[GH-ISSUE #2] Please complete the benchmark desc in the readme #2

New issue

Closed

opened 2026-10-26 21:09:55 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:09:55 +00:00

Owner

Originally created by @jcao-ai on GitHub (Feb 26, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/2

Hi, deepseek team,
This is really a good job. But at the first glance, the benchmark confuses me a bit:

Does Computation column mean the total computation (repeated time) ? or what is this column for ?
What is the baseline of this benchmark ?
Specifically, it is better to describe the quantization strategy related to the performance result in doc for people to get a better understanding. (ie. group-wise/tensor-wise/token-wise, group-size=128/256/512...)

Thank you.

Originally created by @jcao-ai on GitHub (Feb 26, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/2 Hi, deepseek team, This is really a good job. But at the first glance, the benchmark confuses me a bit: 1. Does `Computation` column mean the total computation (repeated time) ? or what is this column for ? 2. What is the baseline of this benchmark ? 3. Specifically, it is better to describe the `quantization` strategy related to the performance result in `doc` for people to get a better understanding. (ie. `group-wise`/`tensor-wise`/`token-wise`, `group-size=128/256/512...`) Thank you.

bsakaguchi closed this issue

2026-10-26 21:09:55 +00:00

bsakaguchi referenced this issue

2026-10-26 21:14:54 +00:00

[PR #28] [CLOSED] Comment more clear, memory consistency model is normal memory model #211

bsakaguchi referenced this issue

2026-10-26 21:14:55 +00:00

[PR #29] [CLOSED] Remove if else check in the loop #212

bsakaguchi referenced this issue

2026-10-26 21:15:03 +00:00

[PR #61] [CLOSED] Add kernel for fp8 gemm with 1x128 LHS scaling and 1x128 RHS scaling. #224