Tag
Tim Dettmers, creator of bitsandbytes, is teasing a new quantization method that reportedly runs GLM 5.3 on a single DGX Spark at 7 tokens/s, with cautious optimism from the community.