RSI-AI V1.0 (GGUF imatrix quants)
mradermacher/RSI-AI-V1.0-i1-GGUF is a GGUF quant pack built from EpistemeAI/RSI-AI-V1.0, with imatrix weighting applied to improve quality at lower bitrates. The upstream model card is sparse (it appears to be a fine-tune in the gpt-oss-20b family with an Apache-2.0 license), so the main value here is the packaging: a wide range of ready-to-run quant variants for llama.cpp-compatible tooling.
The repository includes everything from “tiny but desperate” IQ1/IQ2 options to more practical mid-range quants. The author’s notes highlight Q4_K_S and Q4_K_M as good starting points, balancing speed, memory, and output quality. If you’re evaluating the model itself (not just the quantization), treat this as an experiment: establish a baseline on your prompts, then compare different quant types at the same approximate memory budget.
What to try first: download a Q4_K_M (or Q4_K_S if you’re tighter on VRAM/RAM), run a short prompt suite that includes both factual and instruction-following tasks, and then only move to more aggressive quants once you know what quality loss looks like for your use case.
Quick stats from the listing feed: 1603 downloads.
Source listing: https://huggingface.co/models?sort=modified