34 Pull requests merged by 22 people

• [`MT5`] Fix CONFIG_MAPPING issue leading it to load umt5 class

  #24678 merged Jul 7, 2023

• Fix integration with Accelerate and failing test

  #24691 merged Jul 6, 2023

• Avoid import `sentencepiece_model_pb2` in `utils.__init__.py`

  #24689 merged Jul 6, 2023

• DeepSpeed/FSDP ckpt saving utils fixes and FSDP training args fixes

  #24591 merged Jul 6, 2023

• Add dropouts to GPT-NeoX

  #24680 merged Jul 6, 2023

• LlamaTokenizer should be picklable

  #24681 merged Jul 6, 2023

• Add Nucleotide Transformer notebooks and restructure notebook list

  #24669 merged Jul 5, 2023

• Fix model referenced and results in documentation. Model mentioned was inaccessible

  #24609 merged Jul 5, 2023

• Unpin `huggingface_hub`

  #24667 merged Jul 5, 2023

• Add is_torch_mps_available function to utils

  #24660 merged Jul 5, 2023

• Fix `VisionTextDualEncoderIntegrationTest`

  #24661 merged Jul 5, 2023

• Fix `EncodecModelTest::test_multi_gpu_data_parallel_forward`

  #24663 merged Jul 5, 2023

• Make warning disappear for remote code in pipelines

  #24603 merged Jul 4, 2023

• Add `finetuned_from` property in the autogenerated model card

  #24528 merged Jul 4, 2023

• Update warning messages reffering to post_process_object_detection

  #24649 merged Jul 4, 2023

• documentation_tests.txt - sort filenames alphabetically

  #24647 merged Jul 4, 2023

• llama fp16 torch.max bug fix

  #24561 merged Jul 4, 2023

• Fix audio feature extractor deps

  #24636 merged Jul 4, 2023

• precompiled_charsmap checking before adding to the normalizers' list for XLNetTokenizerFast conversion.

  #24618 merged Jul 4, 2023

• Generate: force cache with `inputs_embeds` forwarding

  #24639 merged Jul 3, 2023

• Generate: multi-device support for contrastive search

  #24635 merged Jul 3, 2023

• Fix loading dataset docs link in run_translation.py example

  #24594 merged Jul 3, 2023

• Pin `Pillow` for now

  #24633 merged Jul 3, 2023

• [Time-Series] Added blog-post to tips

  #24482 merged Jul 3, 2023

• 🌐 [i18n-KO] Translated `perplexity.mdx` to Korean

  #23850 merged Jul 3, 2023

• [`Umt5`] Add google's umt5 to `transformers`

  #24477 merged Jul 3, 2023

• Limit Pydantic to V1 in dependencies

  #24596 merged Jun 30, 2023

• Use protobuf 4

  #24599 merged Jun 30, 2023

• [several models] improve readability

  #24585 merged Jun 30, 2023

• Speed up TF tests by reducing hidden layer counts

  #24595 merged Jun 30, 2023

• Make (TF) CI faster (test only a random subset of model classes)

  #24592 merged Jun 30, 2023

• Show a warning for missing attention masks when pad_token_id is not None

  #24510 merged Jun 30, 2023

• Udate link to RunHouse hardware setup documentation.

  #24590 merged Jun 30, 2023

• ⚠️⚠️[`T5Tokenize`] Fix T5 family tokenizers⚠️⚠️

  #24565 merged Jun 30, 2023

26 Pull requests opened by 23 people

• [WIP] Add Llama Flax Implementation

  #24587 opened Jun 30, 2023

• 🌐 [i18n-KO] Translated`tasks/document_question_answering.md` to Korean

  #24588 opened Jun 30, 2023

• Add forward methods to quantizer that also computes commitment loss

  #24593 opened Jun 30, 2023

• add link to accelerate doc

  #24601 opened Jun 30, 2023

• translate the English documentation into Chinese

  #24611 opened Jul 1, 2023

• [DOC] Clarify relationshi load_best_model_at_end and save_total_limit

  #24614 opened Jul 1, 2023

• [Patch-t5-tokenizer] Patches the changes on T5 to make sure previous behaviour is still valide for beginning of words

