deepspeedai/DeepSpeed

Pull Requests Commits

Refine the FP16-to-FP6 cast logic.

JamesTheZ committed 2 years ago

368a7630

Add FP6 end-to-end unit tests

arashb committed 2 years ago

ff6c3c3a

[WIP] Fix the bug of FP16-to-FP6 data packing.

JamesTheZ committed 2 years ago

1ba45fdd

Make the quantizer on GPU.

JamesTheZ committed 2 years ago

1c2131d2

Update CUDA kernels and clean codes.

JamesTheZ committed 2 years ago

91bb4d79

FP6 quantization end-to-end.

JamesTheZ committed 2 years ago

a4562ab7

Use `deepspeed.comm` instead of `torch.distributed` (#5225)

jinyouzhi committed 2 years ago

Verified acf07398

Fix moe cpu offload (#5220)

RezaYazdaniAminabadi committed 2 years ago

Verified e6e8c137

Rename nv-torch-latest-cpu workflow to cpu-torch-latest (#5226)

loadams committed 2 years ago

Verified 3e06a154

Add fp16 support of Qwen1.5 models (0.5B to 72B) to DeepSpeed-FastGen (#5219)

ZonePG committed 2 years ago

Verified bcc617a0

[zero++] Synchronize at the end of secondary partitioning and simplify the logic (#5216)

ByronHsu committed 2 years ago

Verified 4578c249

Allow specifying MII branch on MII CI (#5208)

mrwyattii committed 2 years ago

Verified 857584fd

Fix assertion to run pipeline engine with a compiled module (#5197)

tohtana committed 2 years ago

Verified aed599b4

Update version.txt after 0.13.4 release (#5196)

mrwyattii committed 2 years ago

Verified 07a1ede4

Add script to check for `--extra-index-url` (#5184)

loadams committed 2 years ago

Verified 5115df3f

[CPUAdam] Update full_precision_optimizer_states in docstring (#5181)

rohan-varma committed 2 years ago

Verified 8c42a302

Disable ninja by default (#5194)

mrwyattii committed 2 years ago

Verified 48749411

allow debug/experimental compiler backends (#5191)

tohtana committed 2 years ago

Verified d274ebf3

Fixes for `--extra-index-url` (#5183)

loadams committed 2 years ago

Verified 5d754606

Update version.txt after v0.13.3 release (#5185)

loadams committed 2 years ago

Verified daa5a05a

Switch cpu-inference workflow from --extra-index-url to --index-url (#5182)

loadams committed 2 years ago

Verified afdf028a

Reduce ds_id name length (#5176)

jomayeri committed 2 years ago

Verified 4ec8762b

Inference V2 Human Eval (#4804)

lekurile committed 2 years ago

Verified 2a6c58df

Fix typos in blogs/ (#5172)

jinyouzhi committed 2 years ago

Verified 8d0150d9

Write multiple items to output file at once, in distributed data analyzer. (#5169)

bm-synth committed 2 years ago

Verified d5fa87ff

Fix allreduce for BF16 and ZeRO0 (#5170)

tohtana committed 2 years ago

Verified dd3690c5

Update flops profiler to handle attn and __matmul__ (#4724)

KimmiShi committed 2 years ago

Verified 98c96e79

Use ninja to speed up build (#5088)

jinzhen-lin committed 2 years ago

Verified b00533e4

Fix gradient clipping (#5150)

tohtana committed 2 years ago

Verified 005afe12

MOE: Fix save checkpoint when TP > 1 (#5157)

mosheisland committed 2 years ago

Verified a84d07c5

Newer Older