microsoft/DeepSpeed

Pull Requests Commits

Test with their re-instated pypi releases

loadams committed 2 years ago

a507e41b

loadams committed 2 years ago

0da9b635

loadams committed 2 years ago

1c81efbb

Revert other changes

loadams committed 2 years ago

c620e498

loadams committed 2 years ago

da37269b

Find matching megatron version

loadams committed 2 years ago

719a901d

Fix pipeline module evaluation when contiguous activation checkpointing is enabled (#3005)

BacharL committed 2 years ago

Verified 7ddc3b01

remove megatron-lm, no longer pip installable (#3389)

jeffra committed 2 years ago

Verified a094c976

Missed updating one tensor-model-parallel-size

loadams committed 2 years ago

adbc7e28

MII tests passed with this before, then failed

loadams committed 2 years ago

c227fe39

Add updates required by Megatron

loadams committed 2 years ago

5d47a016

loadams committed 2 years ago

4958df8b

Use git commit corresponding to 1.1.5

loadams committed 2 years ago

51b9fb98

loadams committed 2 years ago

87336a9b

Force lower torch version

loadams committed 2 years ago

f5a86f2b

Update other flows

loadams committed 2 years ago

ac0ac7d5

Add more to the builds

loadams committed 2 years ago

d4620038

Add more six, regex

loadams committed 2 years ago

6dfa0048

confused pip and apt

loadams committed 2 years ago

7ded2cd4

Add six install to other packages

loadams committed 2 years ago

3f4a987f

Revert to older megatron installed via git

loadams committed 2 years ago

6f420b8b

[docs] add twitter badge to readme

jeffra committed 2 years ago

Verified f7d71ec1

Fix memory leak in zero2 contiguous gradients (#3306)

BacharL committed 2 years ago

Verified 01d17492

stage_1_and_2.py: do gradient scale only for fp16 (#3166)

guoyejun committed 2 years ago

Verified 0e357666

[DRAFT] Tentative implementation of MiCS (#2964)

zarzen committed 2 years ago

Verified 2e99f6ed

Auto TP Tutorial with T5 Example (#2962)

molly-smith committed 2 years ago

Verified d9253950

fixing default communication_data_type for bfloat16_enabled and docs (#3370)

Alexander Jipa committed 2 years ago

Verified d56268f3

Fix redundant shared_params in zero_to_fp32.py (#3149)

ShijieZZZZ committed 2 years ago

Verified 39825a90

add deepspeed chat blog links, add tags (#3369)

conglongli committed 2 years ago

Verified 085981bf

Update DS-Chat issue template (#3368)

mrwyattii committed 2 years ago

Verified 3031eec4

Older