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

model: support nvidia/LocateAnything-3B (#28958)

Browse Source 
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
This commit is contained in:
Jyothirmai Kottu
2026-06-30 00:16:42 +08:00
committed by GitHub
co-authored by Xinyuan Tong
parent 5169df70f6
commit 473a278dd1
11 changed files with 1116 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/docs/supported-models/multimodal_language_models.mdx
+6
View File
@@ -201,6 +201,12 @@ in the GitHub search bar.
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Liquid AI's vision-language model combining a SigLIP2 NaFlex vision encoder (variable resolution, native aspect ratio) with the LFM2 hybrid gated short conv + GQA language model. Supports multi-image inputs.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}></td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><strong>LocateAnything</strong> (3B)</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>nvidia/LocateAnything-3B</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>NVIDIA's visual grounding/detection model (MoonViT vision encoder + Qwen2 backbone) that emits &lt;ref&gt;label&lt;/ref&gt;&lt;box&gt;...&lt;/box&gt; outputs with coordinates normalized to [0, 1000]. Covers object detection, phrase grounding, scene-text detection, GUI grounding, and pointing.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Use <code>--trust-remote-code</code>. Set <code>skip_special_tokens=false</code> so the &lt;ref&gt;/&lt;box&gt; grounding tokens survive in the output. Constrained &lt;box&gt; decoding is opt-in and client-side: start the server with <code>--enable-custom-logit-processor</code>, then pass <code>custom_logit_processor</code> (a top-level request field) and <code>custom_params</code> (inside <code>sampling_params</code>) together — use <code>LocateAnythingBoxGrammarLogitProcessor.build_sampling_params(config)</code> to build both from the config token ids.</td>
</tr>
</tbody>
</table>