← Overview
nemotron-3.5-lightning-dspark
- Quantization
- NVFP4
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-17 – 2026-09-18
- Hardware
- NVIDIA DGX Spark, 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 53.1
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 55.8 |
| Math | 40.4 |
| Instruction following | 46.0 |
| Code | 53.0 |
| German | 70.5 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
71.0 |
± 2.6 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
57.1 |
± 3.5 |
198 |
| Knowledge & reasoning |
musr_gen |
39.2 |
± 3.1 |
250 |
| Math |
omni_math_gen |
40.4 |
± 3.1 |
250 |
| Instruction following |
ifbench |
35.0 |
± 2.8 |
300 |
| Instruction following |
ifeval |
79.3 |
± 3.3 |
150 |
| Code |
livecodebench_gen |
53.0 |
± 3.5 |
200 |
| German |
mmlu_prox_de |
72.0 |
± 2.6 |
300 |
| German |
include_de_gen |
67.5 |
± 3.8 |
154 |
± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.
Speed
| Measure | Context (target) | Context (actual) | Median | p10 | p90 | n |
|---|
| Generation [t/s] |
– |
– |
83.8 |
80.7 |
87.5 |
5 |
| Prompt processing [t/s] |
512 |
528 |
3738.2 |
2851.3 |
3772.4 |
5 |
| Prompt processing [t/s] |
4096 |
4010 |
7236.8 |
6862.4 |
7270.2 |
5 |
| Prompt processing [t/s] |
16384 |
15960 |
7177.3 |
7129.3 |
7180.0 |
5 |
| Time to first token [ms] |
512 |
528 |
141.2 |
140.0 |
185.0 |
5 |
| Time to first token [ms] |
4096 |
4010 |
554.1 |
551.6 |
585.4 |
5 |
| Time to first token [ms] |
16384 |
15960 |
2223.7 |
2222.8 |
2238.7 |
5 |
Measured directly at the inference server.