microsoft/DeepSpeed

Pull Requests Commits

Pin deepspeed, not transformers

loadams committed 349 days ago

bcb86a73

loadams committed 349 days ago

4b50e4c1

Pin version of transformers

loadams committed 349 days ago

b81ac210

Fix windows blog examples (#6934)

loadams committed 349 days ago

Verified 53fb5795

Fix building on Windows with presence of Triton (#6749)

woct0rdho committed 349 days ago

Verified b62c84d8

Update python version classifiers (#6933)

loadams committed 349 days ago

Verified 6628127a

Use `torch.log1p` (#6930)

kit1980 committed 350 days ago

Verified c41b0c28

inference: remove unused _validate_args function (#5505)

nelyahu committed 350 days ago

Verified c7f30322

Check transformers version in BLOOM for inference v1 (#6766)

lekurile committed 351 days ago

Verified f2cc8090

Cleanup ops/transformer/inference tests (#6925)

loadams committed 351 days ago

Verified c348c5b1

Reduce the device bubble introduced by heavy loop synchronization in coalesced fetch/release(z3_leaf_module) (#6694)

inkcherry committed 351 days ago

Verified b0040b6c

Add fp8_gemm fallback for non-triton systems (#6916)

oelayan7 committed 351 days ago

Verified c5e48f49

[BUG FIX]:fix get torch.version.cuda error when cuda is None in rocm (#6909)

kairos-yu committed 351 days ago

Verified f8c9f314

Fix `checkpointable_layers` Logic (#6881)

Quentin-Anthony committed 354 days ago

Verified 0dbbb70b

Cleanup ops/transformer/inference tests (#6830)

loadams committed 354 days ago

Verified a8ede3a9

Stage3: Use new torch grad accumulation hooks API (#6773)

deepcharm committed 354 days ago

Verified 456c9ac6

Change compile for pipeline module torch.compile (#6478)

NirSonnenschein committed 358 days ago

Verified 3573858e

Update Gaudi2 jobs to latest 1.19 build (#6905)

raza-sikander committed 362 days ago

Verified cc03c76d

Fix error caused by all_reduce call in domino (#6880)

hwchen2017 committed 362 days ago

Verified 85cc5f9b

hpu_accelerator: use torch.use_deterministic_algorithms (#6897)

nelyahu committed 1 year ago

Verified eea53048

Zero2: avoid graph breaks in torch.compile by using param_idx (#6803)

nelyahu committed 1 year ago

Verified 00ea0c46

Allow to compile collective for PT>2.3 (#6899)

NirSonnenschein committed 1 year ago

Verified 4fd79205

Update version.txt after 0.16.2 release (#6893)

loadams committed 1 year ago

Verified f9e158a0

Update code owners (#6890)

tjruwase committed 1 year ago

Verified b344c04d

Add arctic model support by adding w2 to all_reduce (#6856)

pi314ever committed 1 year ago

Verified 0b25630a

Don't error out when cpu accelerator doesn't have torch (as default for whl building) (#6886)

loadams committed 1 year ago

Verified 4cd1d974

Update transformers ops unit tests to use `requried_torch_version` (#6884)

loadams committed 1 year ago

Verified 2f32966b

Fix --enable_each_rank_log when used with PDSH multi-node runner (#6863)

akeshet committed 1 year ago

Verified a964e435

Add MLP/lm_head tp grain size setting. (#6828)

Yejing-Lai committed 1 year ago

Verified da771ed4

Remove pin from transformers version and fix Processing/Threading issues in tests (#6822)

loadams committed 1 year ago

Verified 87c65068

Older