MiniMax-M2.1-REAP-30 (imatrix GGUF quants)
mradermacher/MiniMax-M2.1-REAP-30-i1-GGUF is a packaging-and-deployment release: it’s not a new base model, but a curated set of GGUF quantizations for 0xSero/MiniMax-M2.1-REAP-30, intended for llama.cpp-style runtimes. The distinguishing feature here is the use of an importance matrix (“imatrix”) to produce weighted quants that can preserve quality better than naïve quantization at the same size.
The model card is practical: it lists many quant variants (including IQ quants), provides an imatrix file you can use to roll your own, and calls out recommendations like Q4_K_S/Q4_K_M as good size/speed/quality tradeoffs. If you want to try it quickly, pick a single quant (start at Q4_K_M if your machine can handle it), run it in your preferred GGUF runner, and compare against a static-quant baseline. If the outputs degrade too much, step up to Q6_K; if you’re memory-bound, try IQ variants and see whether they retain enough quality for your workload.
Quick stats from the listing feed: pipeline: text-generation · 2 likes · 1957 downloads.
Source listing: https://huggingface.co/models?sort=modified