  #24622 opened Jul 2, 2023

• 🌐 [i18n-KO] Translated `model_summary.md` to Korean

  #24625 opened Jul 2, 2023

• Create SECURITY.md

  #24627 opened Jul 2, 2023

• [`MPT`] Add MosaicML's `MPT` model to transformers

  #24629 opened Jul 3, 2023

• [WIP] Add LaVIN

  #24645 opened Jul 4, 2023

• Enable `conversational` pipeline for `GPTSw3Tokenizer`

  #24648 opened Jul 4, 2023

• fixing name position_embeddings to object_queries

  #24652 opened Jul 4, 2023

• Llama: add RoPE scaling

  #24653 opened Jul 4, 2023

• add CFG for .generate()

  #24654 opened Jul 5, 2023

• 🌐 [i18n-KO] Fixed Korean and English `quicktour.md`

  #24664 opened Jul 5, 2023

• Whisper: fix prompted max length

  #24666 opened Jul 5, 2023

• updating _compute_mask_indices fn to work with torch compile

  #24668 opened Jul 5, 2023

• Remove WWT from README

  #24672 opened Jul 5, 2023

• Fix non-deterministic Megatron-LM checkpoint name

  #24674 opened Jul 5, 2023

• [`T5`] Adding model_parallel = False to `T5ForQuestionAnswering` and `MT5ForQuestionAnswering`

  #24684 opened Jul 6, 2023

• 🌐 [i18n-KO] Updated Korean `serialization.md`

  #24686 opened Jul 6, 2023

• [DO NOT MERGE] Test PR for studying #24622

  #24690 opened Jul 6, 2023

• Removing unnecessary `device=device` in modeling_llama.py

  #24696 opened Jul 6, 2023

• Bump scipy from 1.8.0 to 1.10.0 in /examples/research_projects/decision_transformer

  #24699 opened Jul 6, 2023

• Suppress warnings from LUKE for unexpected keys

  #24703 opened Jul 7, 2023

71 Issues closed by 20 people

• Bug on Gather all remaining tensors and put them back on the CPU

  #24391 closed Jul 7, 2023

• When using model.generate, it does not stop at eos_token, but instead continues until the maximum length.

  #23175 closed Jul 6, 2023

• trainer evaluation strucked when using dynamic padding in distributed evaluation

  #23780 closed Jul 6, 2023

• Sliding window for finetuning

  #23835 closed Jul 6, 2023

• Add EMD loss

  #23838 closed Jul 6, 2023

• InstructBlipProcessor not working with load_in_4bit and load_in_8bit

  #24564 closed Jul 6, 2023

• TrainingArguments not working in transformers v 4.30

  #24676 closed Jul 6, 2023

• AssertionError: Dynamo only supports FSDP with use_orig_params=True

  #24641 closed Jul 6, 2023

• Language Modeling on Already Tokenized Data

  #24673 closed Jul 6, 2023

• Gradient clipping is no longer recommended?

  #24677 closed Jul 6, 2023

• Custom vision encoder-decoder problem

  #24679 closed Jul 6, 2023

• Finetuning Whisper with prompts

  #24272 closed Jul 6, 2023

• RuntimeError

  #23815 closed Jul 5, 2023

• AttributeError: EagerTensor object has no attribute 'size'

  #23819 closed Jul 5, 2023

• index out of range in self torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)

  #23822 closed Jul 5, 2023

• IndexError when training with GLUE dataset using pretrained from scratch ELECTRA.

  #23831 closed Jul 5, 2023

• Deepspeed ZeRO2 + Trainer does not resume training after evaluation

  #24313 closed Jul 5, 2023

• LlamaForCausalLM returning prompt without answer

  #24624 closed Jul 5, 2023

• 'eos_token_id' for llama model.generate is not working

  #24644 closed Jul 5, 2023

• TrainingArguments.report_to is not configured as documented

  #24646 closed Jul 4, 2023

• Simplifying Output from Text Classification Pipelines

  #23092 closed Jul 4, 2023

• `torch.compile` is ignored when using DeepSpeed

  #23095 closed Jul 4, 2023

• Getting ValueError: model.shared.weight doesn't have any device set in running a M2M100's-12B model on colab while using with accelerate

  #23294 closed Jul 4, 2023

• Causal language modeling documentation is wrong?

