microsoft/DeepSpeed

Pull Requests Commits

Install mpi4py requirements in accelerate flow

loadams committed 235 days ago

5b1dee2c

Pin to pre 4.0.0 versio

loadams committed 235 days ago

eb73a08e

Added Intel Gaudi to Accelerator Setup Guide (#6543)

ShifaAbu committed 244 days ago

Verified 2a56f533

Add conditional on torch version for scaled_dot_product_attention (#6517)

loadams committed 249 days ago

Verified 170b46e8

Avoid security issues of subprocess shell (#6498)

tjruwase committed 249 days ago

Verified 659f6be1

wrap include cuda_bf16.h with ifdef BF16_AVAILABLE (#6520)

oelayan7 committed 251 days ago

Verified c2748393

Revert "BF16 optimizer: Clear lp grads after updating hp grads in hook" (#6508)

nelyahu committed 252 days ago

Verified 8fa6b50b

fix environment variable export bug for MultiNodeRunner (#5878)

TideDra committed 253 days ago

Verified fc22d960

Fix the broken url link (#6500)

rogerxfeng8 committed 255 days ago

Verified 2a647c51

fix pipeline eval_batch micro_batches argument for schedule (#6484)

nelyahu committed 255 days ago

Verified 3b09d945

Op_builder->is_compatible quite warning (#6093)

terry-for-github committed 255 days ago

Verified 4f803852

HPU: add required ENV vars to acccelerator init (#6495)

nelyahu committed 256 days ago

Verified 857780a8

Update version.txt after 0.15.1 release (#6493)

loadams committed 256 days ago

Verified c210e601

Handle an edge case where `CUDA_HOME` is not defined on ROCm systems (#6488)

amorehead committed 256 days ago

Verified 10ba3dde

Safe usage of popen (#6490)

tjruwase committed 256 days ago

Verified 662a421b

Add default value to "checkpoint_folder" in "load_state_dict" of bf16_optimizer (#6446)

ljcc0930 committed 256 days ago

Verified ddd35718

DS_BUILD_OPS should build only compatible ops (#6489)

tjruwase committed 256 days ago

Verified 5d1a30c0

Fix patch for parameter partitioning in zero.Init() (#6388)

tohtana committed 256 days ago

Verified ddeb0c19

print warning if actual triton cache dir is on NFS, not just for default (#6487)

jrandall committed 256 days ago

Verified 9d17116f

DeepNVMe tutorial (#6449)

tjruwase committed 257 days ago

Verified 5df12a4a

bf16_optimizer: fixes to different grad acc dtype (#6485)

nelyahu committed 257 days ago

Verified cfc6ed37

Add workflow to build DS without torch to better test before releases (#6450)

loadams committed 262 days ago

Verified 9b7fc545

TestLowCpuMemUsage UT get device by device_name (#6397)

raza-sikander committed 262 days ago

Verified 89c4d9f5

Avoid gds build errors on ROCm (#6456)

rraminen committed 262 days ago

Verified a7ffe540

[CCL] fix condition issue in ccl.py (#6443)

YizhouZ committed 262 days ago

Verified 0cd9bf59

Change GDS to 1 AIO thread (#6459)

jomayeri committed 263 days ago

Verified f2739b4f

Allow triton==3.0.x for fp_quantizer (#6447)

siddartha-RE committed 263 days ago

Verified 4864991f

Add the accelerator setup guide link in Getting Started page (#6452)

rogerxfeng8 committed 263 days ago

Verified 405b6d5e

Store/Load CIFAR from local/offline (#6390)

raza-sikander committed 264 days ago

Verified 9bc4cd01

Dtype support check for accelerator in UTs (#6360)

raza-sikander committed 264 days ago

Verified b5cf30a0

Older