Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Better Hidream O1 mem usage factor for non dynamic vram. (#13864)

comfyanonymous2bd65f2091f0
diff --git a/comfy/supported_models.py b/comfy/supported_models.py
index 8d2e02f68..1e4434fd5 100644
--- a/comfy/supported_models.py
+++ b/comfy/supported_models.py
@@ -1443,7 +1443,7 @@ class HiDreamO1(supported_models_base.BASE):
}
latent_format = latent_formats.HiDreamO1Pixel
- memory_usage_factor = 0.6
+ memory_usage_factor = 0.033
# fp16 not supported: LM MLP down_proj activations fp16 overflow, causing NaNs
supported_inference_dtypes = [torch.bfloat16, torch.float32]