alibaba / BladeDISC

BladeDISC is an end-to-end DynamIc Shape Compiler project for machine learning workloads.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[TorchBench] Performance Signal Detected

zzpmiracle opened this issue · comments

TorchBench CI has detected a performance signal.

Affected Tests:

  • eval-cuda-fp32:
    • attention_is_all_you_need_pytorch[dynamo-blade (latency)] 6.525 -> 5.932, +9.0881%
    • attention_is_all_you_need_pytorch[dynamo-disc (latency)] 5.521 -> 4.842, +12.2985%
    • hf_Bert[dynamo-blade (latency)] 7.911 -> 6.83, +13.6645%
    • hf_Bert[dynamo-disc (latency)] 8.191 -> 7.384, +9.8523%
  • eval-cuda-fp16:
    • attention_is_all_you_need_pytorch[dynamo-blade (latency)] 5.34 -> 4.409, +17.4345%
    • attention_is_all_you_need_pytorch[dynamo-disc (latency)] 3.537 -> 2.782, +21.3458%
    • BERT_pytorch[dynamo-disc (latency)] 16.914 -> 15.829, +6.4148%
    • hf_Bert[dynamo-blade (latency)] 4.444 -> 3.252, +26.8227%
    • hf_Bert[dynamo-disc (latency)] 4.507 -> 3.235, +28.2228%

detail data can be seen in oss://bladedisc-ci/TorchBench/gpu/partial/20230521-04
created by TorchBench CI automatically