fix draft cuda graph capture failure (#3431)

sgl-project · Feb 9, 2025 · 1646149 · 1646149
1 parent bc72e5b
commit 1646149
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/python/sglang/srt/speculative/eagle_draft_cuda_graph_runner.py b/python/sglang/srt/speculative/eagle_draft_cuda_graph_runner.py
@@ -85,6 +85,7 @@ def __init__(self, eagle_worker: EAGLEWorker):
                 "1. disable cuda graph by --disable-cuda-graph\n"
                 "2. set --mem-fraction-static to a smaller value (e.g., 0.8 or 0.7)\n"
                 "3. disable torch compile by not using --enable-torch-compile\n"
+                "4. specify --dtype to the same dtype (e.g. bfloat16)\n"
                 "Open an issue on GitHub https://github.com/sgl-project/sglang/issues/new/choose \n"
             )