  #23841 closed Jul 4, 2023

• Training siamese (biencoder) based transformer model with gradient checkpointing throws error

  #23801 closed Jul 4, 2023

• object of type 'IterableDataset' has no len()

  #23809 closed Jul 4, 2023

• How to convert flax model to pytorch?

  #23810 closed Jul 4, 2023

• LoRA training adapter_model.bin is 888 bytes always

  #24551 closed Jul 4, 2023

• XLNetTokenizerFast conversion fails with identity normalization in Sentencepiece tokenizer

  #24616 closed Jul 4, 2023

• Datasets in run_translation.py

  #24579 closed Jul 3, 2023

• .generate() supports contrastive-search on multi-device?

  #24634 closed Jul 3, 2023

• My QUESTION is how run a very big model like bloom on a cluster of machines ?

  #23761 closed Jul 3, 2023

• Trainer do model generation during evaluation loop

  #23763 closed Jul 3, 2023

• Convert Pre-LN Transformers into equivalent Pre-RMSNorm Transformers to accelerate inference and training

  #23786 closed Jul 3, 2023

• [Bloom] Inconsistent results when testing pretrained model bloom under different dtypes(float16, float32)

  #23791 closed Jul 3, 2023

• Transformer trainer training crashed with GLM models

  #23794 closed Jul 3, 2023

• Type hinting Inconsistency in beam_search.py

  #22856 closed Jul 3, 2023

• Trainer的使用问题

  #24626 closed Jul 3, 2023

• Hi

  #24623 closed Jul 3, 2023

• Problem with Deepspeed integration

  #24438 closed Jul 2, 2023

• no dependency package `accelerate` installed when we install transformers v4.29.1

  #23323 closed Jul 2, 2023

• BertTokenizer.save_vocabulary does not save the full vocab

  #23743 closed Jul 2, 2023

• ImportError: cannot import name 'PartialState' from 'transformers.trainer_pt_utils'

  #23744 closed Jul 2, 2023

• Seems some Bart models from facebook are removed

  #24617 closed Jul 2, 2023

• BART is not found 404

  #24620 closed Jul 2, 2023

• tokenizer = AutoTokenizer.from_pretrained('distilroberta-base') report error

  #24586 closed Jul 1, 2023

• OneFormerImageProcessor does not support passing local config file, always tries to download from repo

  #23116 closed Jul 1, 2023

• llama model can't generate EOS

  #23230 closed Jul 1, 2023

• Save checkpoint asynchronously on cpu to keep GPU training going

  #23419 closed Jul 1, 2023

• Loading LLaMA hf format from local folder is not using GPU in Google Colab

  #23425 closed Jul 1, 2023

• graphormer.collating_graphormer.preprocess_item TypeError: Cannot cast array data from dtype('int64') to dtype('int32') according to the rule 'safe'

  #23474 closed Jul 1, 2023

• `run_mlm.py` doesn't log perplexity to `wandb`

  #23593 closed Jul 1, 2023

• Unable to download Google/vit-base-patch-16-224 / Getting 404 tepo not found error

  #23696 closed Jul 1, 2023

• Help with TrOCR training for Spanish

  #23700 closed Jul 1, 2023

• Can not using LION optimizer

  #23727 closed Jul 1, 2023

• High memory usage for BigBirdForPreTraining

  #23733 closed Jul 1, 2023

• Is it possible to use the transformers library with models, e.g. t5-small, commercially?

  #23734 closed Jul 1, 2023

• PreTrainedTokenizerFast - whitespace merge skipped

  #24610 closed Jul 1, 2023

• Protobuf 4 support (again)

  #24323 closed Jul 1, 2023

• RuntimeError: Could not infer dtype of NoneType

  #24606 closed Jul 1, 2023

• Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:3!

