microsoft/DeepSpeed

Pull Requests Commits

Switch torch latest CPU to latest torch

loadams committed 1 year ago

3ea47178

Revert "allow cuda mismatch exceptions to be triggered"

loadams committed 1 year ago

bbbf04a2

Fix figure of llama2 13B in DS-FlexGen blog (#4625)

tohtana committed 1 year ago

Verified de96e18a

Fix figure in FlexGen blog (#4624)

tohtana committed 1 year ago

Verified d89027be

Update version.txt after 0.12.2 release (#4617)

mrwyattii committed 1 year ago

Verified b51d15b3

allow cuda mismatch exceptions to be triggered

jeffra committed 1 year ago

4f7dd721

Update version.txt after 0.12.1 release (#4615)

mrwyattii committed 1 year ago

Verified 016b7962

Update minor CUDA version compatibility. (#4613)

cmikeh2 committed 1 year ago

Verified 3437a5b1

Add number for latency comparison (#4612)

tohtana committed 1 year ago

Verified ff53c224

Update version.txt after 0.12.0 release (#4611)

mrwyattii committed 1 year ago

Verified 58d3b65a

jeffra committed 1 year ago

Verified 09834bbe

DeepSpeed-FastGen blog (#4607)

jeffra committed 1 year ago

Verified 1d9e256c

DeepSpeed-FastGen (#4604)

cmikeh2 committed 1 year ago

Verified 38b41dff

Pin transformers in nv-inference (#4606)

loadams committed 1 year ago

Verified 737ef296

Update version.txt after 0.11.2 release (#4609)

mrwyattii committed 1 year ago

Verified 2cbfdc70

Disable CPU inference on PRs (#4590)

loadams committed 1 year ago

Verified f0604078

[docs] fix deepspeed.ai links

jeffra committed 1 year ago

4199dc25

[docs] paper updates (#4584)

jeffra committed 1 year ago

Verified 45b07bf9

Public update (#4583)

yaozhewei committed 1 year ago

Verified ff529c91

Don't check overflow for bf16 data type (#4512)

BacharL committed 1 year ago

Verified 244040c1

fix multiple definition while building evoformer (#4556)

fecet committed 1 year ago

Verified 8f168c2f

Added __HIP_PLATFORM_AMD__=1 (#4570)

rraminen committed 1 year ago

Verified 764f5b07

Remove PP Grad Tail Check (#2538)

Quentin-Anthony committed 1 year ago

Verified 3589cad8

engine.py - save_checkpoint: only rank-0 should create the save dir (#4536)

nelyahu committed 1 year ago

Verified b891ec2a

Fix the sequence-parallelism for the dense model architecture (#4530)

RezaYazdaniAminabadi committed 1 year ago

Verified ec029e76

[AutoTP] Make AutoTP work when num_heads not divisible by number of workers (#4011)

delock committed 1 year ago

Verified f15cccfa

Enable universal checkpoint for zero stage 1 (#4516)

tjruwase committed 1 year ago

Verified 8fdd9b35

Add missing RocBlas include (#4557)

loadams committed 1 year ago

Verified 869629c2

pipe engine _aggregate_total_loss: more efficient loss concatenation (#4327)

nelyahu committed 1 year ago

Verified a02de228

Fix RTD builds (#4558)

mrwyattii committed 1 year ago

Verified 0f2338f7

Older