SIGN IN SIGN UP

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

158577 0 0 Python

Fix llama4 bnb mode (#44588)

* check float before using normal op

Signed-off-by: jiqing-feng <[email protected]>

* fix llama4 weight

Signed-off-by: jiqing-feng <[email protected]>

* add bnb quant skip module for llama4

Signed-off-by: jiqing-feng <[email protected]>

* revert bnb integration

Signed-off-by: jiqing-feng <[email protected]>

* revert initialization.py

Signed-off-by: jiqing-feng <[email protected]>

* total revert init

Signed-off-by: jiqing-feng <[email protected]>

* fix _keep_in_fp32_modules

Signed-off-by: jiqing-feng <[email protected]>

* add _modules_to_not_quantize

Signed-off-by: jiqing-feng <[email protected]>

* fix modules_to_not_convert

Signed-off-by: jiqing-feng <[email protected]>

* update bnb quantize condition

Signed-off-by: jiqing-feng <[email protected]>

---------

Signed-off-by: jiqing-feng <[email protected]>
J
jiqing-feng committed
ce4a791c5277840c4c1d74eed03431b674869da5
Parent: cc4ef19
Committed by GitHub <[email protected]> on 3/27/2026, 2:05:40 PM