  #24410 closed Jun 30, 2023

• [BUG] Protobuf not being correctly installed

  #24533 closed Jun 30, 2023

• Add MQTTS

  #24142 closed Jun 30, 2023

• TransformerEngine FP8 inference

  #23660 closed Jun 30, 2023

• AutoTokenizer Encode Error

  #23671 closed Jun 30, 2023

• About Tokenizer

  #23676 closed Jun 30, 2023

• How to check word ids for BartTokenizer?

  #23679 closed Jun 30, 2023

• [RWKV] Inference memory leak unless use_cache=False is specified

  #23687 closed Jun 30, 2023

• Token Alignment

  #23692 closed Jun 30, 2023

• ZeRO 3 error: expected the next 4 parameters in the parameter fetch queue to be ... but got ()

  #23693 closed Jun 30, 2023

• Add warning message if model uses `input_ids` that include padding tokens, but no `attention_mask` is provided.

  #16136 closed Jun 30, 2023

45 Issues opened by 45 people

• bf16 with DeepSpeed stage 3 with CPU offload breaks LLaMA 13b+ training

  #24702 opened Jul 7, 2023

• In RWForCausalLM.prepare_inputs_for_generation, the past_key_values are always None.

  #24701 opened Jul 7, 2023

• Pix2StructImageProcessor does not accept list of PIL Images

  #24700 opened Jul 7, 2023

• Assertion `srcIndex < srcSelectDimSize` failed

  #24698 opened Jul 6, 2023

• `Trainer` class on Mac uses `accelerate` to incorrectly set MPS device

  #24697 opened Jul 6, 2023

• Time Series Transformer - Dynamic Categorical Features

  #24695 opened Jul 6, 2023

• Make correct padding for text generation with GPT-NEO

  #24694 opened Jul 6, 2023

• TF : tensor mismatch error in training with opus100 and t5-small

  #24693 opened Jul 6, 2023

• Breaking change in upcoming PyTorch version for weight norm and loading pretrained models

  #24692 opened Jul 6, 2023

• is there any plan to add falcon to instructblip?

  #24688 opened Jul 6, 2023

• OSError: Error no file named pytorch_model.bin

  #24687 opened Jul 6, 2023

• How to get the last 4 Hidden states from the feature extraction pipeline

  #24685 opened Jul 6, 2023

• Model checkpoint twice as large when saved with safetensors

  #24683 opened Jul 6, 2023

• Unable to use Trainer with T5ForQuestionAnswering

  #24682 opened Jul 6, 2023

• Is there any plan to add kosmos-2 to the transformers.

  #24671 opened Jul 5, 2023

• Unable to Get Decoded Output from Whisper

  #24670 opened Jul 5, 2023

• Add ELECTRA/DeBERTa v3 pretraining script (replaced token detection pretraining)

  #24665 opened Jul 5, 2023

• Loading mT5 checkpoint will load from UMT5 class

  #24662 opened Jul 5, 2023

• Add HyenaDNA model

  #24659 opened Jul 5, 2023

• CUDA error: out of memory with zero3 offload

  #24658 opened Jul 5, 2023

• At least one model's inference seems to have broken from transformers 4.29.2 -> 4.30.*

  #24657 opened Jul 5, 2023

• discontinuity learning rate while resume from checkpoint

  #24656 opened Jul 5, 2023

• Add a mechanism to transform the forward pass on Flax models

  #24655 opened Jul 5, 2023

• CLIP pooling is not compatible with adding new tokens

  #24650 opened Jul 4, 2023

• "RuntimeError: 'weight' must be 2-D" training with DeepSpeed

  #24643 opened Jul 4, 2023

• openlm-research/open_llama_13b_easylm cannot be downloaded

  #24642 opened Jul 4, 2023

• 'DummyOptim' object has no attribute 'step'

  #24640 opened Jul 3, 2023

• attention weight clipping

  #24638 opened Jul 3, 2023

• TFOPTForCausalLM Attention mask size mismatch exception

  #24637 opened Jul 3, 2023

• TrOCRProcessor.from_pretrained raise KeyError(key)

  #24632 opened Jul 3, 2023

• Fine tunning Bloom model - Failed to import transformers.training_args

  #24631 opened Jul 3, 2023

• Loading GPT-Neo-2.7B has error

  #24630 opened Jul 3, 2023

• [i18n-<languageCode>] Translating docs to <languageName>

  #24628 opened Jul 3, 2023

• AutoTokenizer always tries to download from hub even if the model is cached. Thus it fails to run when running in an docker environment without SSL.

