Apple M1 Max (32-core GPU) × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on Apple M1 Max (32-core GPU)
21.4best gen tok/s
211best prompt tok/s
4,352max context tested
19results
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 | 21.4 | 203 | LocalScore ↗ | 512+128 | 5,978 ms | llamafile 0.9.2 |
| Q4_K - Medium | 21.1 | 205 | LocalScore ↗ | 512+128 | 5,969 ms | llamafile 0.9.2 |
| Q4_K - Medium | 21.1 | 205 | LocalScore ↗ | 512+128 | 5,969 ms | llamafile 0.9.2 |
| Q4_K - Medium | 21.0 | 207 | LocalScore ↗ | 512+128 | 5,889 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.9 | 204 | LocalScore ↗ | 512+128 | 5,986 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.9 | 199 | LocalScore ↗ | 4,352 | 6,124 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.9 | 203 | LocalScore ↗ | 512+128 | 6,040 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.7 | 203 | LocalScore ↗ | 4,352 | 6,025 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.7 | 201 | LocalScore ↗ | 512+128 | 6,061 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.6 | 204 | LocalScore ↗ | 512+128 | 6,076 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.5 | 191 | LocalScore ↗ | 512+128 | 6,516 ms | llamafile 0.9.2 |
| Q4_K - Medium | 20.4 | 205 | LocalScore ↗ | 512+128 | 5,923 ms | llamafile 0.9.2 |
| Q4_K - Medium | 19.6 | 197 | LocalScore ↗ | 512+128 | 6,040 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.7 | 211 | LocalScore ↗ | 4,352 | 5,967 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.5 | 188 | LocalScore ↗ | 512+128 | 6,177 ms | llamafile 0.9.2 |
| Q4_K - Medium | 18.1 | 178 | LocalScore ↗ | 512+128 | 7,131 ms | llamafile 0.9.2 |
| Q4_K - Medium | 17.2 | 170 | LocalScore ↗ | 512+128 | 7,410 ms | llamafile 0.9.2 |
| Q4_K - Medium | 16.8 | 162 | LocalScore ↗ | 512+128 | 7,212 ms | llamafile 0.9.2 |
| Q4_K - Medium | 11.9 | 111 | LocalScore ↗ | 512+128 | 12,252 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 M1 Max (32-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.