MVP graph of realized_model_tflop/s/gpu & MFU / 已实现模型 TFLOP/s/GPU 和 MFU 的 MVP 图表
Nobody has claimed this yet.
Assessment
This issue has not been assessed yet.
Description
create these 2 graphs y-selector behind an feature flag (while we calcuate the properly formula with attention flops)
- Model TFLOP/s vs Interactivity
- MFU (percentage) vs Interactivity
- Model TFLOP/s vs e2e Latency
- MFU (percentage) vs e2e Latency
realized_model_tflops = 2 * N * tok/s/gpu
where N is num of active params
num of active params for
llama = 70 billion params
gptoss = 5.1B active params
deepseek = 37B active params
MFU = realized_model_tflops / theortical_tflops
theortical_tflops dense is
h100 fp8 = 1,979 TFLOP/s
h200 fp8 = 1,979 TFLOP/s
b200 fp8 = 4,500 TFLOP/s
b200 fp4 = 9.000 TFLOP/s
b300 fp8 = 4,500 TFLOP/s
b300 fp4 = 13,500 TFLOP/s
gb200 fp8 = 5,000TFLOP/s
gb200 fp4 = 10,000TFLOP/s
gb300 fp8 = 5,000TFLOP/s
gb300 fp4 = 15,000TFLOP/s
mi300 fp8 = 2,615 TFLOP/s
mi325 fp8 = 2,615 TFLOP/s
mi355 fp8 = 5,033 TFLOP/s
mi355 fp4 = 10,066 TFLOP/s
中文说明
跟踪所有集群上的软件版本 (BKC)。
- Dominant language
- Python
- Stars
- 1.7k
- Forks
- 303
- Avg merge
- 1d 13h
- Merged PRs (30d)
- 284
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from SemiAnalysisAI/InferenceX
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
SemiAnalysisAI/InferenceX#2125 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
SemiAnalysisAI/InferenceX#1587 ·
-
Difficulty 1/5 Under an hour Newbie friendliness 78/100
SemiAnalysisAI/InferenceX#1369 · 3 comments ·
-
Difficulty 1/5 1-3 hours Newbie friendliness 76/100
SemiAnalysisAI/InferenceX#1359 · 1 comment ·
-
Difficulty 5/5 Over a week Newbie friendliness 30/100
SemiAnalysisAI/InferenceX#3122 · 3 comments ·
All issues in SemiAnalysisAI/InferenceX
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
bancolombia/sentinel#23 ·
-
test md OpenCI
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
-
integration:quickjs org:external priority:backlog topic:code-interpreter topic:middleware type:feature
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
langchain-ai/deepagents#6450 ·
-
bug client
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100