← Back to community benchmarks
Huihui-Qwen3-VL-4B-Instruct-abliterated-4bit
Performance
16k
tokens
2,036
PP tok/s
52.2
TG tok/s
8048
TTFT (ms)
6.5
Peak mem (GB)
Hardware
Chip
M5 Pro (20c)
Memory
64 GB
GPU Cores
20
Software
oMLX
v0.5.1
macOS
macOS 26.5.2
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 3,221 | 99.1 | 2.9 GB | view |
| 4k | 2,969 | 84.4 | 3.7 GB | view |
| 8k | 2,502 | 69.9 | 4.5 GB | view |
| 16k | 2,036 | 52.2 | 6.5 GB | current |
| 32k | 1,407 | 35.1 | 9.8 GB | view |
| 64k | 843.8 | 21.0 | 16.6 GB | view |
| 128k | 426.3 | 11.0 | 30.5 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 99.1 | 1.00× |
| 2× | 165.2 | 1.67× |
| 4× | 263.7 | 2.66× |
| 8× | 478.0 | 4.82× |