Update README.md
Browse files
README.md
CHANGED
|
@@ -4,6 +4,8 @@ base_model:
|
|
| 4 |
license: mit
|
| 5 |
---
|
| 6 |
|
|
|
|
|
|
|
| 7 |
## Model Description
|
| 8 |
|
| 9 |
**GLM-5.1-NVFP4** is an NVFP4-quantized version of [zai-org/GLM-5.1](https://huggingface.co/zai-org/GLM-5.1), a 744B-parameter Mixture-of-Experts language model with 40B active parameters, 256 experts per MoE layer (8 activated per token), and DeepSeek Sparse Attention (DSA).
|
|
@@ -28,8 +30,6 @@ Three calibration passes were run:
|
|
| 28 |
|
| 29 |
Hardware: 8x RTX PRO 6000 Blackwell 96GB (b12x MoE runner recommended)
|
| 30 |
|
| 31 |
-
Note: You must run with `--disable-shared-experts-fusion` in sglang, otherwise it will incorrectly attempt to fuse the BF16 shared expert.
|
| 32 |
-
|
| 33 |
### Community Testing
|
| 34 |
|
| 35 |
```
|
|
|
|
| 4 |
license: mit
|
| 5 |
---
|
| 6 |
|
| 7 |
+
Note: You must run with --disable-shared-experts-fusion in sglang, otherwise it will incorrectly attempt to fuse the BF16 shared expert.
|
| 8 |
+
|
| 9 |
## Model Description
|
| 10 |
|
| 11 |
**GLM-5.1-NVFP4** is an NVFP4-quantized version of [zai-org/GLM-5.1](https://huggingface.co/zai-org/GLM-5.1), a 744B-parameter Mixture-of-Experts language model with 40B active parameters, 256 experts per MoE layer (8 activated per token), and DeepSeek Sparse Attention (DSA).
|
|
|
|
| 30 |
|
| 31 |
Hardware: 8x RTX PRO 6000 Blackwell 96GB (b12x MoE runner recommended)
|
| 32 |
|
|
|
|
|
|
|
| 33 |
### Community Testing
|
| 34 |
|
| 35 |
```
|