onnxruntime
legacy_megatron-lm/deepspeed_ZERO1&2 FP16_Optimizer wrapper
#9184

Merged

Commits

FP16_Optimizer Wrap

pengwa committed 4 years ago
refine the code

pengwa committed 4 years ago
allow model parallelism aggregation optional

pengwa committed 4 years ago
zero1 optimizer inefficient function override

pengwa committed 4 years ago
add deepspeed zero1 and zero2 - checkoverflow & clip norm

pengwa committed 4 years ago
refine the code

pengwa committed 4 years ago
minor fix

pengwa committed 4 years ago
refine the code after validation

pengwa committed 4 years ago
Merge branch 'master' of https://github.com/microsoft/onnxruntime into pengwa/opt_fp16_optimizer

pengwa committed 4 years ago
fix modifier

pengwa committed 4 years ago
fix build

pengwa committed 4 years ago
fix a indent

pengwa committed 4 years ago
Merge branch 'master' of https://github.com/microsoft/onnxruntime into pengwa/opt_fp16_optimizer

pengwa committed 4 years ago
Merge branch 'master' of https://github.com/microsoft/onnxruntime into pengwa/opt_fp16_optimizer

pengwa committed 4 years ago
resolve comments

pengwa committed 4 years ago
update the document

pengwa committed 4 years ago
refine comment

pengwa committed 4 years ago
refine with comments

pengwa committed 4 years ago
refine after validation

pengwa committed 4 years ago