LLM Inference on WebLLM macOS Apple M2 16GB unified memory
46.4Decode Throughput (tok/s)Qwen2.5-0.5B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5-0.5BBrowser=Chrome 143, Backend=Metal, Quantization=q4f162026.02 | 46.4 | 510 | |
| Qwen2.5-0.5BBrowser=Safari 26.2, Backend=Metal, Quantization=q4f162026.02 | 41.7 | 257 | |
| Qwen2.5-1.5BBrowser=Chrome 143, Backend=Metal, Quantization=q4f162026.02 | 36 | 225 | |
| Qwen2.5-1.5BBrowser=Safari 26.2, Backend=Metal, Quantization=q4f162026.02 | 29.7 | 93 | |
| Qwen2.5-0.5BBrowser=Firefox 147, Backend=Metal, Quantization=q4f162026.02 | 9.6 | 77 | |
| Qwen2.5-1.5BBrowser=Firefox 147, Backend=Metal, Quantization=q4f162026.02 | 9.6 | 58 |