Repository Issue Activity (beta)

#1889 Eval dataset for training docs

#1890 Eval dataset tip: chain-ground-truth finance benchmark

#1876 autoresume_default variable

#1888 Consider using pre-built Flash Attention kernels via `kernels`

#1848 signaling error on shared filesystem is back

#1880 404 error when accessing MPT models on HuggingFace

#1828 Training stuck when auto_resume=True

#1791 memory bound and global_batch_size

#1785 data_prep format

#1716 Flash Attention 3 Usage

#394 Unify export script arguments after next version release

#846 Any plan for supporting DPO?

#1568 MI300X Compatibility

#1732 MBT-30B conflict after Flash Attention 2.7.0

#1744 dmoe from hf

#1746 Using NoConcatDataset

#1736 Usage of composer.utils.dist.get_node_signal_file_name()

#1743 Tensor Parallelism, Pipeline Parallelism, Sequence Parallelism

#1705 Tests fail in local environment due to device type mismatch

#1731 Provide a metric that uses Math-Verify

#1538 Fine-tuning error in conda environment without docker image

#1722 Make metric name handling more flexible and transparent

#1718 quickstart train on cpu with m1?

#1715 TransformerEngine attention

#1706 Unable to reproduce the benchmark