Repository Issue Activity (beta)

#4173 Circular import between accelerate.utils and accelerate.big_modeling breaks concurrent imports

#3041 infer_auto_device_map inefficiently allocates GPU memory for models with imbalanced module sizes

#4111 concatenate() still raises TypeError on non-tensor batch fields when num_processes > 1

#3996 DataLoader shuffle sequence replays from epoch 0 after resuming from a checkpoint

#4100 [Proposal] HeartFlow - Cognitive Engine for Accelerate

#4067 [Security] Path traversal + malicious-model DoS via sharded-checkpoint `weight_map` in ….

#4159 AcceleratedOptimizer.eval() does not reach through DeepSpeed wrapper, silently saving train-view schedule-free weights

#4154 send_to_device crashes on defaultdict batches

#3265 accelerator.prepare

#3547 `datasets` Iterable Dataset sharding support

#4122 BatchSamplerShard.__len__ overcounts for some ranks when split_batches=True and even_batches=False

#3078 [Community Contributions] examples on distributed inference using 🤗 Accelerate

#4102 `--parallelism_config_sp_seq_length_is_variable False` parses as True

#4086 When we use StateDictType.FULL_STATE_DICT to resume training from a checkpoint, it takes over 5 hours to start up.

#1089 how to cleanly exit when using `accelerate launch`

#4080 [bug report] DataSourcesShufflingDisallowed when training using split datasets.IterableDataset

#4099 [Feature] ShardedOptimStateDictConfig + per-rank save/load for large-scale FSDP checkpoint

#2401 SpanMarker support to the estimator tool

#4074 Feature Proposal: Add FunASR ASR Distributed Inference Examples

#4075 [Feature request] Support already-sharded DataLoaders in Accelerator.prepare

#4062 Question: Manual Dataset Sharding per GPU Rank with Accelerate + DistributedSampler (Avoid Double DataLoader Length Split)

#2734 [Feature Request] Allows registering custom trackers to internal tracker type registry

#4029 [Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes

#4088 pre-commit fails on a fresh checkout: pinned ruff v0.2.1 can't parse pyproject.toml (Unknown rule selector: UP045)

#4094 SuperCompress: Cut LLM token costs by ~65% with open-source prompt compression