← Back to community benchmarks
Llama-3.2-3B-Instruct
Performance
16k
tokens
510.5
PP tok/s
18.3
TG tok/s
32097
TTFT (ms)
8.2
Peak mem (GB)
Hardware
Chip
M2 Pro (19c)
Memory
32 GB
GPU Cores
19
Software
oMLX
v0.2.10
macOS
macOS 15.7.1
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 741.0 | 26.1 | 6.5 GB | view |
| 4k | 713.8 | 23.9 | 6.8 GB | view |
| 8k | 633.0 | 21.6 | 7.3 GB | view |
| 16k | 510.5 | 18.3 | 8.2 GB | current |
| 32k | 364.1 | 14.2 | 10.1 GB | view |
| 64k | 229.8 | 9.6 | 13.8 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 26.1 | 1.00× |
| 2× | 25.7 | 0.98× |
| 4× | 47.3 | 1.81× |