Skip to content

[Minor] fix: do not requantize the scales in FP8 scale sweep calibration #825

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

Open
Fridah-nv wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Minor] fix: do not requantize the scales in FP8 scale sweep calibration #825

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ce8c562
fix FP8 sweep, do not requantize the scales
Fridah-nv Jan 28, 2026
1caa24f
default skip_scale_quant to False
Fridah-nv Jan 28, 2026
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
4 changes: 2 additions & 2 deletions modelopt/torch/quantization/model_calib.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -245,10 +245,10 @@ def mse_calibrate(
step_size: Step size for amax search (default: 0.1).
start_multiplier: Starting multiplier for amax search (default: 0.25).
stop_multiplier: Ending multiplier for amax search (default: 4.0).
fp8_scale_sweep: If True, sweep over all 128 possible FP8 E4M3 scale values
fp8_scale_sweep: If True, sweep over all 126 valid FP8 E4M3 scale values
for NVFP4 per-block quantization instead of using multipliers.
This is specifically designed for optimizing the FP8-quantized
per-block scales in NVFP4 format (default: False).
per-block scales in NVFP4 format.

See :class:`MseCalibConfig <modelopt.torch.quantization.config.MseCalibConfig>` for
details on the remaining arguments.
Expand Down
2 changes: 1 addition & 1 deletion modelopt/torch/quantization/nn/modules/tensor_quantizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -768,7 +768,7 @@ def _fake_quantize(self, inputs):
inputs,
None, # scale
None, # scale_fp8_quant_amax
False, # skip_scale_quant
True, # skip_scale_quant
inputs.dtype, # out_dtype
self._pass_through_bwd, # pass_through_bwd
amax, # amax
Expand Down
Loading