Skip to content

[quantization] Microscaling (MX) Quantization for LayerNorm in Qwen3-vl #723

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

Draft
Torrero wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

[quantization] Microscaling (MX) Quantization for LayerNorm in Qwen3-vl #723

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
17 changes: 17 additions & 0 deletions tico/quantization/config/builders.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@
from tico.quantization.wrapq.dtypes import DType
from tico.quantization.wrapq.observers.base import ObserverBase
from tico.quantization.wrapq.observers.minmax import MinMaxObserver
from tico.quantization.wrapq.observers.mx import MXObserver
from tico.quantization.wrapq.qscheme import QScheme


Expand Down Expand Up @@ -533,6 +534,22 @@ def _build_qwen3_vl_norm_override(
]:
override[obs_name] = {"qscheme": norm_qscheme}

# LayerNorm observers that benefit from microscaling
mx_observers = [
"inv_std",
"act_in",
"centered",
"square",
"act_out",
"norm",
]
for obs_name in mx_observers:
override[obs_name] = {
"observer": MXObserver,
"elem_format": "int8",
"axis": 1,
}

if norm_weight_dtype is not None:
weight_qscheme = auto_qscheme_for(norm_weight_dtype, "weight")
override["weight"] = {
Expand Down
Loading