← Overview
zgcm-1-7b
- Quantization
- NVFP4
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-23
- Hardware
- NVIDIA DGX Spark, 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 33.4
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 34.7 |
| Math | 40.0 |
| Instruction following | 44.3 |
| Code | 23.0 |
| German | 24.8 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
25.3 |
± 2.5 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
32.3 |
± 3.3 |
198 |
| Knowledge & reasoning |
musr_gen |
46.4 |
± 3.2 |
250 |
| Math |
omni_math_gen |
40.0 |
± 3.1 |
250 |
| Instruction following |
ifbench |
33.0 |
± 2.7 |
300 |
| Instruction following |
ifeval |
78.7 |
± 3.4 |
150 |
| Code |
livecodebench_gen |
23.0 |
± 3.0 |
200 |
| German |
mmlu_prox_de |
21.0 |
± 2.4 |
300 |
| German |
include_de_gen |
32.5 |
± 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] |
– |
– |
33.7 |
33.7 |
33.7 |
5 |
| Prompt processing [t/s] |
512 |
526 |
3487.7 |
3466.3 |
3494.1 |
5 |
| Prompt processing [t/s] |
4096 |
4047 |
3957.9 |
3932.1 |
3965.1 |
5 |
| Prompt processing [t/s] |
16384 |
16107 |
3928.9 |
3924.4 |
3936.6 |
5 |
| Time to first token [ms] |
512 |
526 |
150.8 |
150.5 |
151.7 |
5 |
| Time to first token [ms] |
4096 |
4047 |
1022.5 |
1020.7 |
1029.2 |
5 |
| Time to first token [ms] |
16384 |
16107 |
4099.6 |
4091.6 |
4104.3 |
5 |
Measured directly at the inference server.