Просмотр исходного кода

vision 2x: cudagraph cap from sizes list (pydantic consistency); MAX env kept

Alexander Korolev 1 месяц назад
Родитель
Сommit
41f2808d7a
1 измененных файлов с 1 добавлено и 1 удалено
  1. 1 1
      recipes/dsv4-vision-aiden-2x/recipe.toml

+ 1 - 1
recipes/dsv4-vision-aiden-2x/recipe.toml

@@ -56,7 +56,7 @@ args = [
     "--max-model-len", "393216",
     "--max-model-len", "393216",
     "--max-num-seqs", "8",
     "--max-num-seqs", "8",
     "--max-num-batched-tokens", "8192",
     "--max-num-batched-tokens", "8192",
-    "--max-cudagraph-capture-size", "56",
+# capture cap comes from the sizes list (max 8) — the 56 value would need 56 in the list
     "--compilation-config", """
     "--compilation-config", """
 '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"],"cudagraph_capture_sizes":[1,2,3,4,5,6,7,8]}'""",
 '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"],"cudagraph_capture_sizes":[1,2,3,4,5,6,7,8]}'""",
     "--async-scheduling",
     "--async-scheduling",