RTX 3090 × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on RTX 3090
68.1best gen tok/s
2,189best prompt tok/s
4,352max context tested
39results
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 | 68.1 | 2,187 | LocalScore ↗ | 4,352 | 588 ms | llamafile 0.9.2 |
| Q4_K - Medium | 65.1 | 2,185 | LocalScore ↗ | 512+128 | 592 ms | llamafile 0.9.2 |
| Q4_K - Medium | 63.7 | 2,135 | LocalScore ↗ | 512+128 | 601 ms | llamafile 0.9.2 |
| Q4_K - Medium | 63.6 | 2,151 | LocalScore ↗ | 512+128 | 597 ms | llamafile 0.9.2 |
| Q4_K - Medium | 63.4 | 2,131 | LocalScore ↗ | 512+128 | 610 ms | llamafile 0.9.2 |
| Q4_K - Medium | 62.0 | 2,126 | LocalScore ↗ | 512+128 | 601 ms | llamafile 0.9.2 |
| Q4_K - Medium | 61.9 | 2,003 | LocalScore ↗ | 4,352 | 636 ms | llamafile 0.9.2 |
| Q4_K - Medium | 61.8 | 2,106 | LocalScore ↗ | 512+128 | 605 ms | llamafile 0.9.2 |
| Q4_K - Medium | 61.6 | 2,083 | LocalScore ↗ | 4,352 | 619 ms | llamafile 0.9.2 |
| Q4_K - Medium | 61.0 | 1,917 | LocalScore ↗ | 512+128 | 651 ms | llamafile 0.9.2 |
| Q4_K - Medium | 60.6 | 1,911 | LocalScore ↗ | 512+128 | 652 ms | llamafile 0.9.2 |
| Q4_K - Medium | 60.5 | 2,034 | LocalScore ↗ | 512+128 | 641 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.9 | 2,005 | LocalScore ↗ | 512+128 | 640 ms | llamafile 0.9.1 |
| Q4_K - Medium | 59.8 | 2,030 | LocalScore ↗ | 4,352 | 631 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.8 | 2,010 | LocalScore ↗ | 512+128 | 641 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.7 | 1,837 | LocalScore ↗ | 512+128 | 700 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.6 | 2,049 | LocalScore ↗ | 512+128 | 631 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.4 | 1,944 | LocalScore ↗ | 512+128 | 660 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.1 | 1,876 | LocalScore ↗ | 512+128 | 676 ms | llamafile 0.9.2 |
| Q4_K - Medium | 59.0 | 2,007 | LocalScore ↗ | 512+128 | 632 ms | llamafile 0.9.2 |
| Q4_K - Medium | 58.6 | 1,878 | LocalScore ↗ | 4,352 | 679 ms | llamafile 0.9.2 |
| Q4_K - Medium | 58.4 | 1,873 | LocalScore ↗ | 4,352 | 683 ms | llamafile 0.9.2 |
| Q4_K - Medium | 58.4 | 1,938 | LocalScore ↗ | 512+128 | 657 ms | llamafile 0.9.3 |
| Q4_K - Medium | 58.3 | 1,854 | LocalScore ↗ | 512+128 | 675 ms | llamafile 0.9.2 |
| Q4_K - Medium | 58.2 | 1,821 | LocalScore ↗ | 4,352 | 699 ms | llamafile 0.9.2 |
| Q4_K - Medium | 56.7 | 1,877 | LocalScore ↗ | 512+128 | 671 ms | llamafile 0.9.2 |
| Q4_K - Medium | 55.5 | 1,884 | LocalScore ↗ | 512+128 | 684 ms | llamafile 0.9.2 |
| Q4_K - Medium | 55.2 | 1,728 | LocalScore ↗ | 512+128 | 693 ms | llamafile 0.9.3 |
| Q4_K - Medium | 53.2 | 1,882 | LocalScore ↗ | 512+128 | 665 ms | llamafile 0.9.3 |
| Q4_K - Medium | 52.3 | 1,719 | LocalScore ↗ | 512+128 | 738 ms | llamafile 0.9.2 |
| Q4_K - Medium | 52.1 | 1,705 | LocalScore ↗ | 512+128 | 740 ms | llamafile 0.9.2 |
| Q4_K - Medium | 50.7 | 2,189 | LocalScore ↗ | 512+128 | 578 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.7 | 1,957 | LocalScore ↗ | 512+128 | 653 ms | llamafile 0.9.3 |
| Q4_K - Medium | 48.7 | 1,943 | LocalScore ↗ | 512+128 | 665 ms | llamafile 0.9.3 |
| Q4_K - Medium | 48.1 | 1,916 | LocalScore ↗ | 4,352 | 675 ms | llamafile 0.9.3 |
| Q4_K - Medium | 48.0 | 1,923 | LocalScore ↗ | 4,352 | 663 ms | llamafile 0.9.3 |
| Q4_K - Medium | 40.9 | 1,852 | LocalScore ↗ | 512+128 | 680 ms | llamafile 0.9.3 |
| Q4_K - Medium | 27.1 | 1,692 | LocalScore ↗ | 512+128 | 772 ms | llamafile 0.9.2 |
| Q4_K - Medium | 13.9 | 1,383 | LocalScore ↗ | 4,352 | 934 ms | llamafile 0.9.3 |
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.