  #24619 opened Jul 1, 2023

• Cannot load BART model

  #24615 opened Jul 1, 2023

• Fine-tune T5 on SQuAD

  #24613 opened Jul 1, 2023

• ValueError: An instance of tokenizer class BioGptTokenizer cannot be converted in a Fast tokenizer instance. No converter was found.

  #24612 opened Jul 1, 2023

• CUDA error: an illegal memory access was encountered

  #24608 opened Jul 1, 2023

• `logging_dir` is not being generated.

  #24607 opened Jul 1, 2023

• Contradictory information in documentation about the ability to push qunatized models to hub

  #24604 opened Jun 30, 2023

• Support gradient checkpointing for ESM models

  #24602 opened Jun 30, 2023

• IndexError: index -1 is out of bounds for dimension 1 with size 0

  #24600 opened Jun 30, 2023

• Falcon-40b-instruct on Runpod

  #24598 opened Jun 30, 2023

• Test with Pydantic V2

  #24597 opened Jun 30, 2023

• stuck in the evaluation_loop of trainer.py when training

  #24589 opened Jun 30, 2023

120 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.

• Falcon port

  #24523 commented on Jul 6, 2023 • 105 new comments

• Add Pop2Piano

  #21785 commented on Jul 6, 2023 • 83 new comments

• add VITS model

  #24085 commented on Jul 6, 2023 • 71 new comments

• Add bark

  #24086 commented on Jul 6, 2023 • 66 new comments

• Add Classifier-Free Guidance sampling

  #24536 commented on Jul 6, 2023 • 41 new comments

• Add UDOP

  #22940 commented on Jul 5, 2023 • 17 new comments

• Add Image Completion Transformer (ICT)

  #21990 commented on Jul 4, 2023 • 14 new comments

• Add image object

  #24577 commented on Jul 3, 2023 • 9 new comments

• Add Compact Convolutional Transformer model (CCT)

  #24507 commented on Jul 6, 2023 • 7 new comments

• 📓 Text Generation docs rework

  #24575 commented on Jul 6, 2023 • 6 new comments

• Add Multi Resolution Analysis (MRA) (New PR)

  #24513 commented on Jul 6, 2023 • 6 new comments

• to is not supported for `8-bit` models

  #23336 commented on Jul 6, 2023 • 5 new comments

• FP-16 training producing nans on t5-large/flan-t5-xl

  #23918 commented on Jul 6, 2023 • 5 new comments

• [Feature Request] Add timestamp prediction for TF Whisper

  #23928 commented on Jul 6, 2023 • 5 new comments

• Contrastive Search peak memory reduction

  #24120 commented on Jul 6, 2023 • 5 new comments

• LlamaTokenizer: Slow implementation opts for whitespace-lead token (different from fast)

  #24569 commented on Jul 4, 2023 • 4 new comments

• Add ViViT

  #22518 commented on Jun 30, 2023 • 4 new comments

• Separate kwargs of tokenizer and feature_extractor in `ClapProcessor`

  #24503 commented on Jul 3, 2023 • 4 new comments

• Failed to import transformers.pipelines because of the following error (look up to see its traceback): cannot import name 'PartialState' from 'accelerate'

