From 276099bd64eb5b43fe7ea7fdaeb97af31ea608e5 Mon Sep 17 00:00:00 2001 From: bong-water-water-bong <277547417+bong-water-water-bong@users.noreply.github.com> Date: Thu, 24 Sep 2026 19:23:52 -0300 Subject: [PATCH] blog: MiniMax-H3 is a video model, not an LLM the build failed to run Co-Authored-By: Claude Opus 5.5 --- blog/2026-09-24-hrx-long-prompts-and-ud-sweet-spots.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/blog/2026-09-24-hrx-long-prompts-and-ud-sweet-spots.md b/blog/2026-09-24-hrx-long-prompts-and-ud-sweet-spots.md index 0bb4ea7a..d7448478 100644 --- a/blog/2026-09-24-hrx-long-prompts-and-ud-sweet-spots.md +++ b/blog/2026-09-24-hrx-long-prompts-and-ud-sweet-spots.md @@ -46,4 +46,4 @@ For each model we measured every Unsloth Dynamic quant from 2 to 5 bits against - **UD-Q5_K_XL is the accurate pick:** under 0.01 on four of the five models, for 8-17% slower decode. Qwen3-Coder-30B-A3B is the exception: 0.012 is as close as any Unsloth file gets it. - **Skip the 2-bit files at these sizes:** KL 0.11-0.23, with the top prediction changing on 13-21% of tokens. -Two popular models are missing: gemma-4-12b-it and MiniMax-H3 do not run correctly on the llama.cpp build we measured with, so there was nothing fair to compare. Every row, with prompt speeds and ROCm, is on the [Quantization page](https://github.com/1bit-MONSTER/engine/wiki/Quantization) of the wiki. +One popular model is missing: gemma-4-12b-it does not run correctly on the llama.cpp build we measured with, so there was nothing fair to compare. (MiniMax-H3 was on our download list too, by mistake: it generates video, it is not a text model.) Every row, with prompt speeds and ROCm, is on the [Quantization page](https://github.com/1bit-MONSTER/engine/wiki/Quantization) of the wiki.