Issues

• BOFT on MPS: safe_merge raises LinAlgError instead of ValueError when the adapter has a NaN

  huggingface/peft#4116 · Filter by author Alex-Wengg opened on Nov 4, 2026
  1

• C3ALinear.forward chains multiple active adapters through x instead of summing them

  huggingface/peft#4115 · Filter by author vineethsaivs opened on Nov 4, 2026

• Add Context Tuning as a new PEFT method, recently presented at ICML 2026

  huggingface/peft#4113 · Filter by author Jacklu0831 opened on Nov 4, 2026

• VeRA and PVeRA fail in forward() and merge() when adding a second adapter with a lower rank

  huggingface/peft#4109 · Filter by author apuravmanhas opened on Nov 3, 2026
  1

• Proposal: Add RAPA — Recursively Aligned Pathway Adaptation

  huggingface/peft#4106 · Filter by author freeSoul-SNU opened on Nov 3, 2026

• infer_device() returns "cpu" on torch.accelerator backends (e.g., torch_tpu on Cloud TPU)

  huggingface/peft#4101 · Filter by author jigyasa-grover opened on Nov 2, 2026
  1

• Update FSDP docs

  huggingface/peft#4096 · Filter by author BenjaminBossan opened on Nov 2, 2026
  2

• layers_to_transform without layers_pattern matches nothing when the layer list is at the model root

  huggingface/peft#4090 · Filter by author udsy19 opened on Oct 31, 2026
  6

• [v1 #4071] Disjoint auxiliary trainable tokens activate only the first adapter

  huggingface/peft#4088 · Filter by author DaoyuanLi2816 opened on Oct 31, 2026
  1

• add_weighted_adapter(combination_type="svd"): exact factorized SVD instead of decomposing the dense delta

  huggingface/peft#4079 · Filter by author jontb opened on Oct 29, 2026
  3

• layers_to_transform is silently ignored with regex target_modules in 15 PEFT configs (13 others raise)

  huggingface/peft#4070 · Filter by author Shaurya55555 opened on Oct 29, 2026
  2

• Error message for unmatched target_modules does not reveal that patterns are matched with re.fullmatch

  huggingface/peft#4067 · Filter by author jwh-yf opened on Oct 28, 2026
  4

• Mixed models: merge_adapter() merges nested adapters in reverse order, unmerge_adapter() then corrupts base weights

  huggingface/peft#4048 · Filter by author ATW117 opened on Oct 24, 2026
  5

• Cache the frozen LoRA-FA preconditioner to avoid repeated pseudoinverses

  huggingface/peft#4041 · Filter by author gorkadabo1 opened on Oct 23, 2026
  3

• LoRA-FA skips active adapter updates when a second adapter is present

  huggingface/peft#4039 · Filter by author cjwystz opened on Oct 23, 2026
  1

• [Feature Proposal] S-Dory-DoRA: Sparse Latent-Recurrent Directional Adaptation with QR Null-Space Guard

  huggingface/peft#4037 · Filter by author navadeep-9849 opened on Oct 23, 2026

• Loading an adapter from a Hub subfolder fails offline and with AutoPeftModel*

  huggingface/peft#4035 · Filter by author superdiaodiao opened on Oct 22, 2026
  2

• Prompt tuning: batched generate output depends on left padding

  huggingface/peft#4033 · Filter by author tivris opened on Oct 22, 2026
  3

• AdaLoRA Conv2d ignores padding_mode

  huggingface/peft#4032 · Filter by author Sergasgr opened on Oct 21, 2026
  3

• AdaLoRA Conv2d crashes in forward and merge with groups > 1

  huggingface/peft#4031 · Filter by author Sergasgr opened on Oct 21, 2026
  1

• Performance improvements for intruder dimension reduction

  huggingface/peft#4030 · Filter by author causeri3 opened on Oct 21, 2026
  3

• 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

• 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
  2

• 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
  5