Apple M4 Pro (20-core GPU) × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on Apple M4 Pro (20-core GPU)
18.6best gen tok/s
197best prompt tok/s
4,352max context tested
26results
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 | 18.6 | 194 | LocalScore ↗ | 512+128 | 6,517 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.3 | 197 | LocalScore ↗ | 512+128 | 6,377 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.3 | 185 | LocalScore ↗ | 512+128 | 6,740 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 197 | LocalScore ↗ | 512+128 | 6,384 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 187 | LocalScore ↗ | 512+128 | 6,634 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 194 | LocalScore ↗ | 512+128 | 6,526 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.2 | 187 | LocalScore ↗ | 512+128 | 6,820 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.1 | 195 | LocalScore ↗ | 512+128 | 6,448 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.1 | 189 | LocalScore ↗ | 512+128 | 6,659 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.1 | 195 | LocalScore ↗ | 512+128 | 6,485 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 192 | LocalScore ↗ | 512+128 | 6,562 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 196 | LocalScore ↗ | 4,352 | 6,404 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 192 | LocalScore ↗ | 512+128 | 6,567 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 191 | LocalScore ↗ | 512+128 | 6,677 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 192 | LocalScore ↗ | 512+128 | 6,600 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.0 | 189 | LocalScore ↗ | 512+128 | 6,661 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.9 | 187 | LocalScore ↗ | 4,352 | 6,724 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.9 | 188 | LocalScore ↗ | 512+128 | 6,760 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.9 | 180 | LocalScore ↗ | 512+128 | 7,083 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.9 | 192 | LocalScore ↗ | 512+128 | 6,569 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.8 | 184 | LocalScore ↗ | 512+128 | 6,785 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.8 | 189 | LocalScore ↗ | 512+128 | 6,724 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.8 | 180 | LocalScore ↗ | 512+128 | 6,959 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.7 | 178 | LocalScore ↗ | 512+128 | 7,102 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.6 | 187 | LocalScore ↗ | 512+128 | 6,766 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.1 | 180 | LocalScore ↗ | 512+128 | 6,850 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 Apple M4 Pro (20-core GPU) 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.