  #23340 commented on Jul 3, 2023 • 3 new comments

• Whisper with Elastic Weight Consolidation

  #23880 commented on Jul 3, 2023 • 3 new comments

• TF Swiftformer

  #22771 commented on Jul 6, 2023 • 3 new comments

• Cannot reproduce results for Pix2struct on InfographicVQA

  #23877 commented on Jul 6, 2023 • 3 new comments

• [WIP]Add TF BEiT Implementation

  #18559 commented on Jul 6, 2023 • 3 new comments

• Add DINOv2

  #24016 commented on Jul 6, 2023 • 3 new comments

• [InstructBLIP] Fix bos token of LLaMa checkpoints

  #24492 commented on Jul 6, 2023 • 3 new comments

• MT5 data padding not working

  #24567 commented on Jun 30, 2023 • 2 new comments

• QLoRA Training does not give expected results

  #24212 commented on Jun 30, 2023 • 2 new comments

• Regression Models

  #23189 commented on Jul 1, 2023 • 2 new comments

• `.to_dict` does not correctly serialize `torch.dtype` in some cases (e.g., vision models)

  #23876 commented on Jul 2, 2023 • 2 new comments

• Get output_hidden_state and output_scores from Flax whisper model

  #22612 commented on Jul 3, 2023 • 2 new comments

• `prompt_ids` does not seem to work with `repetition_penalty`

  #23951 commented on Jul 3, 2023 • 2 new comments

• Add support for BLIP and GIT in image-to-text and VQA pipelines

  #21110 commented on Jul 5, 2023 • 2 new comments

• [WIP] Add CLIPViP

  #23802 commented on Jul 6, 2023 • 2 new comments

• Addition of test code for GPTNeoX Flax support

  #24002 commented on Jul 5, 2023 • 2 new comments

• Add Flax diverse group search

  #24508 commented on Jun 30, 2023 • 2 new comments

• 🌐 [i18n-KO] Translated `custom_tools.mdx` to Korean

  #24580 commented on Jul 3, 2023 • 2 new comments

• Model resources contribution

  #20055 commented on Jun 30, 2023 • 1 new comment

• Open AI GPT Model Implementation in Flax

  #22647 commented on Jun 30, 2023 • 1 new comment

• Add training support for EnCodec

  #24295 commented on Jun 30, 2023 • 1 new comment

• fsdp support bool type in trainArgs, use len(args.fsdp) would evoke TypeError when set fsdp=True

  #24584 commented on Jun 30, 2023 • 1 new comment

• [XLMModel, FlaubertModel] Compatibility with torch make_fx

  #23907 commented on Jun 30, 2023 • 1 new comment

• OSError: When i am loading pszemraj\flan-t5-large-grammar-synthesis from hugging face

  #23902 commented on Jun 30, 2023 • 1 new comment

• MarkupLM: feature_extraction_markuplm.py only extracts SearchableText

  #23887 commented on Jun 30, 2023 • 1 new comment

• MarkupLM: TypeError: unsupported operand type(s) for +: 'Tensor' and 'NoneType' in modeling_markuplm.py", line 217

  #23886 commented on Jun 30, 2023 • 1 new comment

• distillation training for arabic langauge

  #23879 commented on Jun 30, 2023 • 1 new comment

• importing of transformers 4.29.2 slows down PyToch DataLoader's multi-processing significantly

  #23870 commented on Jun 30, 2023 • 1 new comment

• forced_decoder_ids in Whisper models significantly impacts performance, use decoder_input_ids instead

  #23845 commented on Jun 30, 2023 • 1 new comment

• RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation

  #22994 commented on Jun 30, 2023 • 1 new comment

• GPU memory not completely free after one BERT/RoBERTa fwd pass

  #23931 commented on Jul 1, 2023 • 1 new comment

• 4.29.0 bug

  #23927 commented on Jul 1, 2023 • 1 new comment

• AttributeError: 'Wav2Vec2Processor' object has no attribute 'set_lang'

  #23925 commented on Jul 1, 2023 • 1 new comment

• Training "microsoft/beit-large-finetuned-ade-640-640" on my dataset present some worry can not to solve

  #23924 commented on Jul 1, 2023 • 1 new comment

• Adding support for 3D deep learning models.

  #23923 commented on Jul 1, 2023 • 1 new comment

• I want to use 'from_ Pretrained' to read the '.safetensors' model file. What should I do?

