Skip to content

fp8 quantize rowwise speedup#133

Open
coreyhu wants to merge 1 commit intometa-pytorch:mainfrom
coreyhu:export-D91731095
Open

fp8 quantize rowwise speedup#133
coreyhu wants to merge 1 commit intometa-pytorch:mainfrom
coreyhu:export-D91731095

Conversation

@coreyhu
Copy link
Copy Markdown

@coreyhu coreyhu commented Feb 5, 2026

Summary:
Rowwise Quantization:
Average Speedup: 1.08x
Median Speedup: 1.04x
Max Speedup: 1.41x
Min Speedup: 0.98x
Win Rate: 94.3% (33/35 cases)
Avg Throughput Gain: +193.0 GB/s

Differential Revision: D91731095

@meta-cla meta-cla Bot added the cla signed label Feb 5, 2026
@meta-codesync
Copy link
Copy Markdown

meta-codesync Bot commented Feb 5, 2026

@coreyhu has exported this pull request. If you are a Meta employee, you can view the originating Diff in D91731095.

coreyhu added a commit to coreyhu/MSLK that referenced this pull request Feb 6, 2026
Summary:

Rowwise Quantization:
  Average Speedup:       1.08x
  Median Speedup:        1.04x
  Max Speedup:           1.41x
  Min Speedup:           0.98x
  Win Rate:             94.3% (33/35 cases)
  Avg Throughput Gain:   +193.0 GB/s

Differential Revision: D91731095
Summary:

Rowwise Quantization:
  Average Speedup:       1.08x
  Median Speedup:        1.04x
  Max Speedup:           1.41x
  Min Speedup:           0.98x
  Win Rate:             94.3% (33/35 cases)
  Avg Throughput Gain:   +193.0 GB/s

Differential Revision: D91731095
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant