All pull requests

Pull requests

• ENH Add replace_lora_weights_pissa for 4-bit models

  triaged
  #4135·jietinatian0803 opened Nov 8, 2026·

  ·
  1

  10

• Add Bayesian-LoRA as a LoRA variant

  triaged
  #4131·moulelin opened Nov 6, 2026·

  ·

• docs: assorted fixes

  triaged
  #4124·githubnemo opened Nov 5, 2026Collaborator·

  ·
  2

• Use the peft-internal-testing tiny random Llama model in the parallelism tests

  triaged
  #4121·albertvillanova opened Nov 5, 2026Member·

  ·
  2

• [Distributed] embedding_colwise for LoRA adapters

  triaged
  #4112·michaelbenayoun opened Nov 3, 2026Member·

  ·
  8

• Pass token_type_ids and position_ids in PeftModelForQuestionAnswering

  triaged
  #4110·Sergasgr opened Nov 3, 2026Contributor·

  ·
  1

• FIX Empty adapter state dict with nested FSDP or checkpoint_wrapper

  triaged
  #4108·hivaze opened Nov 3, 2026Contributor·

  ·
  1

  8

• DOC Update FSDP guide and add an FSDP2 SFT configuration

  triaged
  #4107·DaoyuanLi2816 opened Nov 3, 2026Contributor·

  ·
  1

  10

• [WIP] Option to allow exact merge-unmerge roundtrip

  triaged
  #4098·BenjaminBossan opened Nov 2, 2026Member·

  ·
  16

• TST: merge/unmerge round trip test for base weights (CI check only, do not merge)

  triaged
  #4097·peft-jambot opened Nov 2, 2026Contributor·

  ·
  2

• CI: Bump the third-party-actions group with 5 updates

  dependenciesgithub_actionstriaged
  #4093·dependabot[bot] opened Nov 1, 2026Bot·

  ·
  1

• FIX AdaLoRA Conv2d with groups and padding_mode

  triaged
  #4084·Sergasgr opened Oct 30, 2026Contributor·

  ·
  2

  13

• FIX MHA weight restoration issues

  triaged
  #4083·BenjaminBossan opened Oct 30, 2026Member·

  ·
  1

• ENH Handle dtype consistently when merging and unmerging

  triaged
  #4082·BenjaminBossan opened Oct 30, 2026Member·

  ·
  1

• ENH: Svd lowrank option for lora intruder dimension reduction

  triaged
  #4081·causeri3 opened Oct 29, 2026·

  ·
  1

  5

• FIX Make FSDP2 merges reach the sharded parameters

  triaged
  #4077·RerankerGuo opened Oct 29, 2026·

  ·
  1

  24

• [WIP] [V1] Allow setting multiple active adapters on PeftModel

  PEFT v1triaged
  #4071·BenjaminBossan opened Oct 29, 2026Member·

  ·
  2

• fix: reuse the 4-bit quantization scale when unmerging LoRA

  triaged
  #4069·peft-jambot opened Oct 29, 2026Contributor·

  ·
  29

• FEAT Add Activation Steering

  triaged
  #4066·BenjaminBossan opened Oct 28, 2026Member·

  ·
  10

• FIX Gather FSDP2 adapter DTensors when saving

  triaged
  #4065·Levius-Fubuki opened Oct 28, 2026·

  ·
  1

  24

  

• FEAT Add Cortex-Structural adapters for dense and MoE models

  triaged
  #4064·MikeDean2367 opened Oct 27, 2026·7 tasks done·

  ·
  1

  15

• Add temporarily_merge_adapter for exact LoRA merge/unmerge round trips

  triaged
  #4062·winglian opened Oct 26, 2026Contributor·

  ·
  2

• Use the peft-internal-testing tiny Llama4 model

  triaged
  #4057·albertvillanova opened Oct 26, 2026Member·

  ·
  4

• ENH Add generic PEFT loss and optmizer hook

  triaged
  #4036·BenjaminBossan opened Oct 23, 2026Member·

  ·
  1

• AGENTS.md: PR description should fit git history

  triaged
  #4034·githubnemo opened Oct 22, 2026Collaborator·

  ·
  2