  #23176 commented on Jul 1, 2023 • 1 new comment

• Name Error: "Partial State" is not defind

  #22816 commented on Jul 1, 2023 • 1 new comment

• Two tokenizer initialization methods result in inconsistent segmentation results for special words

  #23930 commented on Jul 2, 2023 • 1 new comment

• https://huggingface.co/sentence-transformers/clip-ViT-B-32 license?

  #23956 commented on Jul 2, 2023 • 1 new comment

• Got errors when calling AutoModelForCausalLM related APIs

  #23950 commented on Jul 2, 2023 • 1 new comment

• Unable to load from pretrained by inputting state_dict directly

  #23947 commented on Jul 2, 2023 • 1 new comment

• Dataset load train data have values but when training got num_examples 0

  #23893 commented on Jul 2, 2023 • 1 new comment

• How to generate one token after the other when no past_key_values is returned?

  #23639 commented on Jul 2, 2023 • 1 new comment

• libssl.so.10: cannot open shared object file: No such file or directory

  #21805 commented on Jul 2, 2023 • 1 new comment

• Adding GPTNeoX (Tensorflow version)

  #23814 commented on Jul 2, 2023 • 1 new comment

• Improve EncoderDecoderModel docs

  #16135 commented on Jul 3, 2023 • 1 new comment

• DeepSpeed ZeRO stage3+huggyllama/llama: RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)

  #24468 commented on Jul 3, 2023 • 1 new comment

• Inference API takes forever and output: "Model ... is currently loading"

  #23793 commented on Jul 3, 2023 • 1 new comment

• Inconsistent issue in multi gpu training single machine

  #22511 commented on Jul 3, 2023 • 1 new comment

• load_in_8bit=True returns gibberish when inferencing on multi GPU

  #23989 commented on Jul 4, 2023 • 1 new comment

• Bug Report - Tokenizer Issue with Tensor Device Assignment in transformers/pipelines/text_generation.py

  #23988 commented on Jul 4, 2023 • 1 new comment

• How can I using deepspeed along with LION optimizer/.?

  #23987 commented on Jul 4, 2023 • 1 new comment

• Improve PreTrainedModel.from_pretrained return type

  #23980 commented on Jul 4, 2023 • 1 new comment

• Incorrect handling of EOS tokens in DataCollatorForLanguageModeling when pad_token is set to eos_token

  #23530 commented on Jul 4, 2023 • 1 new comment

• Implement QFormer for pretrain

  #22645 commented on Jul 4, 2023 • 1 new comment

• VideoMAEForVideoClassification does not support `device_map='auto'` yet.

  #23086 commented on Jul 4, 2023 • 1 new comment

• Handle `g_state` in RWKV's customized CUDA kernel to overcome sequence length limitation

  #23979 commented on Jul 4, 2023 • 1 new comment

• Unexpect behaviour

  #24006 commented on Jul 5, 2023 • 1 new comment

• Zero-shot classification pipeline does not appear to batch examples

  #24005 commented on Jul 5, 2023 • 1 new comment

• Use python generator instead of streamer for generation

  #23640 commented on Jul 5, 2023 • 1 new comment

• MaskFormerSwin shows as unsupported the index

  #22948 commented on Jul 5, 2023 • 1 new comment

• PhraseConstraints apearing only directly after input or at the end of the generated sentence

  #19070 commented on Jul 5, 2023 • 1 new comment

• VideoMAE pretraining error when customizing compute_metrics

  #24474 commented on Jul 6, 2023 • 1 new comment

• How to use LogitsWarper within .generate()?

  #24021 commented on Jul 6, 2023 • 1 new comment

• Parameter: encoder_no_repeat_ngram_size or something that makes model not repeat input tokens in the output.

