[Feature]: AutoDeploy: Check Nemotron Nano V3 prefill perf
Open
@galagam is already working on this.
Since Mar 18, 2026.
AutoDeploy
feature request
- Dominant language
- Python
- Stars
- 14.7k
- Forks
- 2.8k
- Avg merge
- 2d 23h
- Merged PRs (30d)
- 489
Description
🚀 The feature, motivation and pitch
This is the local test for Nemotron Nano v3 FP8 perf -
| Concurrency | TTFT avg (ms) | Prefill TPS avg (tok/s) | TPOT avg (ms) | ||||||
|---|---|---|---|---|---|---|---|---|---|
| no_pcg | pcg | Δ | no_pcg | pcg | Δ | no_pcg | pcg | Δ | |
| 1 | 50.58 | 45.70 | -9.6% | 19,781 | 21,896 | +10.7% | 3.436 | 3.397 | -1.1% |
| 4 | 100.72 | 78.29 | -22.3% | 10,006 | 13,142 | +31.4% | 4.779 | 4.652 | -2.7% |
| 16 | 3,798.15 | 1,210.56 | -68.1% | 4,066 | 4,949 | +21.7% | 7.882 | 8.065 | +2.3% |
| 64 | 550.60 | 405.54 | -26.3% | 2,537 | 2,938 | +15.8% | 13.020 | 13.128 | +0.8% |
| 256 | 1,251.85 | 1,482.86 | +18.5% | 1,464 | 1,109 | -24.3% | 22.032 | 21.827 | -0.9% |
Looks like the prefill perf is bad when the concurrency is 16. (3798 ms without the piecewise cudagraph, 1210.56 with the piecewise cudagraph)
Alternatives
No response
Additional context
No response
Before submitting a new issue...
- Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.
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.
Assessment
This issue has not been assessed yet.