Repository Issue Activity (beta)

#3635 X-LoRA: disable_adapter does not disable the experts, and generate() freezes the classifier / switches the model to eval mode

#3660 Supertuning: save_precomputed_indices for later adapters follows the first adapter

#3650 Trainable Tokens random initialization prevents unmerge from restoring base weights

#3620 [Bug] ShadowPEFT sequence classification pools the wrong token with left padding

#3644 Docstring examples in peft_model.py: four of six PeftModelFor* examples fail on removed config fields

#3682 BOFT Conv2d: safe_merge=True skips the non-finite check and corrupts the base layer

#3631 Fix state-dict round-trip for directly injected shared tuners

#3306 LoraConfig(bias="lora_only") trains base_layer.bias but save_pretrained/get_peft_model_state_dict drops it

#3638 Adaption prompt saves the wrong weights for inactive adapters; delete_adapter raises a transformers TypeError for prompt learning and adaption prompt

#3654 PeftConfig.from_pretrained raises a bare KeyError for an invalid peft_type

#3670 Add NoRA and NoRA-init support for LoRA

#3676 [Bug] ignore_mismatched_sizes crashes on scalar adapter state entries

#3478 Prefix sizing from per_layer_config ignores KV head count

#3675 Four of the five diffusers-style tuner docstring examples (LoHa, LoKr, OFT, MiSS) fail as written

#3582 rank_pattern / alpha_pattern keys that match zero layers are applied silently; anchored and fully-qualified keys can never match

#3580 Loading a trainable adapter changes the `requires_grad` state of the active adapter

#3658 target_modules can match a subset of layers with no warning

#3646 save_pretrained treats empty name_or_path as a Hub repo id

#3662 Add Astra (Activation-Space Tail-Eigenvector LoRA) initialization

#3615 [RFC, I can do a PR] Make `lora_alpha` a float

#3626 [Bug] ShadowPEFT LM-head trainability freeze after set_adapter — modules_to_save=["lm_head"] silently stops receiving grads

#3584 save_pretrained writes an empty/truncated adapter checkpoint (silently) when another adapter name equals a base-model module path component

#3574 LN Tuning: KeyError on forward/merge when the active adapter does not target all adapted layers

#3512 MixedModel.delete_adapter: stale active adapter and leaked auxiliary modules

#3480 Expand model test matrices to cover LoHa, LoKr, and LN Tuning