Update README.md
Browse files
README.md
CHANGED
|
@@ -15,7 +15,7 @@ See GLM-5.2 in action: [demonstration videos](https://youtube.com/xcreate)
|
|
| 15 |
|
| 16 |
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](https://huggingface.co/models?search=inferencerlabs/glm-5.2) model as a speculative decoder.
|
| 17 |
|
| 18 |
-
Q4 quant typically achieves higher throughput with less RAM usage at no loss in quality.
|
| 19 |
|
| 20 |
#### Tested on a M3 Ultra 512GB RAM using [Inferencer app v2.0.6](https://inferencer.com)
|
| 21 |
<table style="border-collapse: collapse; border: none; text-align:left; margin-top:10px; margin-bottom:0px;">
|
|
|
|
| 15 |
|
| 16 |
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](https://huggingface.co/models?search=inferencerlabs/glm-5.2) model as a speculative decoder.
|
| 17 |
|
| 18 |
+
Q4 quant typically achieves higher throughput with less RAM usage (compared to base MTP) at no loss in quality.
|
| 19 |
|
| 20 |
#### Tested on a M3 Ultra 512GB RAM using [Inferencer app v2.0.6](https://inferencer.com)
|
| 21 |
<table style="border-collapse: collapse; border: none; text-align:left; margin-top:10px; margin-bottom:0px;">
|