Rowwise F8F8BF16 GEMMs - Auto-generate kernel library, auto-generated heuristics cache, add to FBGEMM quantize_bench #3210
Conversation
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D63744054 |
… heuristics cache, add to FBGEMM quantize_bench (pytorch#3210) Summary: # Summary - Auto-generated F8F8BF16 Rowwise Scaled Kernels. - Auto-generation of Heuristic Cache. - Add to quantize_bench # Performance Improvements ## DisaggBench Cultass Prefill B=1 T=2048: Elapsed: 109.13ms FLOPs: 333.74TF/s Prefill B=1 T=4928: Elapsed: 272.55ms FLOPs: 338.62TF/s Prefill B=1 T=6336: Elapsed: 354.93ms FLOPs: 342.55TF/s Prefill B=1 T=8192: Elapsed: 468.64ms FLOPs: 346.06TF/s Cultass extensions Prefill B=1 T=2048: Elapsed: 108.83ms FLOPs: 334.66TF/s Prefill B=1 T=4928: Elapsed: 260.46ms FLOPs: 354.34TF/s Prefill B=1 T=6336: Elapsed: 336.39ms FLOPs: 361.43TF/s Prefill B=1 T=8192: Elapsed: 442.64ms FLOPs: 366.39TF/s Differential Revision: D63744054
manishucsd
force-pushed
the
export-D63744054
branch
from
8d65a52 to
33855a3
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63744054 |
|
Hi @manishucsd! Thank you for your pull request. We require contributors to sign our Contributor License Agreement, and yours needs attention. You currently have a record in our system, but the CLA is no longer valid, and will need to be resubmitted. ProcessIn order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with If you have received this in error or have any questions, please contact us at [email protected]. Thanks! |
Summary:
Summary
Performance Improvements
DisaggBench
Cultass
Prefill B=1 T=2048: Elapsed: 109.13ms FLOPs: 333.74TF/s
Prefill B=1 T=4928: Elapsed: 272.55ms FLOPs: 338.62TF/s
Prefill B=1 T=6336: Elapsed: 354.93ms FLOPs: 342.55TF/s
Prefill B=1 T=8192: Elapsed: 468.64ms FLOPs: 346.06TF/s
Cultass extensions
Prefill B=1 T=2048: Elapsed: 108.83ms FLOPs: 334.66TF/s
Prefill B=1 T=4928: Elapsed: 260.46ms FLOPs: 354.34TF/s
Prefill B=1 T=6336: Elapsed: 336.39ms FLOPs: 361.43TF/s
Prefill B=1 T=8192: Elapsed: 442.64ms FLOPs: 366.39TF/s
Differential Revision: D63744054