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.

People should update their pytorch versions. (#10618)

comfyanonymous958a17199ac5
diff --git a/comfy/quant_ops.py b/comfy/quant_ops.py
index 873f173ed..5af6f118e 100644
--- a/comfy/quant_ops.py
+++ b/comfy/quant_ops.py
@@ -418,6 +418,10 @@ def fp8_linear(func, args, kwargs):
scale_b=scale_b,
out_dtype=out_dtype,
)
+
+ if isinstance(output, tuple): # TODO: remove when we drop support for torch 2.4
+ output = output[0]
+
if not tensor_2d:
output = output.reshape((-1, input_shape[1], weight.shape[0]))