bigscience-workshop/Megatron-DeepSpeed

Pull Requests Commits

Integrate code from t5_main into existing code.

jaredcasper committed 4 years ago

48a5e0da

Merge branch 'main' into github-main

jaredcasper committed 4 years ago

aed2f75e

Merge branch 'add_ref' into 'main'

jaredcasper committed 4 years ago

f32a638d

added link to the pipeline papers

shoeybi committed 4 years ago

9ec547c9

Merge branch 'release_fixes' into 'main'

deepakn94 committed 4 years ago

8cfef1bf

shoeybi committed 4 years ago

50a4b5fa

Merge branch 'interleaved_bugfix' into 'main'

shoeybi committed 4 years ago

23632ee5

Small bugfix to make sure refactored code works with interleaved schedule

deepakn94 committed 4 years ago

6fd78189

Merge branch 'pipeline_refactor' into 'main'

shoeybi committed 4 years ago

3fc035d7

Addressed MR comments, mostly adding comments to code.

jaredcasper committed 4 years ago

e270f68a

Merge branch 'main' into main_dedup

Mostofa Patwary committed 4 years ago

ee7b19e7

More features added

Mostofa Patwary committed 4 years ago

d413bd5f

updated filter_ngrams.py

Mostofa Patwary committed 4 years ago

f559787d

Merge branch 'bfloat_jit' into 'main'

shoeybi committed 4 years ago

f2d64c00

removed the checks for bfloat jitting

shoeybi committed 4 years ago

d28716e8

fix one more issue

Mostofa Patwary committed 4 years ago

0c01c2fe

added parallelism for computing jaccard similaity

mpatwary committed 4 years ago

43d307d5

Fixing text generation and zeroshot eval and addressing comments.

jaredcasper committed 4 years ago

64a83fb5

Tasks seems to be working.

jaredcasper committed 4 years ago

b938ec51

pipeline code simplification

kvareddy committed 4 years ago

3b91262e

Merge branch 'extra_assertion' into 'main'

jaredcasper committed 4 years ago

2f3a2d68

Cherry-pick fix from development repo.

jaredcasper committed 4 years ago

8aa4619f

Make sure pipeline-model-parallel size is greater than 2 for interleaved schedule

deepakn94 committed 4 years ago

182841f7

Added more feature in train data deduplication

Mostofa Patwary committed 4 years ago

882683dc

Merge branch 'main_retriver_merge_ict_eval' into 'main'

jaredcasper committed 4 years ago

a6e00d97

ICT zeroshot evaluation

Mostofa Patwary committed 4 years ago

fcfd0949

Merge branch 'bfloat_fused_softmax' into 'main'

jaredcasper committed 4 years ago

c5346794

Bfloat fused softmax + fused layer norm

shoeybi committed 4 years ago

0fa7175f

Merge branch 'ninja_compilation_fix' into 'main'

jaredcasper committed 4 years ago

d9b1c681

refactored the fused kernels build

shoeybi committed 4 years ago

0d5188c1

Newer Older