Ecosyste.ms: Timeline
Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.
shamanez pushed 77 commits to main shamanez/verl
- [megatron] docs: clean up unused code, update megatron backend docs and installation docs (#89) * [megatron] style: ... e88cf81
- [misc] feat: spport rmpad/data-packing in FSDP with transformers (#91) * init commit of rmpad * add rmpad test ... 569210e
- [example] docs: add getting started notebook with free GPUs from lightning (#92) e6b089c
- [example] fix: fix notebook link due to username update (#94) * update lightning link * Update verl_getting_start... e08c428
- fix readme and add back citation (#98) 53c3ff4
- [misc] fix reward model issue with TokenClassification model and support running particular steps instead of epochs (... a0e8ed2
- [misc] feat: support different flash_attn versions with variable num returns (#100) * add ci * fix reward model a... 1facb9d
- Fix loss value for gradient accumulation > 1 (#102) e230de8
- refact: hybrid_engine dir to sharding_manager for more general representation (#103) 6a9f6e1
- [misc] feat: add Ray Summit Youtube video link (#105) As title c1d5e76
- [misc] fix: fix license (#110) - fix license - add license ci a33a3ba
- [readme] docs: add acknowledgement (#107) 4566cfb
- Update README.md installation link d0152e1
- [ci] fix: add force stop in ray e2e ci to clean env (#112) - As titled ff0c7cc
- [misc] chore: refactor and add several metrics (#111) - Add format script - Move save_checkpoint to a separate func... 018b0d7
- [ci] fix: change VLLM_ATTENTION_BACKEND to XFORMERS to avoid illegal memory access (#113) 594d80a
- [misc][Long Context] feat: support ulysses for long context training (#109) e8eb9e4
- [perf] fix: set use_reentrant=False when enable gradient checkpointing (#114) - Set use_reentrant=False to avoid dup... 5a94e14
- [dataproto] fix: add assertion for uneven chunk (#115) - forbid uneven chunk for DataProto 1ec5eb5
- [misc] feat: support mfu calculation (#117) 41f645d
- and 57 more ...