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

Disable multi-threaded load by default when prefetch is on (#30146)

Browse Source
This commit is contained in:
Mohammad Miadh Angkad
2026-07-09 00:28:53 -07:00
committed by GitHub
parent 64e2a73c80
commit 666a09fe2a
5 changed files with 224 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/docs/advanced_features/model_loading.mdx
+2 -2
View File
@@ -130,7 +130,7 @@ python -m sglang.launch_server \
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>auto</code> / <code>safetensors</code> / <code>pt</code> / <code>npcache</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>enable_multithread_load</code> (bool)</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Read weight shards with a thread pool instead of sequentially.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Read weight shards with a thread pool instead of sequentially. Disabled by default when <code>--weight-loader-prefetch-checkpoints</code> is set (to avoid I/O oversubscription with the prefetch threads); set this to <code>true</code> to opt back in.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>true</code></td>
</tr>
<tr>
@@ -190,7 +190,7 @@ Top-level arguments that tune how safetensors weights are read, independent of `
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>--weight-loader-prefetch-checkpoints</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Prefetch checkpoint files into the OS page cache before loading. Each rank prefetches a fraction of the shards, cutting total network I/O on shared filesystems (NFS/Lustre) from N×checkpoint to 1×checkpoint. Recommended for models on network storage.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Prefetch checkpoint files into the OS page cache before loading. Each rank prefetches a fraction of the shards, cutting total network I/O on shared filesystems (NFS/Lustre) from N×checkpoint to 1×checkpoint. Recommended for models on network storage. When enabled, multi-threaded safetensors loading is disabled by default to avoid I/O oversubscription with the prefetch threads; set <code>enable_multithread_load=true</code> in <code>--model-loader-extra-config</code> to keep multi-threaded loading (e.g. on local NVMe where prefetch is a no-op).</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>off</td>
</tr>
<tr>