Apple M3 Pro (14-core GPU) × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on Apple M3 Pro (14-core GPU)
12.2best gen tok/s
121best prompt tok/s
4,352max context tested
8results
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 | 12.2 | 121 | LocalScore ↗ | 512+128 | 10,562 ms | llamafile 0.9.2 |
| Q4_K - Medium | 12.2 | 120 | LocalScore ↗ | 4,352 | 10,669 ms | llamafile 0.9.2 |
| Q4_K - Medium | 12.0 | 119 | LocalScore ↗ | 512+128 | 10,734 ms | llamafile 0.9.2 |
| Q4_K - Medium | 11.9 | 117 | LocalScore ↗ | 512+128 | 10,755 ms | llamafile 0.9.2 |
| Q4_K - Medium | 11.9 | 117 | LocalScore ↗ | 512+128 | 10,830 ms | llamafile 0.9.2 |
| Q4_K - Medium | 11.9 | 119 | LocalScore ↗ | 512+128 | 10,790 ms | llamafile 0.9.2 |
| Q4_K - Medium | 11.8 | 115 | LocalScore ↗ | 512+128 | 11,300 ms | llamafile 0.9.2 |
| Q4_K - Medium | 9.9 | 102 | LocalScore ↗ | 4,352 | 13,165 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 M3 Pro (14-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.