Skip to content

move ste from quant to round for nvfp4 #889

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
Oct 15, 2025
Merged

move ste from quant to round for nvfp4 #889

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
18 changes: 6 additions & 12 deletions auto_round/data_type/nvfp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@

from auto_round.data_type.fp8 import float8_e4m3fn_ste
from auto_round.data_type.register import register_dtype
from auto_round.data_type.utils import reshape_pad_tensor_by_group_size, revert_tensor_by_pad
from auto_round.data_type.utils import reshape_pad_tensor_by_group_size, revert_tensor_by_pad, round_ste
from auto_round.logger import logger


Expand All @@ -26,9 +26,9 @@ def cast_to_fp4(x):
sign = torch.sign(x)
x = torch.abs(x)

step1 = torch.round(2.0 * x) / 2.0
step2 = torch.round(x)
step3 = 2.0 * torch.round(x / 2.0)
step1 = round_ste(2.0 * x) / 2.0
step2 = round_ste(x)
step3 = 2.0 * round_ste(x / 2.0)

mask1 = x < 2.0
mask2 = x < 4.0
Expand All @@ -38,12 +38,6 @@ def cast_to_fp4(x):
return x * sign


def cast_to_fp4_ste(x):
fp4 = (cast_to_fp4(x).to(x.dtype) - x).detach() + x

return fp4


def get_reciprocal(x):
if isinstance(x, torch.Tensor):
return torch.where(x == 0, torch.zeros_like(x, dtype=x.dtype), 1.0 / x)
Expand Down Expand Up @@ -81,7 +75,7 @@ def ref_nvfp4_quant(x, global_scale, block_size=16, v=0, scale_coeff=1.0):
output_scale = get_reciprocal(scale * get_reciprocal(global_scale))
scaled_x = x.to(torch.float32) * output_scale + v
clipped_x = torch.clamp(scaled_x, -6.0, 6.0)
return (cast_to_fp4_ste(clipped_x) * get_reciprocal(output_scale)).reshape(m, n), scale
return (cast_to_fp4(clipped_x) * get_reciprocal(output_scale)).reshape(m, n), scale


@register_dtype("nv_fp4")
Expand Down Expand Up @@ -208,7 +202,7 @@ def ref_fp4_quant(x, global_scale, block_size=16, v=0, max_scale=1.0):
output_scale = get_reciprocal(scale * get_reciprocal(global_scale))
scaled_x = x.to(torch.float32) * output_scale + v
clipped_x = torch.clamp(scaled_x, -6.0, 6.0)
return (cast_to_fp4_ste(clipped_x) * get_reciprocal(output_scale)).reshape(m, n), scale
return (cast_to_fp4(clipped_x) * get_reciprocal(output_scale)).reshape(m, n), scale


@register_dtype("fp4_v2_with_global_scale")
Expand Down
Loading