huggingface/peft issues and pull requests

#1058 - Numerical instabilities with LoHa

Issue - State: closed - Opened by BenjaminBossan 11 months ago - 4 comments

#1057 - Forward Methods seems weird

Issue - State: closed - Opened by KimJaehee0725 11 months ago

#1056 - fix backward error in stack llama2 DPO example if checkpointing is used

Pull Request - State: closed - Opened by sywangyi 11 months ago - 18 comments

#1055 - Fine tune models in the exl2 format

Issue - State: closed - Opened by psych0v0yager 11 months ago - 3 comments

#1054 - [DOCS] README.md

Pull Request - State: open - Opened by Akash190104 11 months ago - 2 comments

#1053 - Prompt tuning: Allow to pass additional args to AutoTokenizer.from_pretrained

Pull Request - State: closed - Opened by BenjaminBossan 11 months ago - 1 comment

#1052 - RuntimeError: "svd_cuda_gesvdj" not implemented for 'BFloat16'

Issue - State: closed - Opened by LuciferianInk 11 months ago - 2 comments

#1051 - FIX: Setting active adapter correctly

Pull Request - State: closed - Opened by BenjaminBossan 11 months ago - 1 comment

#1050 - [`tests`] Update Dockerfile to use cuda 12.2

Pull Request - State: closed - Opened by younesbelkada 11 months ago - 2 comments

#1049 - Fix P-tuning for sequence classification docs

Pull Request - State: closed - Opened by ehcalabres 11 months ago - 1 comment

#1048 - Question about GPU Memory Usage with LoRA in VisionEncoderDecoderModel

Issue - State: closed - Opened by eclickECNU 11 months ago - 4 comments

#1047 - docs(README): bit misspell current path link StackLLaMa

Pull Request - State: closed - Opened by guspan-tanadi 11 months ago - 1 comment

#1046 - Fix target_modules type in config.from_pretrained

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago

#1045 - add_weighted_adapter() is unusable, throws error: "Invalid type <class 'list'> found in target_modules"

Issue - State: closed - Opened by LuciferianInk 12 months ago - 6 comments

#1044 - Peft is not working with stabilityai/stablelm-3b-4e1t

Issue - State: closed - Opened by apcameron 12 months ago - 3 comments

#1043 - Merge and unload changes inference a lot on quantized llama 2

Issue - State: closed - Opened by Lanzelot-Moll 12 months ago - 9 comments

#1042 - Load PeftModel in another PretrainedModel

Issue - State: closed - Opened by 648652443 12 months ago - 10 comments

#1041 - fix loading pretrained config

Pull Request - State: closed - Opened by pacman100 12 months ago - 1 comment

#1040 - Multiple Adapters merging question

Issue - State: closed - Opened by AlbertoZerbinati 12 months ago - 5 comments

#1039 - [WIP] Adds Vera (Vector Based Random Matrix Adaption)

Pull Request - State: closed - Opened by vvvm23 12 months ago - 59 comments

#1038 - Add VeRA (Vector Based Random Matrix Adaption)

Issue - State: closed - Opened by vvvm23 12 months ago - 4 comments
Labels: PRs welcome to address this

#1037 - Conv2d.forward() takes 2 positional arguments but 3 were given

Issue - State: closed - Opened by hoang1007 12 months ago - 9 comments

#1036 - [`core`] Fix `use_reentrant` issues

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 7 comments

#1035 - Quality of inference severely changed after merge_and_unload

Issue - State: closed - Opened by blaze7451 12 months ago - 5 comments

#1034 - [`LoRA`] Raise error when adapter name not found in `set_scale`

Pull Request - State: closed - Opened by younesbelkada 12 months ago

#1033 - Add save_function parameter to peftmodel.save_pretrained(). (Fix #939)

Pull Request - State: closed - Opened by Imagium719 12 months ago - 4 comments

#1032 - about some small bug of prompt_tuning.py

Issue - State: closed - Opened by duxiaoyu555 12 months ago

#1031 - Add testing for regex matching and other custom kwargs

Pull Request - State: closed - Opened by SumanthRH 12 months ago - 4 comments

#1030 - 'PeftModelForCausalLM' object has no attribute 'merge_and_unload' when attempting to merge (BloomForCausalLM)

Issue - State: closed - Opened by gmanlan 12 months ago - 4 comments

#1029 - [`LoRA`] Revert original behavior for scale / unscale

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 1 comment

#1028 - Enhance `scaling`/`unscaling` methods of lora/loha

Pull Request - State: closed - Opened by pacman100 12 months ago - 1 comment

#1027 - fix lora scaling and unscaling

Pull Request - State: closed - Opened by pacman100 12 months ago - 1 comment

#1026 - using peft with autogptq and gpt2 model causes "RuntimeError: self and mat2 must have the same dtype"

