Commits

Commits on Apr 29, 2024

Un-pin torch version in nv-torch-latest back to latest and skip test_compile_zero tests on v100 (#5459)
loadams
authored
f32ad3e
fix: swapping order of parameters in create_dir_symlink method. (#5465)

authored
059bb20

Commits on Apr 24, 2024

Add getter and setter methods for compile_backend across accelerators. (#5299)

authored
fa8458b

Commits on Apr 23, 2024

Update version.txt after 0.14.2 release (#5458)

mrwyattii
and
loadams
authored
fbdf0ea
Update PyTest torch version to match PyTorch latest official (2.3.0) (#5454)
loadams
authored
5f631ab
Update README.md (#5453)

Jhonso7393
and
loadams
authored
ad20279
OptimizedLinear implementation (#5355)

authored
5e6c9b9

Commits on Apr 22, 2024

set the default to use set_to_none for clearing gradients in BF16 optimizer. (#5434)

inkcherry
and
loadams
authored
c66bc42
Improve parallel process of universal checkpoint conversion (#5343)

tohtana
and
tjruwase
authored
c292b03
64bit indexing fused adam (#5187)

authored
9b6ef9e

Commits on Apr 20, 2024

add device config env for the accelerator (#5396)

authored
3f875d9

Commits on Apr 19, 2024

Fix sorting of shard optimizer states files for universal checkpoint (#5395)

tohtana
and
tjruwase
authored
99951ca
Selective dequantization (#5375)

authored
c632ea0

Commits on Apr 18, 2024

Parallel map step for `DistributedDataAnalyzer` map-reduce (#5291)

authored
64defe6
Bug fix in `split_index` method (#5292)

bm-synth
and
loadams
authored
aaaf8bc
Add required paths to trigger AMD tests on PRs (#5406)

loadams
and
root
authored
3194fe8

Commits on Apr 17, 2024

[CPU] Support SHM based inference_all_reduce in TorchBackend (#5391)

authored
b22706a
Add more tests into XPU CI (#5427)
Liangliang-Ma
authored
4d4c99d

Commits on Apr 16, 2024

Update _sidebar.scss (#5293)

fasterinnerlooper
and
loadams
authored
a9cbd68
Update engine.py to avoid torch warning (#5408)

authored
bc0f774
Update 'create-pr' action in release workflow to latest (#5415)
loadams
authored
34969d2
Fix a convergence issues in TP topology caused by incorrect grad_norm. (#5411)

inkcherry
and
conglongli
authored
0896503
[XPU] Use non_daemonic_proc by default on XPU device (#5412)

yisustc
and
loadams
authored
258e500

Commits on Apr 15, 2024

Remove dtype(fp16) condition check for residual_add unit test (#5329)

authored
8949105
Update version.txt after 0.14.1 release (#5413)

mrwyattii
and
loadams
authored
2b3d31f
Fix the FP6 kernels compilation problem on non-Ampere GPUs. (#5333)

authored
e3d873a

Commits on Apr 14, 2024

stage3: efficient compute of scaled_global_grad_norm (#5256)

nelyahu
and
tjruwase
authored
54c0687
fix pagable h2d memcpy (#5301)

authored
7b5b066

Commits on Apr 13, 2024

Removal of cuda hardcoded string with get_device function (#5351)

raza-sikander
and
loadams
authored
f69f884
Add custom reshaping for universal checkpoint (#5390)

tohtana
and
tjruwase
authored
2c51aba

Commits on Apr 10, 2024

Optimize the fp-dequantizer to get high memory-BW utilization (#5373)

RezaYazdaniAminabadi
and
sfc-gh-reyazda
authored
a8b8215