  #35012 commented on Dec 10, 2024 • 0 new comments

• switch from `training_args.bin` `training_args.json`

  #35010 commented on Dec 12, 2024 • 0 new comments

• Refactoring `AssistedCandidateGenerator` for Improved Modularity and Reusability

  #35009 commented on Dec 10, 2024 • 0 new comments

• Make `test_generate_with_static_cache` even less flaky

  #34995 commented on Dec 10, 2024 • 0 new comments

• Deprecate _is_quantized_training_enabled

  #34991 commented on Dec 10, 2024 • 0 new comments

• [ `Core`] Refactor modeling code

  #34987 commented on Dec 11, 2024 • 0 new comments

• Add the Bamba Model

  #34982 commented on Dec 11, 2024 • 0 new comments

• [`ESM`] Add support for sdpa.

  #34954 commented on Dec 9, 2024 • 0 new comments

• Add sdpa for Beit

  #34941 commented on Dec 11, 2024 • 0 new comments

• Implement AsyncTextIteratorStreamer for asynchronous streaming

  #34931 commented on Dec 9, 2024 • 0 new comments

• [tests] fix "Tester object has no attribute '_testMethodName'"

  #34910 commented on Dec 10, 2024 • 0 new comments

• Add Relation DETR

  #34900 commented on Dec 9, 2024 • 0 new comments

• [WIP] Add flex attention for gpt2

  #34861 commented on Dec 6, 2024 • 0 new comments

• Add Flex Attention for Mistral along with refactoring

  #34845 commented on Dec 5, 2024 • 0 new comments

• [`GPTQ`, `CompressedTensors`] Fix unsafe imports and metada check

  #34815 commented on Dec 8, 2024 • 0 new comments

• Adding support for OpenLMForCausalLM from DataComp

  #34081 commented on Dec 5, 2024 • 0 new comments

• The dot in the model name when using auto_map will cause a path parsing error.

  #35082 commented on Dec 10, 2024 • 0 new comments

• Add Flax diverse group search

  #25355 commented on Dec 9, 2024 • 0 new comments

• Resuming from checkpoint runs into OOM

  #30822 commented on Dec 9, 2024 • 0 new comments

• NaN model parameter found in meta-llama/Llama-3.2-11B-Vision under 4.46.1 version

  #34602 commented on Dec 9, 2024 • 0 new comments

• Trying to train a model using automatic1111. Error - Exception training model: 'module 'transformers.integrations' has no attribute 'deepspeed''.

  #34427 commented on Dec 9, 2024 • 0 new comments

• `dataloader_persistent_workers=True` causes fork-bomb due to repeated creation of `eval_dataloader`

  #28469 commented on Dec 9, 2024 • 0 new comments

• bus error on version 4.43.0 with pretrained community CLIP model - MacOS

  #33357 commented on Dec 8, 2024 • 0 new comments

• Accelerate x Trainer issue tracker:

  #33345 commented on Dec 8, 2024 • 0 new comments

• Passing nn.Parameter values within the model architecture as deep copies.

  #34643 commented on Dec 8, 2024 • 0 new comments

• ValueError: Architecture deepseek2 not supported

  #34335 commented on Dec 7, 2024 • 0 new comments

• Does per_device_train_batch_size have a loss error similar to that of GA?

  #34579 commented on Dec 7, 2024 • 0 new comments

• Enhancing Hugging Face Models with Tensor Parallelism for Large-Scale Model Support 🚀

  #32470 commented on Dec 7, 2024 • 0 new comments

• xpu device is not used running pipeline(device_map="auto")

  #31922 commented on Dec 6, 2024 • 0 new comments

• Is there a way to find the earliest version of transformers that has a certain model?

