dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

FIX: Fix 8-bit serialization tests (#30051) 

* fix 8-bit serialization tests

* add more clarification

* Update src/transformers/quantizers/quantizer_bnb_8bit.py

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

---------

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
This commit is contained in:
Younes Belkada 2024-04-16 12:28:10 +02:00 committed by GitHub
parent ddf5f2588f
commit b86d0f4eca
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 8 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/transformers/quantizers/quantizer_bnb_8bit.py
View File

@ -171,7 +171,10 @@ class Bnb8BitHfQuantizer(HfQuantizer):
import bitsandbytes as bnb
fp16_statistics_key = param_name.replace("weight", "SCB")
fp16_weights_format_key = param_name.replace("weight", "weight_format")
fp16_statistics = state_dict.get(fp16_statistics_key, None)
fp16_weights_format = state_dict.get(fp16_weights_format_key, None)
module, tensor_name = get_module_from_name(model, param_name)
if tensor_name not in module._parameters:
@ -210,6 +213,11 @@ class Bnb8BitHfQuantizer(HfQuantizer):
if unexpected_keys is not None:
unexpected_keys.remove(fp16_statistics_key)
# We just need to pop the `weight_format` keys from the state dict to remove unneeded
# messages. The correct format is correctly retrieved during the first forward pass.
if fp16_weights_format is not None and unexpected_keys is not None:
unexpected_keys.remove(fp16_weights_format_key)
def _process_model_after_weight_loading(self, model: "PreTrainedModel", **kwargs):
model.is_loaded_in_8bit = True
model.is_8bit_serializable = self.is_serializable