RTX A4000 × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on RTX A4000
34.8best gen tok/s
1,100best prompt tok/s
4,352max context tested
17results
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 | 34.8 | 1,086 | LocalScore ↗ | 512+128 | 1,189 ms | llamafile 0.9.2 |
| Q4_K - Medium | 34.1 | 1,100 | LocalScore ↗ | 512+128 | 1,201 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.6 | 1,076 | LocalScore ↗ | 512+128 | 1,197 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.2 | 1,062 | LocalScore ↗ | 4,352 | 1,230 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.2 | 1,044 | LocalScore ↗ | 4,352 | 1,220 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.2 | 1,052 | LocalScore ↗ | 4,352 | 1,230 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.1 | 1,057 | LocalScore ↗ | 512+128 | 1,227 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.1 | 1,050 | LocalScore ↗ | 4,352 | 1,236 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.9 | 1,040 | LocalScore ↗ | 4,352 | 1,253 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.9 | 1,045 | LocalScore ↗ | 4,352 | 1,231 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.9 | 1,040 | LocalScore ↗ | 4,352 | 1,245 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.8 | 1,029 | LocalScore ↗ | 4,352 | 1,241 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.8 | 1,028 | LocalScore ↗ | 4,352 | 1,264 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.7 | 1,002 | LocalScore ↗ | 4,352 | 1,266 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.4 | 1,016 | LocalScore ↗ | 4,352 | 1,274 ms | llamafile 0.9.2 |
| Q4_K - Medium | 26.6 | 1,018 | LocalScore ↗ | 512+128 | 1,253 ms | llamafile 0.9.3 |
| Q4_K - Medium | 18.8 | 963 | LocalScore ↗ | 4,352 | 1,325 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 RTX A4000 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.