Drop SGLang exclusion
Browse files
README.md
CHANGED
|
@@ -229,8 +229,6 @@ network.
|
|
| 229 |
|
| 230 |
#### SGLang
|
| 231 |
|
| 232 |
-
> **NVFP4 is not currently working on SGLang.** On Blackwell GPUs the model loads but its NVFP4 (W4A4) fused-MoE kernel produces NaN and degenerate output, so generations are corrupted. This is a SGLang engine issue, not a checkpoint problem: the same weights serve correctly on vLLM and TRT-LLM, and we have a fix in progress upstream. For SGLang today, serve the [FP8 checkpoint](https://huggingface.co/poolside/Laguna-S-2.1-FP8) instead.
|
| 233 |
-
|
| 234 |
The Laguna S 2.1 architecture is supported in SGLang via [sgl-project/sglang#24204](https://github.com/sgl-project/sglang/pull/24204). Quantization is detected automatically from `quantization_config`, so no extra flags are required. See the [SGLang cookbook entry](https://docs.sglang.io/cookbook/autoregressive/Poolside/Laguna-S-2.1) and the main [Laguna S 2.1 model card](https://huggingface.co/poolside/Laguna-S-2.1) for a serving recipe.
|
| 235 |
|
| 236 |
#### Transformers
|
|
|
|
| 229 |
|
| 230 |
#### SGLang
|
| 231 |
|
|
|
|
|
|
|
| 232 |
The Laguna S 2.1 architecture is supported in SGLang via [sgl-project/sglang#24204](https://github.com/sgl-project/sglang/pull/24204). Quantization is detected automatically from `quantization_config`, so no extra flags are required. See the [SGLang cookbook entry](https://docs.sglang.io/cookbook/autoregressive/Poolside/Laguna-S-2.1) and the main [Laguna S 2.1 model card](https://huggingface.co/poolside/Laguna-S-2.1) for a serving recipe.
|
| 233 |
|
| 234 |
#### Transformers
|