I only have 32GB DDR4 and a 8GB GPU. Despite that, I downloaded Qwen 3.8 Flash Q4 quant 93GB in size from unsloth. Guess what, it gets up to 3.5t/s on Debian, freshly compiled. Way better than expected, you can actually chat with that model. This is fucking nuts. edit: I just run llama-cli --model b
