Issues

• convert_to_lora silently converts variants that cannot be merged (aLoRA), producing a non-equivalent adapter

  huggingface/peft#4170 · Filter by author Y9Y123 opened on Nov 13, 2026

• ties and dare_ties elect the majority sign before applying adapter weights

  huggingface/peft#4169 · Filter by author JaypeeLan opened on Nov 13, 2026

• get_additive_delta / convert_to_lora ignore the DoRA magnitude after training

  huggingface/peft#4168 · Filter by author anh-nguyenn opened on Nov 13, 2026

• target_parameters: LoRA parametrization is not removed after forward on 3-D expert weights that already have a parametrization

  huggingface/peft#4162 · Filter by author Shivansh237 opened on Nov 12, 2026
  1

• LoRA on MoE experts (target_parameters) fails under FSDP + expert parallelism

  huggingface/peft#4161 · Filter by author qgallouedec opened on Nov 12, 2026

• Add Deletion Tuning as a new PEFT method, to be presented at NeurIPS 2026

  huggingface/peft#4158 · Filter by author Jacklu0831 opened on Nov 12, 2026
  1

• BOFT fails on Conv2d layers with non-square kernels

  huggingface/peft#4152 · Filter by author MaiAditya opened on Nov 11, 2026
  5

• HRA forward crashes for Conv2d padding='same' and padding='valid'

  huggingface/peft#4151 · Filter by author gorkadabo1 opened on Nov 11, 2026
  2

• merge_adapter() / merge_and_unload() silently drop adapter when base layer is parametrized (e.g. weight_norm)

  huggingface/peft#4148 · Filter by author Shivansh237 opened on Nov 11, 2026
  3

• CorDA and LoRA-GA preprocessing fail with a cache filename in the current directory

  huggingface/peft#4144 · Filter by author ananya-kaul opened on Nov 10, 2026
  2

• add_weighted_adapter silently drops source layers when targeting filters differ (LoRA and IA3)

  huggingface/peft#4133 · Filter by author Ovi-Poddar opened on Nov 8, 2026
  2

• FSDP checkpoint resume deadlocks across ranks during distributed optimizer state sync (Llama-405B, 8x H200)

  huggingface/peft#4132 · Filter by author utkarshxdev opened on Nov 7, 2026
  5

• Warn when LoRA SFT trains on special tokens whose input embedding is all zero (silent failure on Qwen2.5-Coder tool calls)

  huggingface/peft#4130 · Filter by author euisuh opened on Nov 6, 2026
  11

• load_adapter leaves a selectable adapter after a safetensors read failure

  huggingface/peft#4127 · Filter by author Max-Reisinger opened on Nov 6, 2026
  1

• RandLoRA: wrong matrix multiplication in 8-bit/4-bit get_delta_weight() and broken multi-adapter rank handling

  huggingface/peft#4125 · Filter by author apuravmanhas opened on Nov 5, 2026
  4

• IA3 feedforward_modules=['fc'] also marks modules like not_fc as feedforward

  huggingface/peft#4118 · Filter by author mengchen-atx opened on Nov 5, 2026
  4

• LoHa rank_dropout can produce NaN when the keep-mask is all zeros

  huggingface/peft#4117 · Filter by author mengchen-atx opened on Nov 5, 2026
  3

• 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
  8

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

  huggingface/peft#4115 · Filter by author vineethsaivs opened on Nov 4, 2026
  1

• 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
  3

• 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
  8

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

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