Tags: TJX2014/Megatron-LM
Tags
chore: Fix version number Signed-off-by: Oliver Koenig <[email protected]>
Merge pull request NVIDIA#1126 from mikolajblaz/mblaz/fix-fp8-exchange Fix parallel load convergence issue by ensuring BF16 exchange for FP8 tensors
Merge branch 'ko3n1g/chore/remove-prerelease' into 'core_r0.8.0' chore: Remove prerelease See merge request ADLR/megatron-lm!1922
Merge branch 'cherry-pick-c4d12e26' into 'core_r0.7.0' Merge branch 'xuwenc/moe_gmm_infer_fix' into 'main' See merge request ADLR/megatron-lm!1519
Merge branch 'rachitg/ag' into 'core_r0.7.0.beta' allow disabling qkv or fc1 overlap See merge request ADLR/megatron-lm!1398
Merge branch 'fix_overlap_param_gather' into 'main' fix EP distopt with overlap param gather See merge request ADLR/megatron-lm!1345 (cherry picked from commit ccfeda4) ac93d847 fix EP distopt with overlap param gather bb7b4307 change golden metrics 0ff731ff Minor fix to thrown value error
Merge branch 'update_version' into 'core_r0.5.0' Update 0.5.0 version in package info See merge request ADLR/megatron-lm!1158
Update retro doc following the suggestion of Wei and Lawrence
Merge branch 'cherry-pick-25b2723a' into 'core_r0.4.0' Merge branch 'fix_bert' into 'main' See merge request ADLR/megatron-lm!1006
PreviousNext