microsoft/DeepSpeed

Pull Requests Commits

Merge branch 'master' into staging-zero-inference-v1

awan-10 committed 2 years ago

Verified 9723d93f

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

skipping redundant MoE optimizer state loading (#4120)

Alexander Jipa committed 2 years ago

Verified e801e6d7

Code Refactoring (#4262)

tosemml committed 2 years ago

Verified 9894c06a

Add check that opening issues on CI failure requires schedule (#4242)

loadams committed 2 years ago

Verified c93e89a3

Checks for user injection policy (#3052)

satpalsr committed 2 years ago

Verified 430510bf

Allow modification of zero partitioned parameters (#4192)

tjruwase committed 2 years ago

Verified a23cda6c

Pin Triton version to >=2.0.0 and <2.1.0 (#4251)

lekurile committed 2 years ago

Verified f96c1c0a

Update README.md (#4244)

awan-10 committed 2 years ago

Verified 844eb685

DeepSpeed-Chat Llama2/stability release (#4240)

awan-10 committed 2 years ago

Verified 2420e23f

Add Mixed Precision ZeRO++ tutorial (#4241)

HeyangQin committed 2 years ago

Verified 69e89dd5

jeffra committed 2 years ago

dc7b65f6

Enable hpz when running with torch.no_grad (#4232)

HeyangQin committed 2 years ago

Verified 462def45

fix MegatronLayerPolicy to be compatible with the newest ParallelTransformerLayer (#4236)

Dino Chen committed 2 years ago

Verified 6cbf6661

Enable AMD MI200 and H100 to run on branches for testing (#4238)

loadams committed 2 years ago

Verified 5dbc5313

Merge branch 'master' into staging-zero-inference-v1

tjruwase committed 2 years ago

Verified a63e92b9

fix: linker issues in conda environments #3929 (#4235)

maximegmd committed 2 years ago

Verified 194b0043

Empty tensor size check (#4186)

jomayeri committed 2 years ago

Verified 57d629a1

added port argument for ssh (#4117)

Hiromasa-H committed 2 years ago

Verified 8145b5e4

Fix fused qkv sizing for bloom (#4161)

molly-smith committed 2 years ago

Verified 042115c8

add npu support dtypes (#4223)

CurryRice233 committed 2 years ago

Verified 60d7b0a3

feat(activation_checkpointing): add `non_reentrant_checkpoint` to support inputs require no grad (#4118)

hughpu committed 2 years ago

Verified 42c1e916

Older