Qwen3.8 Flash Quants

~20–30GB smaller than Unsloth/AesSedai Q4 at similar PPL. After several days of testing I released a set of mainline-compatible imatrix quants for Qwen3.8-Flash-Next. Goal: same quality band as the popular Unsloth / AesSedai Q4 builds, less disk and RAM. Savings are roughly 20–30GB depending on the file you compare against. PPL is in the model card and is competitive with both. Repo: huggingface.co/agentionai/Qwen3.8-Flash-Next-AP-GGUF Q4 quants are the ones I would start with. Q3 and Q5 are coming. Recipe is per-layer / tailored, not a blanket lower bpw. AMD / Strix Halo: separate ROCmFP4 build that is a bit better and faster than the Q4_XS on that hardware. ( huggingface.co/agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF ) If you try it, post your quant, RAM/VRAM, tok/s, and whether quality felt on par with Unsloth IQ4_XS / Q4_K. That is the comparison I care about.