Issues

• Performance improvements for intruder dimension reduction

  huggingface/peft#4030 · Filter by author causeri3 opened on Oct 21, 2026

• LoftQ init raises TypeError for Conv, Embedding, target_parameters and Megatron LoRA layers

  huggingface/peft#4028 · Filter by author nettanvirdev opened on Oct 21, 2026

• Nightly regression: bf16 state dict cases fail due to CPU kernel differences (not a PEFT bug)

  huggingface/peft#4019 · Filter by author peft-jambot opened on Oct 20, 2026

• examples/adamss_finetuning/test_adamss_quick.py fails: outdated imports and config args

  huggingface/peft#4013 · Filter by author sarthakuwar opened on Oct 19, 2026
  1

• Memory-efficient PiSSA initialization for 4-bit models (replace_lora_weights_pissa, analogous to replace_lora_weights_loftq)

  huggingface/peft#4010 · Filter by author jietinatian0803 opened on Oct 19, 2026
  1

• AdaMSS: forward crashes after casting the PEFT model's dtype (e.g. model.to(torch.bfloat16)) because the layer caches the base weight dtype at init

  huggingface/peft#4003 · Filter by author ghaith-mq opened on Oct 17, 2026
  2

• RoAd mixed-adapter batches fail for float16/bfloat16 models

  huggingface/peft#3997 · Filter by author mottopanikeiku opened on Oct 16, 2026
  4

• Loading an adapter from a repo subfolder fails with HF_HUB_OFFLINE=1

  huggingface/peft#3995 · Filter by author mottopanikeiku opened on Oct 16, 2026
  3

• [New method] Bayesian-LoRA: Probabilistic Low-Rank Adaptation (ICML 2026)

  huggingface/peft#3994 · Filter by author moulelin opened on Oct 16, 2026
  1

• [Proposal] Add Cortex-Structural: parameter-efficient FFN expansion for dense and MoE models

  huggingface/peft#3990 · Filter by author MikeDean2367 opened on Oct 16, 2026
  3

• n_frequency_pattern, block_size_pattern and lambda_pattern lose their key order on save

  huggingface/peft#3985 · Filter by author devYRPauli opened on Oct 15, 2026
  2

• Revalidate Transformer Engine GPU CI support

  huggingface/peft#3983 · Filter by author sbhavani opened on Oct 14, 2026
  1

• OFT Conv2d: unmerged and merged outputs differ for kernels larger than 1x1

  huggingface/peft#3982 · Filter by author Kayvan-Zahiri opened on Oct 14, 2026
  2

• Mixed adapter batch raises KeyError when only one adapter has modules_to_save

  huggingface/peft#3967 · Filter by author mottopanikeiku opened on Oct 14, 2026
  4

• unload() returns the trained modules_to_save copy instead of the original module

  huggingface/peft#3955 · Filter by author Sergasgr opened on Oct 13, 2026
  1

• Prompt-learning methods crash on transformers 4.43/4.44: the pre-get_text_config fallback uses in on a PretrainedConfig

  huggingface/peft#3951 · Filter by author hazelxue opened on Oct 13, 2026

• get_peft_model_state_dict gathers the whole sharded base model on DTensor-sharded models

  huggingface/peft#3930 · Filter by author qgallouedec opened on Oct 11, 2026
  5

  

• ensure_weight_tying=True silently stops tying the LoRA weights when the model is in bf16/fp16

  huggingface/peft#3929 · Filter by author superdiaodiao opened on Oct 11, 2026
  3

• LoRA on MoE experts (target_parameters) fails under tensor/expert parallelism

  huggingface/peft#3914 · Filter by author qgallouedec opened on Oct 9, 2026
  2

• Default CPTConfig masks all input labels and returns NaN training loss

  huggingface/peft#3899 · Filter by author aswanth-07 opened on Oct 6, 2026
  3

• Adapters are saved under transformers v5's renamed module paths, while save_pretrained writes base models in the original checkpoint layout

  huggingface/peft#3896 · Filter by author eilamc14 opened on Oct 5, 2026
  9

• Add support for BoRA

  huggingface/peft#3877 · Filter by author uis246 opened on Oct 3, 2026
  1

• get_peft_model_state_dict returns an empty dict for FSDP1-wrapped models since 0.21.0 (adapter checkpoints / weight syncs silently empty)

  huggingface/peft#3876 · Filter by author murphy977 opened on Oct 3, 2026
  4

• MoE recipe in the LoRA docs gives adapters that vLLM serves with the wrong expert scaling

  huggingface/peft#3874 · Filter by author surajsharan opened on Oct 2, 2026
  1

• Astra preprocessing scales activations by max value, not max abs value, can overflow fp16 covariance

  huggingface/peft#3871 · Filter by author rupeshpoojary9 opened on Oct 2, 2026
  6