Repository Issue Activity (beta)

#3942 Security: request to open a private disclosure channel

#3930 [Question] How automatic gradient accumulation clear the activation saved on forward pass

#3860 Stuck when continuing training with auto_resume: true

#3920 OptimizerMonitor Error with FSDP2

#3902 Incorrect autoresumption behavior with non-default CheckpointSaver

#3916 FSDP2 Not Compatible w/ HSDP (DeviceMesh Ignored By State)

#3905 MLFlow logger: log system metrics for all nodes when rank_zero_only is True

#3892 Can't log to S3 during training

#3868 Chat template is causing HuggingFaceModel checkpointing to fail

#3801 Support for FSDP2

#3658 Make Composer compatible with Python 3.12.

#3776 MPT-30B conflict with Flash attention>=2.7.0

#3170 Remote file name in `MemorySnapshot` not being formatted

#3761 Bring back DeepSpeed 🚀

#2778 InContextLearning*Dataset Default padding sides hardcoded?

#2564 Is FlashAttention really used while using HuggingFaceModel supported as one of ComposerModel types.

#3583 Torch script export not working with hugging face model

#3504 Eval Loss

#3762 Add support for `torch` 2.6

#3699 `DataSpec.device_transforms` do not run on device

#3725 Diverging runs when multiple resume

#864 SWA - RuntimeError: Only Tensors created explicitly by the user (graph leaves) support the deepcopy protocol at the moment

#3723 DistributedSampler seeding

#3690 Loosen `torchmetrics` requirements

#3680 [Question] Gradient Accumulation issue