ByteShape Qwen3.6-35B-A3B: 30% faster than Unsloth IQ on 6GB VRAM laptop

Article automatically generated from technical news.

A few days ago I posted about my experiments with MTP on a 6GB VRAM laptop. That didn't work so well; CPU offload hurts MTP performance badly. But now I've tried out the [new ByteShape quants](https://byteshape.com/blogs/Qwen3.6-35B-A3B/) for Qwen3.6-35B-A3B that are claimed to be both smaller and f