From c876108059ec1796417bc1d9d407bd6d30931317 Mon Sep 17 00:00:00 2001 From: kijai <40791699+kijai@users.noreply.github.com> Date: Thu, 16 Oct 2025 13:08:42 +0300 Subject: [PATCH] Update fp8_e4m3fn compile warning on older arch to indicate it should now work with latest Triton https://github.com/woct0rdho/triton-windows/releases/tag/v3.5.0-windows.post21 --- nodes_model_loading.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/nodes_model_loading.py b/nodes_model_loading.py index 8683325..3d425bc 100644 --- a/nodes_model_loading.py +++ b/nodes_model_loading.py @@ -1104,7 +1104,7 @@ class WanVideoModelLoader: major, minor = torch.cuda.get_device_capability(device) log.info(f"CUDA Compute Capability: {major}.{minor}") if compile_args is not None and "e4" in quantization and (major, minor) < (8, 9): - log.warning("WARNING: Torch.compile with fp8_e4m3fn weights on CUDA compute capability < 8.9 is not supported. Please use fp8_e5m2, GGUF or higher precision instead.") + log.warning("WARNING: Torch.compile with fp8_e4m3fn weights on CUDA compute capability < 8.9 may not be supported. Please use fp8_e5m2, GGUF or higher precision instead, or check the latest triton version that adds support for older architectures https://github.com/woct0rdho/triton-windows/releases/tag/v3.5.0-windows.post21") if "scaled_fp8" in sd and "scaled" not in quantization: raise ValueError("The model is a scaled fp8 model, please set quantization to '_scaled'")