RTX 5090 × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on RTX 5090
65.3best gen tok/s
5,328best prompt tok/s
640max context tested
9results
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 | 65.3 | 4,777 | LocalScore ↗ | 512+128 | 281 ms | llamafile 0.9.2 |
| Q4_K - Medium | 65.1 | 4,787 | LocalScore ↗ | 512+128 | 279 ms | llamafile 0.9.2 |
| Q4_K - Medium | 65.0 | 4,758 | LocalScore ↗ | 512+128 | 719 ms | llamafile 0.9.2 |
| Q4_K - Medium | 62.3 | 5,328 | LocalScore ↗ | 512+128 | 239 ms | llamafile 0.9.3 |
| Q4_K - Medium | 55.6 | 5,069 | LocalScore ↗ | 512+128 | 252 ms | llamafile 0.9.3 |
| Q4_K - Medium | 49.5 | 4,295 | LocalScore ↗ | 512+128 | 631 ms | llamafile 0.9.2 |
| Q4_K - Medium | 27.4 | 2,671 | LocalScore ↗ | 512+128 | 689 ms | llamafile 0.9.2 |
| Q4_K - Medium | 26.1 | 2,626 | LocalScore ↗ | 512+128 | 493 ms | llamafile 0.9.2 |
| Q4_K - Medium | 24.3 | 2,451 | LocalScore ↗ | 512+128 | 754 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.
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.