microsoft/DeepSpeed

Pull Requests Commits

Merge branch 'master' into styoun/triton-flash2

lekurile committed 1 year ago

Verified 544f1ddc

Fix llama meta tensor loading in AutoTP and kernel injected inference (#3608)

zeyugao committed 1 year ago

Verified 4fc2c8e7

Fix min torch version (#4375)

tjruwase committed 1 year ago

Verified 463dea27

Fix multinode runner to properly append to PDSH_SSH_ARGS_APPEND (#4373)

loadams committed 1 year ago

Verified 17957728

add the missing method (#4363)

cli99 committed 1 year ago

Verified 727609df

Openfold fix (#4368)

cctry committed 1 year ago

Verified c5814647

deepspeed4science japanese blog (#4369)

conglongli committed 1 year ago

Verified a99e5d3f

deepspeed4science chinese blog (#4366)

conglongli committed 1 year ago

Verified 3592a22c

Enable workflow dispatch on Torch 1.10 CI tests (#4361)

loadams committed 1 year ago

Verified dcd3ae19

Update conda env to have max pydantic version (#4362)

loadams committed 1 year ago

Verified dcf649c3

add deepspeed4science blog link (#4364)

conglongli committed 1 year ago

Verified da7a1851

added check to avoid undefined behavior when the input_id length is greater than max_tokens (#4349)

mrwyattii committed 1 year ago

Verified 85334238

Add the policy to run llama model from the official repo (#4313)

RezaYazdaniAminabadi committed 1 year ago

Verified 468882fb

fix deepspeed4science links (#4358)

conglongli committed 1 year ago

Verified 78c3b148

DeepSpeed4Science (#4357)

conglongli committed 1 year ago

Verified f876d81d

Support InternLM (#4137)

wangruohui committed 1 year ago

Verified 367d6f9c

Pass base_dir to model files can be loaded for auto-tp/meta-tensor. (#4348)

awan-10 committed 1 year ago

Verified b9d719a6

fix according to the PR comment

styoun committed 1 year ago

4e6c1646

Merge branch 'master' into styoun/triton-flash2

styoun committed 1 year ago

4bae6079

cleanup flash kernel

styoun committed 1 year ago

fae2ab9a

Fix autotune to support Triton 2.1 (#4340)

stephen-youn committed 1 year ago

Verified 3f3e9fb1

Fix skipped inference tests (#4336)

mrwyattii committed 1 year ago

Verified 8c7f7fd2

styoun committed 1 year ago

401b3d5a

styoun committed 1 year ago

f30f64ca

styoun committed 1 year ago

26853e9d

triton supports the flash attention when compute cap > 8.0

styoun committed 1 year ago

1ec42552

Suppress noise (#4310)

tjruwase committed 1 year ago

Verified 3ea42c6e

Fix a bug in the implementation of dequantization for inference (#3433)

sakogan committed 1 year ago

Verified 9bf77782

DS-Chat BLOOM: Fix Attention mask (#4338)

lekurile committed 1 year ago

Verified 0c83436a

is_triton_supported added to Accelerator class

styoun committed 1 year ago

b6c47f6b

Older