Patch release v5.8.1

<h2>What's Changed</h2> <p>This release candidate was focused mostly on <code>quantization</code> support with the new dynamic weight loader, and a few notable 🚨 breaking

<h2>What's Changed</h2> <p>Another fix for qwen vl models that prevented correctly loading the associated model type - this works together with <a class="issue-link js-issue-link"

<p>Small patch dedicated to optimizing gemma4, fixing inference with <code>use_cache=False</code> due to k/v states sharing between layers, as well as conversion mappings for some

<h1>Release v5.20.0</h1> <h2>New Model additions</h2> <h3>Muse Spark Open</h3> <a

<h1>Patch release v5.14.1</h1> <p>This patch solves a few issues which appeared when integrating Inkling model, most notably an issue affecting models using EncoderDecoderCache

<h1>Release v5.8.0</h1> <h2>New Model additions</h2> <h3>DeepSeek-V4</h3> <a

<h1>Patch release v5.12.1</h1> <p>Updated the lower bound for PEFT and a fix for auto tokenizer to properly resolve the mistral tokenizer (when <code>mistral-common</code> is

<h1>Patch release v5.6.2</h1> <p>Qwen 3.5 and 3.6 MoE (text-only) were broken when using with FP8. It should now work again with this 🫡</p> <ul> <li>Fix configuration reading and