From e439214b764c78772253a4f4e6c55f74495bf2fb Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 29 Oct 2026 11:21:09 +0000 Subject: [PATCH] MXFP8: drop the claim that a forced compute dtype costs the speed (measured: same step times) Co-Authored-By: Claude Sonnet 5.5 Claude-Session: https://claude.ai/code/session_01Tw9fjWwkXVabCNSQ53xaW1 --- modules/loaders/diffusion_model/nodes.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/modules/loaders/diffusion_model/nodes.py b/modules/loaders/diffusion_model/nodes.py index e2243763..ba70e339 100644 --- a/modules/loaders/diffusion_model/nodes.py +++ b/modules/loaders/diffusion_model/nodes.py @@ -260,8 +260,6 @@ class FunUIDiffusionModelLoader(io.ComfyNode): else: _, note = mxfp8_module.quantize_state_dict(state_dict, comfy.model_management.get_torch_device()) notes.append(note) - if compute_dtype != "default": - notes.append("mxfp8: compute dtype is forced, so the matmuls run in that dtype -- memory is saved, speed is not") for line in notes[mark:]: # the status output is rarely read; the log is log.info("FunUI Diffusion Model Loader", line) # Read before the load: comfy may consume the dict.