From 9f3a5962e6606e3c04543f315605ca8bddf4f067 Mon Sep 17 00:00:00 2001 From: kijai <40791699+kijai@users.noreply.github.com> Date: Fri, 15 Aug 2025 14:56:09 +0300 Subject: [PATCH] Update nodes_model_loading.py --- nodes_model_loading.py | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/nodes_model_loading.py b/nodes_model_loading.py index 7a2158d..1ed6494 100644 --- a/nodes_model_loading.py +++ b/nodes_model_loading.py @@ -750,13 +750,6 @@ class WanVideoModelLoader: compile_args=None, attention_mode="sdpa", block_swap_args=None, lora=None, vram_management_args=None, vace_model=None, fantasytalking_model=None, multitalk_model=None, fantasyportrait_model=None): assert not (vram_management_args is not None and block_swap_args is not None), "Can't use both block_swap_args and vram_management_args at the same time" - - if torch.cuda.is_available(): - #only warning for now - 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("Torch.compile with fp8_e4m3fn weights on CUDA compute capability < 8.9 is not supported. Please use fp8_e5m2, GGUF or higher precision instead.") lora_low_mem_load = merge_loras = False if lora is not None: @@ -824,6 +817,13 @@ class WanVideoModelLoader: if "scaled_fp8" in sd: quantization = "fp8_e5m2_scaled" break + + if torch.cuda.is_available(): + #only warning for now + 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("Torch.compile with fp8_e4m3fn weights on CUDA compute capability < 8.9 is not supported. Please use fp8_e5m2, GGUF or higher precision instead.") if "scaled_fp8" in sd and "scaled" not in quantization: raise ValueError("The model is a scaled fp8 model, please set quantization to '_scaled'")