onnxruntime
Implement reuse past and present tensor in Attention Ops.
#13791

Merged

Commits

Implement reuse kv_cache for past and present tensor in Attention Ops.

zhanghuanrong committed 3 years ago
Python file format

zhanghuanrong committed 3 years ago
Fix vocab_size not set issue which cause random crash.

zhangleihuanrong committed 3 years ago
Merge branch 'main' into zhalei/attention_reuse_past_present

zhangleihuanrong committed 3 years ago
according pr feedbacks.

zhanghuanrong committed 3 years ago
correct some on the LaunchAddBiasTransAppendKvToPresent

zhanghuanrong committed 3 years ago
support past present share buffer for attention when bias is nullptr.

zhanghuanrong committed 3 years ago
Merge branch 'main' into zhalei/attention_reuse_past_present

zhanghuanrong committed 3 years ago
Some condition check

zhanghuanrong committed 3 years ago
optimized matrix number on qkv

zhanghuanrong committed 3 years ago
Fix CI build issues

zhanghuanrong committed 3 years ago
another python lint error

zhanghuanrong committed 3 years ago
Make past_present_share_buffer change do put into result model.

zhanghuanrong committed 3 years ago
fix prefast warnings.

zhanghuanrong committed 3 years ago
update op docs.

zhanghuanrong committed 3 years ago
Using CI pipelines doc gen, local one seems not good.

zhanghuanrong committed 3 years ago