  #35097 commented on Dec 6, 2024 • 0 new comments

• RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)

  #34695 commented on Dec 6, 2024 • 0 new comments

• How to specific customized force_token_ids in whisper

  #34107 commented on Dec 6, 2024 • 0 new comments

• Unexpected output of _flash_attention_forward() for cross attention

  #35032 commented on Dec 6, 2024 • 0 new comments

• Duplicate ZeRo 3 Global Step Checkpoint Saves

  #34534 commented on Dec 6, 2024 • 0 new comments

• ValueError: You are trying to save a non-contiguous tensor in MT5 finetunning

  #34623 commented on Dec 6, 2024 • 0 new comments

• Make it possible to save and evaluate checkpoint on CTRL+C / `KeyboardInterrupt` with Hugging Face Trainer

  #35033 commented on Dec 6, 2024 • 0 new comments

• The same situation as #31377 occurred when using Qwen/Qwen2-VL-7B-Instruct

  #33399 commented on Dec 5, 2024 • 0 new comments

• When extending embeddings, multivariate distribution isn't correctly estimated even when the calculated sigma matrix is symmetric and positive definite

  #35075 commented on Dec 5, 2024 • 0 new comments

• Documentation for SWAG contradicts itself when constructing the first sentence.

  #35095 commented on Dec 5, 2024 • 0 new comments

• AssertionError for Pytorch PiPPy example

  #34600 commented on Dec 5, 2024 • 0 new comments

• Add Molmo (7B-D, 7B-O, 70B)

  #33962 commented on Dec 11, 2024 • 0 new comments

• #33512 handle last element out of range error

  #33625 commented on Dec 9, 2024 • 0 new comments

• [WIP] - Enable speculative decoding with batch size >1

  #32189 commented on Dec 11, 2024 • 0 new comments

• [docs] Redesign

  #31757 commented on Dec 11, 2024 • 0 new comments

• Support Kosmos-2.5

  #31711 commented on Dec 6, 2024 • 0 new comments

• Implement SuperGlue model

  #29886 commented on Dec 6, 2024 • 0 new comments

• Fix model code to accurately convert fairseq wav2vec2 model

  #28250 commented on Dec 12, 2024 • 0 new comments

• Add Model Support for xLSTM

  #27011 commented on Dec 11, 2024 • 0 new comments

• Confusing error message

  #34658 commented on Dec 11, 2024 • 0 new comments

• AutoModelForDepthEstimation/DepthAnythingDepthEstimationHead unexpected behavior in JIT

  #34679 commented on Dec 11, 2024 • 0 new comments

• Discrepancy in Training Loss Behavior with Gradient Accumulation using DeepSpeed

  #34694 commented on Dec 11, 2024 • 0 new comments

• trainer resume from checkpoint，the learning rate is not the same as retraining,learning rate is discontinuous

  #34053 commented on Dec 11, 2024 • 0 new comments

• [i18n-ar] Translating docs to Arabic (العربية)

  #32435 commented on Dec 10, 2024 • 0 new comments

• Verify interpolation of image processors

  #28180 commented on Dec 10, 2024 • 0 new comments

• Bug in running facebook/wav2vec2-xlsr-53-espeak-cv-ft

  #35064 commented on Dec 10, 2024 • 0 new comments

• rework `test_multi_gpu_data_parallel_forward`

  #31087 commented on Dec 10, 2024 • 0 new comments

• Padding error when using Universal Assisted Generation with ASR pipeline

  #34639 commented on Dec 10, 2024 • 0 new comments

• Silent failure in generation parameters

  #33690 commented on Dec 10, 2024 • 0 new comments

• BarkProcessor voice_preset doesn't work

  #34634 commented on Dec 10, 2024 • 0 new comments

• about gradient accumulation

  #34648 commented on Dec 10, 2024 • 0 new comments

• Neftune computation is probably wrong with packed training

  #34659 commented on Dec 10, 2024 • 0 new comments

• FlaxWhisperForConditionalGeneration Out Of Memory Error

  #34668 commented on Dec 10, 2024 • 0 new comments

• Vision Encoder-Decoder fails with LLaMA decoder due to missing cross-attention implementation

  #34674 commented on Dec 10, 2024 • 0 new comments

• tokenizer.json modified after tokenizer.save_pretrained of OLMO models

  #34744 commented on Dec 10, 2024 • 0 new comments