Apple M4 Max (40-core GPU) × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on Apple M4 Max (40-core GPU)
32.2best gen tok/s
378best 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 | 32.2 | 378 | LocalScore ↗ | 512+128 | 3,256 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.3 | 378 | LocalScore ↗ | 512+128 | 3,244 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.2 | 378 | LocalScore ↗ | 512+128 | 3,259 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.1 | 360 | LocalScore ↗ | 512+128 | 3,392 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.1 | 359 | LocalScore ↗ | 512+128 | 3,431 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.0 | 351 | LocalScore ↗ | 512+128 | 3,452 ms | llamafile 0.9.2 |
| Q4_K - Medium | 30.8 | 343 | LocalScore ↗ | 512+128 | 3,591 ms | llamafile 0.9.2 |
| Q4_K - Medium | 30.8 | 368 | LocalScore ↗ | 512+128 | 3,358 ms | llamafile 0.9.2 |
| Q4_K - Medium | 30.8 | 338 | LocalScore ↗ | 512+128 | 3,679 ms | llamafile 0.9.2 |
| Q4_K - Medium | 30.5 | 349 | LocalScore ↗ | 512+128 | 3,519 ms | llamafile 0.9.2 |
| Q4_K - Medium | 30.4 | 310 | LocalScore ↗ | 512+128 | 3,948 ms | llamafile 0.9.2 |
| Q4_K - Medium | 29.6 | 339 | LocalScore ↗ | 512+128 | 3,550 ms | llamafile 0.9.2 |
| Q4_K - Medium | 29.3 | 324 | LocalScore ↗ | 512+128 | 3,902 ms | llamafile 0.9.2 |
| Q4_K - Medium | 29.1 | 321 | LocalScore ↗ | 512+128 | 3,930 ms | llamafile 0.9.2 |
| Q4_K - Medium | 28.5 | 302 | LocalScore ↗ | 4,352 | 4,132 ms | llamafile 0.9.2 |
| Q4_K - Medium | 28.5 | 309 | LocalScore ↗ | 512+128 | 4,022 ms | llamafile 0.9.2 |
| Q4_K - Medium | 28.3 | 308 | LocalScore ↗ | 512+128 | 3,867 ms | llamafile 0.9.2 |
| Q4_K - Medium | 28.1 | 304 | LocalScore ↗ | 512+128 | 4,094 ms | llamafile 0.9.2 |
| Q4_K - Medium | 28.0 | 328 | LocalScore ↗ | 512+128 | 3,602 ms | llamafile 0.9.2 |
| Q4_K - Medium | 26.6 | 301 | LocalScore ↗ | 512+128 | 3,999 ms | llamafile 0.9.2 |
| Q4_K - Medium | 26.5 | 299 | LocalScore ↗ | 512+128 | 3,988 ms | llamafile 0.9.2 |
| Q4_K - Medium | 26.4 | 299 | LocalScore ↗ | 512+128 | 3,872 ms | llamafile 0.9.2 |
| Q4_K - Medium | 24.8 | 268 | LocalScore ↗ | 512+128 | 4,445 ms | llamafile 0.9.2 |
| Q4_K - Medium | 23.6 | 257 | LocalScore ↗ | 512+128 | 4,789 ms | llamafile 0.9.2 |
| Q4_K - Medium | 23.1 | 255 | LocalScore ↗ | 512+128 | 4,815 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.6 | 208 | LocalScore ↗ | 512+128 | 6,203 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 Max (40-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.