Skip to content

Don't try fp8 matrix mult in quantized ops if not supported by hardware. #10874

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 25, 2025
Merged

Don't try fp8 matrix mult in quantized ops if not supported by hardware. #10874

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions comfy/ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -646,11 +646,12 @@ def forward(self, input, *args, **kwargs):
return MixedPrecisionOps

def pick_operations(weight_dtype, compute_dtype, load_device=None, disable_fast_fp8=False, fp8_optimizations=False, scaled_fp8=None, model_config=None):
fp8_compute = comfy.model_management.supports_fp8_compute(load_device) # TODO: if we support more ops this needs to be more granular

if model_config and hasattr(model_config, 'layer_quant_config') and model_config.layer_quant_config:
logging.info(f"Using mixed precision operations: {len(model_config.layer_quant_config)} quantized layers")
return mixed_precision_ops(model_config.layer_quant_config, compute_dtype)
return mixed_precision_ops(model_config.layer_quant_config, compute_dtype, full_precision_mm=not fp8_compute)

fp8_compute = comfy.model_management.supports_fp8_compute(load_device)
if scaled_fp8 is not None:
return scaled_fp8_ops(fp8_matrix_mult=fp8_compute and fp8_optimizations, scale_input=fp8_optimizations, override_dtype=scaled_fp8)

Expand Down