DGX Spark × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on DGX Spark
20.0best gen tok/s
1,174best prompt tok/s
4,352max context tested
14results
Measured · all numbers below come from the linked sources
llama.cpp
| Quant | Gen tok/s | Prompt tok/s | Source | Context tested | TTFT | Build |
|---|---|---|---|---|---|---|
| Q4_K - Medium | 20.0 | 1,174 | LocalScore ↗ | 4,352 | 1,164 ms | llamafile 0.9.2 |
| Q4_K - Medium | 19.7 | 1,156 | LocalScore ↗ | 512+128 | 1,184 ms | llamafile 0.9.2 |
| Q4_K - Medium | 19.1 | 1,151 | LocalScore ↗ | 512+128 | 1,183 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.8 | 1,152 | LocalScore ↗ | 512+128 | 1,186 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.8 | 1,149 | LocalScore ↗ | 512+128 | 1,187 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.8 | 1,145 | LocalScore ↗ | 512+128 | 1,196 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.8 | 1,138 | LocalScore ↗ | 512+128 | 1,197 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.7 | 1,157 | LocalScore ↗ | 512+128 | 1,179 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 1,106 | LocalScore ↗ | 512+128 | 1,240 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 1,078 | LocalScore ↗ | 512+128 | 1,260 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.9 | 1,102 | LocalScore ↗ | 512+128 | 1,243 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.6 | 1,072 | LocalScore ↗ | 512+128 | 1,260 ms | llamafile 0.9.2 |
| Q4_K - Medium | 16.6 | 1,014 | LocalScore ↗ | 512+128 | 1,352 ms | llamafile 0.9.2 |
| Q4_K - Medium | 15.8 | 1,018 | LocalScore ↗ | 512+128 | 1,259 ms | llamafile 0.9.2 |
Ollama
No measured Ollama results yet. Submit yours →
MLX
No measured MLX results yet. Submit yours →
"Context tested" is the largest prompt+generation length actually benchmarked (tokens), not a memory-fit estimate.
What else can DGX Spark run? →
Includes results from LocalScore, a Mozilla Builders project by cjpais, mirrored in the cn0303/local-llm-speed-benchmarks dataset. LocalScore runs use llamafile, which is built on llama.cpp.