DeepSpeed
[ROCm] Relax tolerances for FP8 unit test for fp16 and bf16 cases
#7655
Merged
Merged
[ROCm] Relax tolerances for FP8 unit test for fp16 and bf16 cases #7655
sfc-gh-truwase
merged
24 commits into
deepspeedai:master
from
rraminen:relax_tol_testFP8_ROCm
jithunnair-amd
commented
on 2025-10-30
Relax tolerance
80e6f533
ALST/UlyssesSP: more intuitive API wrt variable seqlen (#7656)
160730e4
Fix misplaced overflow handling return in fused_optimizer.py (#7645)
b01045b7
[bug]: fixed comm_dtype in extra_large_param_to_reduce (#7660)
a21a0747
UlyssesSP: TiledMLP doc - recomputes forward twice (#7664)
dd2e1474
resolved a 0-dim tensor slicing bug from _get_state_without_padding (…
56ca87be
Fix typo in pytorch-profiler.md documentation (#7652)
b3fa61f2
README refresh (#7668)
014ee5fc
Update version.txt after release (#7675)
d324f97d
[modal ci] fixes (#7676)
2d8d5238
leaf modules: explain better (#7674)
97301535
disable nv-lightning-v100.yml cI (#7681)
24990451
allow seperate learning rate "muon_lr" and "adam_lr" for muon optimiz…
08c6d1de
see_mem_usage: make always work (#7688)
a0fde72a
make debug utils more resilient (#7690)
0387a0a1
stage 1-2: don't pin memory if not configured (#7689)
c69ab198
modal ci: fix group concurrency (#7691)
7a94820a
Use pytorch utils to detect ninja (#7687)
f00b3887
Update SECURITY.md to point to GitHub reporting rather than Microsoft…
767fe524
Add Qwen2.5 to AutoTP model list (#7696)
e96064e3
Trust intel server for XPU tests (#7698)
2972ef84
PyTorch-compatible backward API (#7665)
a7ea3f6c
Merge branch 'master' into relax_tol_testFP8_ROCm
ae3ae053
sfc-gh-truwase
commented
on 2025-12-02
Apply suggestion from @sfc-gh-truwase
72d4e997
sfc-gh-truwase
approved these changes
on 2025-12-02
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub