The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Users report gfx1201 is buggy on flux with pytorch attention. (#9244)
comfyanonymous735bb4bdb186
diff --git a/comfy/model_management.py b/comfy/model_management.pyindex 9e6149d60..dc5b4711d 100644--- a/comfy/model_management.py+++ b/comfy/model_management.py@@ -321,9 +321,9 @@ try:if torch_version_numeric >= (2, 7): # works on 2.6 but doesn't actually seem to improve muchif any((a in arch) for a in ["gfx90a", "gfx942", "gfx1100", "gfx1101", "gfx1151"]): # TODO: more arches, TODO: gfx950ENABLE_PYTORCH_ATTENTION = True- if torch_version_numeric >= (2, 8):- if any((a in arch) for a in ["gfx1201"]):- ENABLE_PYTORCH_ATTENTION = True+# if torch_version_numeric >= (2, 8):+# if any((a in arch) for a in ["gfx1201"]):+# ENABLE_PYTORCH_ATTENTION = Trueif torch_version_numeric >= (2, 7) and rocm_version >= (6, 4):if any((a in arch) for a in ["gfx1201", "gfx942", "gfx950"]): # TODO: more archesSUPPORT_FP8_OPS = True