inferencerlabs/GLM-5.2-MTP-MLX

GLM-5.2 MTP

See GLM-5.2 in action: demonstration videos

This draft model contains the Multi-Token Prediction (MTP) layers from [zai-org/GLM-5.2](https://huggingface.co/zai-org/GLM-5.2) for use alongside the GLM-5.2-MLX model as a speculative decoder.

Q4 quant typically achieves higher throughput with less RAM usage (compared to base MTP) at no loss in quality.

Tested on a M3 Ultra 512GB RAM using Inferencer app v2.0.6

<table style="border-collapse: collapse; border: none; text-align:left; margin-top:10px; margin-bottom:0px;"> <tr><td><strong>C++</strong></td><td>~77.9% correct predictions at 3 steps</td></tr> <tr><td><strong>HTML/ECMAScript</strong></td><td>~70% correct predictions at 3 steps</td></tr> <tr><td><strong>Maths</strong></td><td>~76.6% correct predictions at 3 steps</td></tr> <tr><td><strong>Encyclopedic</strong></td><td>~71.4% correct predictions at 3 steps</td></tr> <tr><td><strong>Creative Writing</strong></td><td>~46.2% correct predictions at 3 steps</td></tr> </table>

<p style="margin-bottom:0;padding-bottom:0;">Enable speculative decoding in Inference Contols:</p> <div style="margin-left:10px;"> <img src="https://cdn-uploads.huggingface.co/production/uploads/688479d616f1ec82fa645019/weJgx7wZI7VgOdQe827yS.jpeg" style="display:block;margin:0;"> </div>

Disclaimer

We are not the creator, originator, or owner of any model listed. Each model is created and provided by third parties. Models may not always be accurate or contextually appropriate. You are responsible for verifying the information before making important decisions. We are not liable for any damages, losses, or issues arising from its use, including data loss or inaccuracies in AI-generated content.