All issues

Search results

prepare_model unconditionally imports DTensor, breaking single-GPU training on torch builds without a distributed backend

Status: Open.

#4249 In huggingface/accelerate;

· LeoBorcherding opened on Sep 11, 2026

[Bug] Fused AdamW AMP overflow leaves step_was_skipped false and advances the scheduler

Status: Open.

#4242 In huggingface/accelerate;

· gss10282023 opened on Sep 10, 2026

ACCELERATE_BYPASS_DEVICE_MAP is the only boolean env flag read case-sensitively, so =True is silently ignored

Status: Open.

#4240 In huggingface/accelerate;

· Sreekant13 opened on Sep 10, 2026

use_stateful_dataloader + num_workers>0 draws epoch-0 permutation during prepare() before cross-rank RNG sync — corrupts DDP data partition

Status: Open.

#4204 In huggingface/accelerate;

· GoldenStain opened on Sep 2, 2026

Skip automatic DataLoader sharding when the loader is already rank-sharded

Status: Open.

#4199 In huggingface/accelerate;

· YeonwooSung opened on Sep 2, 2026

use_stateful_dataloader under multi-process training restores the cursor, not the shuffle order — sampler permutation is never serialized

Status: Open.

#4195 In huggingface/accelerate;

· GoldenStain opened on Aug 30, 2026

Feature request: FSDP2 QLoRA

feature requestRequest for a new feature to be added to AccelerateRequest for a new feature to be added to Accelerate

Status: Open.

#3874 In huggingface/accelerate;

· ojh31 opened on Dec 5, 2025

Feature Request: Add API to Materialize Meta Tensors for Device Movement

feature requestRequest for a new feature to be added to AccelerateRequest for a new feature to be added to Accelerate

Status: Open.

#3866 In huggingface/accelerate;

· EricRollei opened on Nov 30, 2025

datasets Iterable Dataset sharding support

feature requestRequest for a new feature to be added to AccelerateRequest for a new feature to be added to Accelerate

Status: Open.

#3547 In huggingface/accelerate;

· ValMystletainn opened on May 5, 2025

[CPU offload hooks] hooks with overlapped transfers and computations

feature requestRequest for a new feature to be added to AccelerateRequest for a new feature to be added to Accelerate

Status: Open.

#3267 In huggingface/accelerate;

· sayakpaul opened on Nov 29, 2024

accelerator.prepare

Status: Open.

#3265 In huggingface/accelerate;

· wfan1203 opened on Nov 26, 2024

[BUG] Accelerate 1.0.1 failed to train multiple zero-3 models

wipWork in progressWork in progress

Status: Open.

#3170 In huggingface/accelerate;

· nrailg opened on Oct 15, 2024