  #23834 commented on Jul 6, 2023 • 1 new comment

• sequences and scores dimensions are mismatch when using generate()

  #23750 commented on Jul 6, 2023 • 1 new comment

• lr_scheduler not updated when auto_find_batch_size set to True and batch_size decays

  #21521 commented on Jul 6, 2023 • 1 new comment

• Question about resum_from_checkpoint in run_translation_no_trainer.py

  #23213 commented on Jul 6, 2023 • 1 new comment

• set fsdp and bf16 don't save memory

  #22821 commented on Jul 6, 2023 • 1 new comment

• Add `MegatronT5ForConditionalGeneration`

  #22317 commented on Jul 2, 2023 • 1 new comment

• Add support of output_scores to flax models

  #22700 commented on Jun 30, 2023 • 1 new comment

• Changed "perplexity" to "eval_perplexity"

  #23875 commented on Jun 30, 2023 • 1 new comment

• [`bnb`] Fix blip2 4bit

  #23895 commented on Jul 6, 2023 • 1 new comment

• fixed direct input of state dict functionality

  #23948 commented on Jul 2, 2023 • 1 new comment

• Improve typing for bitsandbytes util

  #23981 commented on Jul 4, 2023 • 1 new comment

• Add overloads for from_pretrained and from_dict on PretrainedConfig

  #23983 commented on Jul 4, 2023 • 1 new comment

• Create overload signatures for cached_file

  #23984 commented on Jul 4, 2023 • 1 new comment

• Add none check when instantiating tokenizer from auto

  #24019 commented on Jul 6, 2023 • 1 new comment

• Fixes all hidden states output in FlaxT5

  #24027 commented on Jul 6, 2023 • 1 new comment

• Add overloads for PretrainedModel.from_pretrained

  #24035 commented on Jul 6, 2023 • 1 new comment

• RWKV can't stop correctly.

  #23852 commented on Jun 30, 2023 • 0 new comments

• Please fix Lora model resume in transformers when using DeepSpeed

  #23881 commented on Jun 30, 2023 • 0 new comments

• CLIP image processor fails when resizing a 1x1 image

  #21638 commented on Jun 30, 2023 • 0 new comments

• Function infer_channel_dimension_format has a bug

  #21981 commented on Jun 30, 2023 • 0 new comments

• HF CLIP image features different from OpenAI CLIP image features

  #22505 commented on Jun 30, 2023 • 0 new comments

• Going above version 4.21.3 gives UnicodeDecodeError

  #22675 commented on Jun 30, 2023 • 0 new comments

• OwlVit gives different results compared to original colab version

  #21206 commented on Jun 30, 2023 • 0 new comments

• 🌐 [i18n-KO] Translating docs to Korean

  #20179 commented on Jul 1, 2023 • 0 new comments

• MPT

  #23174 commented on Jul 3, 2023 • 0 new comments

• Add DINOv2 to Transformers

  #23739 commented on Jul 3, 2023 • 0 new comments

• Error projecting concatenated Fourier Features.

  #22769 commented on Jul 3, 2023 • 0 new comments

• Model outputs are impacted by the aspect ratios of other images in a batch

  #23218 commented on Jul 3, 2023 • 0 new comments

• _keys_to_ignore_on_load_unexpected not working with GPT2 model

  #24581 commented on Jul 4, 2023 • 0 new comments

• [Bug] Failiure to generate Diffusion images / AI language responses when upgrading past 4.19.2

  #23238 commented on Jul 4, 2023 • 0 new comments

• OneFormer processor does not return correctly formatted class_labels tensors

  #23058 commented on Jul 4, 2023 • 0 new comments

• RWKV - Inference NF4 quantization broken, also Int8 quantization weirdness.

  #23848 commented on Jul 6, 2023 • 0 new comments

• save_pretrained 4-bit models with bitsandbytes

  #23904 commented on Jul 6, 2023 • 0 new comments

• 4bit Blip2 compatibility

  #23839 commented on Jul 6, 2023 • 0 new comments

• Add keypoint-detection task

  #24044 commented on Jul 6, 2023 • 0 new comments

• Add accelerate support - vision MAE models

  #23114 commented on Jun 30, 2023 • 0 new comments

• [`config.to_dict()`] update test and models that have a composition

  #23888 commented on Jul 2, 2023 • 0 new comments

• owl-vit-eval-postprocessor

  #23982 commented on Jul 4, 2023 • 0 new comments