Issue - State: closed - Opened by drimeF0 12 months ago - 3 comments

#1025 - Value Error while load_adapter

Issue - State: closed - Opened by Ali1858 12 months ago - 7 comments

#1024 - get_peft_model doesn't work when both target_modules and layers_to_transform mentioned

Issue - State: closed - Opened by datta-nimmaturi 12 months ago - 3 comments

#1023 - ESM-V2 + QLoRa: high memory usage

Issue - State: closed - Opened by mahdip72 12 months ago - 18 comments

#1022 - Can I use p-tuning and LoRA jointly in PEFT ?

Issue - State: closed - Opened by Eikor 12 months ago - 1 comment

#1021 - Fixed wrong construction of LoHa weights, updated adapters conversion script

Pull Request - State: closed - Opened by kovalexal 12 months ago - 3 comments

#1020 - print_trainable_parameters(model) gives different number from foundational pertained model

Issue - State: closed - Opened by tom192180 12 months ago - 2 comments

#1017 - FEAT: Add fp16 + cpu merge support

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 2 comments

#1016 - FIX stale.py uses timezone-aware datetime

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 1 comment

#1015 - TST: Cover checking requires_grad for Conv1D & Conv1d for LoRA & IA³

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 3 comments

#1014 - Fix Conv1D merge error for IA3

Pull Request - State: closed - Opened by SumanthRH 12 months ago - 4 comments

#1013 - Creating a conv2d LoRA layer always sets the bias, even if original layer had none

Issue - State: closed - Opened by BenjaminBossan 12 months ago - 10 comments

#1012 - FIX Don't assume model_config contains the key model_type

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 1 comment

#1011 - Create_or_update_model_card

Issue - State: closed - Opened by eliepark 12 months ago - 2 comments

#1010 - previous_dtype is now inferred from F.linear's result output type.

Pull Request - State: closed - Opened by MFajcik 12 months ago - 19 comments

#1009 - [`core` / `LoRA`] Add `safe_merge` to bnb layers

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 2 comments

#1008 - Merge function in Linear4bit class wasn't updated

Issue - State: closed - Opened by khiempm209 12 months ago - 2 comments

#1007 - Fix IA3 config for Falcon models

Pull Request - State: closed - Opened by SumanthRH 12 months ago - 3 comments

#1006 - Ability to switch between different adapter types and mix different adapter types

Issue - State: closed - Opened by kovalexal 12 months ago - 5 comments

#1005 - Support Multi-LoRA/qLoRA in PEFT

Issue - State: closed - Opened by merlintang 12 months ago - 7 comments

#1004 - Prompt tuning for causal language modeling tutorial got wrong prediction

Issue - State: closed - Opened by fancyerii 12 months ago - 2 comments

#1003 - eos_token not being encoded the same after finetuning and hence not being detected at inference time

Issue - State: closed - Opened by iliasmiraoui 12 months ago - 9 comments

#1002 - TST: Comment out flaky LoHA test

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 3 comments

#1001 - FEAT: Add `safe_merge` option in `merge`

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 8 comments

#1000 - Fix word_embeddings match for deepspeed wrapped model

Pull Request - State: closed - Opened by mayank31398 12 months ago - 3 comments

#998 - What I am doing wrong - load_adapter can't find key

Issue - State: closed - Opened by FartyPants 12 months ago - 1 comment

#996 - Fix BigCode models not working with prompt tuning

Issue - State: closed - Opened by mayank31398 12 months ago - 2 comments

#995 - TST: Add regression tests

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 3 comments

#994 - ENH: Refactor LoRA bnb layers for faster initialization

Pull Request - State: closed - Opened by BenjaminBossan 12 months ago - 1 comment

#993 - Fix lora creation

Pull Request - State: closed - Opened by pacman100 12 months ago - 1 comment

#992 - [`core`/ DO NOT MERGE] Attempts to fix multi-adapter inference through `PeftModel` interface

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 3 comments

#991 - [`core` / FEAT] Allow loading from meta device

Pull Request - State: closed - Opened by younesbelkada 12 months ago - 4 comments

