Skip to content
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

FIX: Fix 8-bit serialization tests #30051

Merged
merged 5 commits into from
Apr 16, 2024
Merged

FIX: Fix 8-bit serialization tests #30051

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2aeaca2
fix 8-bit serialization tests
younesbelkada Apr 4, 2024
6a9cf04
Merge remote-tracking branch 'upstream/main' into fix-int8-new-st
younesbelkada Apr 8, 2024
1f7e3f6
add more clarification
younesbelkada Apr 8, 2024
8d2bbff
Merge remote-tracking branch 'upstream/main' into fix-int8-new-st
younesbelkada Apr 10, 2024
e455882
Update src/transformers/quantizers/quantizer_bnb_8bit.py
younesbelkada Apr 16, 2024
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
8 changes: 8 additions & 0 deletions src/transformers/quantizers/quantizer_bnb_8bit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,7 +171,10 @@ def create_quantized_param(
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:
Expand Down Expand Up @@ -210,6 +213,11 @@ def create_quantized_param(
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
Expand Down
Loading