minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Model] Laguna-XS.2 Model Support (#24204)

Browse Source
This commit is contained in:
Jimmy Shong
2026-05-09 05:43:13 +08:00
committed by GitHub
parent 7b707c9222
commit 096ad02b06
12 changed files with 2023 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/docs/supported-models/generative_models.mdx
+5
View File
@@ -283,5 +283,10 @@ in the GitHub search bar.
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>sarvamai/sarvam-2</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Sarvam's Mixture-of-Experts models. The 105B variant uses MLA (Multi-head Latent Attention) and the 30B variant uses GQA, both with 128 routed experts.</td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><strong>Laguna XS.2</strong> (poolside)</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>poolside/Laguna-XS.2</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Poolside's hybrid sliding-window-attention MoE model (256 routed experts, 1 shared expert, sigmoid router) with per-layer-type RoPE (YARN on full-attention layers, default on sliding-attention layers) and a softplus per-head attention gate.</td>
</tr>
</tbody>
</table>