#990 - Loading multiple adapters fails when quantized

Issue - State: closed - Opened by bhperry 12 months ago - 3 comments

#989 - update BibTeX

Pull Request - State: closed - Opened by pacman100 almost 1 year ago - 1 comment

#988 - the test of get_peft_model_state_dict

Issue - State: closed - Opened by dzonesasaki about 1 year ago - 4 comments

#987 - Weight Tying

Issue - State: open - Opened by thistleknot about 1 year ago - 4 comments

#986 - QaLora

Issue - State: closed - Opened by thistleknot about 1 year ago - 6 comments

#985 - Can't understand lora_dreambooth_inference after multi adapter was added

Issue - State: closed - Opened by Dentoty about 1 year ago - 1 comment

#984 - Improve documentation for IA³

Pull Request - State: closed - Opened by SumanthRH about 1 year ago - 5 comments

#983 - LoRA A and LoRA B dimensions mentioned in the paper are different from the implementation here.

Issue - State: closed - Opened by s3pi about 1 year ago - 3 comments

#982 - add lora target modules for stablelm models

Pull Request - State: closed - Opened by kbulutozler about 1 year ago - 1 comment

#981 - Adding LoRA adapters gets slower and slower with more adapters

Issue - State: closed - Opened by BenjaminBossan about 1 year ago - 4 comments

#980 - Combine multiple (IA)^3 Adapters and delete (IA)^3 adapters

Pull Request - State: closed - Opened by alexrs about 1 year ago - 12 comments

#979 - MNT Make .merged a property

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 1 comment

#978 - Add implementation of LyCORIS LoKr (KronA-like adapter) for SD&SDXL models

Pull Request - State: closed - Opened by kovalexal about 1 year ago - 21 comments

#977 - Fix missing tokenizer attribute in test

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 1 comment

#976 - FIX: issues with (un)merging multiple LoRA and IA³ adapters

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 1 comment

#975 - Add base model metadata to model card

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 8 comments

#974 - add the lora target modules for Mistral Models

Pull Request - State: closed - Opened by pacman100 about 1 year ago - 3 comments

#973 - Support multi-adpater concurrent inferencing

Issue - State: closed - Opened by whybeyoung about 1 year ago - 3 comments

#972 - ENH Support Conv2d layers for IA³

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 11 comments

#971 - What's up with autocast in PEFT?

Issue - State: closed - Opened by MFajcik about 1 year ago - 8 comments

#970 - model.save_pretrained returns empty dict

Issue - State: closed - Opened by Thewillman about 1 year ago - 2 comments

#969 - TST: Fix broken save_pretrained tests

Pull Request - State: closed - Opened by BenjaminBossan about 1 year ago - 1 comment

#968 - Wrong file read in test_save_pretrained_selected_adapters

Issue - State: closed - Opened by dzonesasaki about 1 year ago - 2 comments

#967 - Cannot make several adapters active through PeftModel interface

Issue - State: closed - Opened by kovalexal about 1 year ago - 6 comments

#966 - Update integrations_tests.yml

Pull Request - State: closed - Opened by younesbelkada about 1 year ago - 1 comment

#965 - Fix integrations_tests.yml

Pull Request - State: closed - Opened by younesbelkada about 1 year ago - 1 comment

#964 - Fix typo in custom_models.mdx

Pull Request - State: closed - Opened by Pairshoe about 1 year ago - 1 comment

#963 - [`tests`] Add transformers integration tests

Pull Request - State: closed - Opened by younesbelkada about 1 year ago

#962 - [`tests`] add transformers & diffusers integration tests

Pull Request - State: closed - Opened by younesbelkada about 1 year ago - 2 comments

#961 - [tests] add multiple active adapters tests

Pull Request - State: closed - Opened by pacman100 about 1 year ago - 2 comments

#960 - Training with peft costs more time than training the whole model

Issue - State: closed - Opened by Nidhogg-lyz about 1 year ago - 6 comments

#959 - Merging AdaLora module with 8bit base model and 4bit basemodel

Issue - State: closed - Opened by LokeshJatangi about 1 year ago - 2 comments

#958 - Implement LongLoRA trick for efficient tuning of long-context models

Issue - State: closed - Opened by casper-hansen about 1 year ago - 19 comments

#957 - Issues with loading and unload multiple adapter for inference?

Issue - State: closed - Opened by user06039 about 1 year ago - 14 comments

#956 - Add implementation of LyCORIS LoHa (FedPara-like adapter) for SD&SDXL models

Pull Request - State: closed - Opened by kovalexal about 1 year ago - 23 comments

#955 - How can I turn off wandb training?

Issue - State: closed - Opened by nameless0704 about 1 year ago - 4 comments

GitHub / huggingface/peft issues and pull requests