Update references to new GitHub org (deepspeedai)#461

Closed

loadams wants to merge 21 commits intomainfrom loadams/update-gh-org

+51,585-279

Commits on Feb 7, 2025

Update/add GPT/Llama universal checkpointing scripts (#391 )

lekurile
authored and
loadams
committed
fixing the bug of flash_attn import and the wrong gather index when using flash_attn_cuda in sequence parallel (#406 )

authored and
loadams
committed
add fused_rms_norm support on XPU device (#431 )

ys950902
authored and
loadams
committed
pass batch_dim_idx to deepspeed sequence parallel distributed attention for supporting batch size larger than 1 (#433 )

authored and
loadams
committed
[LLaMa] Adding support converting checkpoint from mds to hf (#432 )

billishyahao
authored and
loadams
committed
add device check when import ipex (#436 )

ys950902
authored and
loadams
committed
fix TFLOPs calculation (#371 )

polisettyvarma
authored and
loadams
committed
fix nan issue when running megatron-deepspeed (#434 )

ys950902
authored and
loadams
committed
enable empty cache on XPU device (#438 )

ys950902
authored and
loadams
committed
[wandb] disable wandb more gracefully (#422 )

billishyahao
and
loadams
committed
[Bug] Fix crash when logging optimizer state to tb (#417 )

billishyahao
authored and
loadams
committed
Enable Sequence Parallelism (#429 )

polisettyvarma
authored and
loadams
committed
grad_wei can't be NoneType when running with DeepSpeed, for zero3 will divided the gradient (#428 )

ys950902
authored and
loadams
committed
fix init issue for rms_norm in squence_parallel (#448 )

ys950902
authored and
loadams
committed
enable profiler for specific ranks (#451 )

ranzhejiang
authored and
loadams
committed
fix init issue for silently ignoring the deepspeed config (#452 )

xylian86
authored and
loadams
committed
fix moe tflops (#445 )

ranzhejiang
authored and
loadams
committed
Adding the new feature of FPDT (#441 )

authored and
loadams
committed
[tool]GQA convert support (#454 )

inkcherry
authored and
loadams
committed
Fix import error in deepspeed_to_megatron.py (#455 )

hotsuyuki
authored and
loadams
committed
Update GH org
loadams
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update references to new GitHub org (deepspeedai)#461

Commits on Feb 7, 2025

Update/add GPT/Llama universal checkpointing scripts (#391 )

fixing the bug of flash_attn import and the wrong gather index when using flash_attn_cuda in sequence parallel (#406 )

add fused_rms_norm support on XPU device (#431 )

pass batch_dim_idx to deepspeed sequence parallel distributed attention for supporting batch size larger than 1 (#433 )

[LLaMa] Adding support converting checkpoint from mds to hf (#432 )

add device check when import ipex (#436 )

fix TFLOPs calculation (#371 )

fix nan issue when running megatron-deepspeed (#434 )

enable empty cache on XPU device (#438 )

[wandb] disable wandb more gracefully (#422 )

[Bug] Fix crash when logging optimizer state to tb (#417 )

Enable Sequence Parallelism (#429 )

grad_wei can't be NoneType when running with DeepSpeed, for zero3 will divided the gradient (#428 )

fix init issue for rms_norm in squence_parallel (#448 )

enable profiler for specific ranks (#451 )

fix init issue for silently ignoring the deepspeed config (#452 )

fix moe tflops (#445 )

Adding the new feature of FPDT (#441 )

[tool]GQA convert support (#454 )

Fix import error in `deepspeed_to_megatron.py` (#455 )

Update GH org