microsoft/DeepSpeed

Pull Requests Commits

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

awan-10 committed 2 years ago

5e09acea

Fix autotune to support Triton 2.1 (#4340)

stephen-youn committed 2 years ago

Verified 3f3e9fb1

Fix skipped inference tests (#4336)

mrwyattii committed 2 years ago

Verified 8c7f7fd2

Suppress noise (#4310)

tjruwase committed 2 years ago

Verified 3ea42c6e

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

sakogan committed 2 years ago

Verified 9bf77782

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

lekurile committed 2 years ago

Verified 0c83436a

clear redundant timers (#4308)

starkhu committed 2 years ago

Verified e20e4a9d

Add release version checking (#4328)

loadams committed 2 years ago

Verified b5977d41

Fix Zero3 contiguous grads, reduce scatter false accuracy issue (#4321)

nelyahu committed 2 years ago

Verified 11aa880e

Clean up modeling code (#4320)

loadams committed 2 years ago

Verified 180dd397

Handle empty parameter groups (#4277)

tjruwase committed 2 years ago

Verified 9adc73ff

Update README.md (#4316)

NinoRisteski committed 2 years ago

Verified 48f2192a

README update (#4303)

tjruwase committed 2 years ago

Verified 581e44dd

Update release and bump patch versioning flow (#4286)

loadams committed 2 years ago

Verified 8c1eed2e

added a bert-model check for triton (#4266)

stephen-youn committed 2 years ago

Verified ffd82bb0

ZeRO-Inference v2 release

tjruwase committed 2 years ago

94174f88

jeffra committed 2 years ago

c700195d

Update index.md (#4297)

NinoRisteski committed 2 years ago

Verified 04a6fedf

fix user args parsing of string with spaces on runner (#4265)

YudiZh committed 2 years ago

Verified e75c285a

ZeRO-Inference refresh (#4197)

tjruwase committed 2 years ago

Verified aa4a7401

AMD Kernel Compatibility Fixes (#3180)

cmikeh2 committed 2 years ago

Verified 542dc0d5

Support iterators with incompletely defined __len__ functions (#2445)

codedecde committed 2 years ago

Verified e1949565

Keep hpz secondary tensor in forward pass (#4288)

HeyangQin committed 2 years ago

Verified 1f0a44d9

update deepspeed to run with the most recent triton 2.1.0 (#4278)

stephen-youn committed 2 years ago

Verified e8ed7419

Update README.md (#4284)

NinoRisteski committed 2 years ago

Verified c55fad9f

fix iteration timing used in autotuning when gradient_accumulation_steps > 1 (#2888)

cli99 committed 2 years ago

Verified bce6ed1c

Use clearer naming (#2548)

Crispig committed 2 years ago

Verified 18a04d04

use ```non_reentrant_checkpoint``` fix requires_grad of input must be true for activation checkpoint layer in pipeline train. (#4224)

inkcherry committed 2 years ago

Verified 60a3e89e

check for zero (empty) param groups in llama + hf/accelerate. (#4270)

awan-10 committed 2 years ago

Verified 16d8953a

Fix nv-inference/un-pin transformers (#4269)

loadams committed 2 years ago

Verified 55d9964c

Older