Skip to content

Add quantization support for bridge mode weight update #1564

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

Open
jairuigou wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add quantization support for bridge mode weight update #1564

Show file tree
Hide file tree
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
5 changes: 2 additions & 3 deletions slime/backends/megatron_utils/megatron_to_hf/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,10 +11,9 @@


# TODO unify w/ `convert_to_hf`
def postprocess_hf_param(args, megatron_param_name, hf_param_name, param):
def postprocess_hf_param(args, megatron_param_name, hf_param_name, param, quantization_config=None):
param = remove_padding(megatron_param_name, param, args.vocab_size)
# TODO support quant
return param
return quantize_params(args, megatron_param_name, [(hf_param_name, param)], quantization_config)


# TODO optimize code details
Expand Down
21 changes: 10 additions & 11 deletions slime/backends/megatron_utils/update_weight/hf_weight_iterator_bridge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,20 +27,19 @@ def get_hf_weight_chunks(self, megatron_local_weights):

named_weights = self._bridge.export_hf_weights(self.model, cpu=False, conversion_tasks=conversion_tasks)

named_weights = (
(
hf_param_name,
postprocess_hf_param(
args=self.args,
megatron_param_name=megatron_param_name,
hf_param_name=hf_param_name,
param=weight,
),
)
processed_weights = (
processed
for hf_param_name, weight, megatron_param_name in named_weights
for processed in postprocess_hf_param(
args=self.args,
megatron_param_name=megatron_param_name,
hf_param_name=hf_param_name,
param=weight,
quantization_config=self.quantization_config,
)
)

yield from chunk_named_params_by_size(named_weights, chunk_size=self.args.update_weight_buffer_size)
yield from chunk_named_params_by_size(processed_weights, chunk_size=self.args.update_weight_buffer_size)


def _process_conversion_tasks(vanilla_conversion_tasks, new_weight_dict):
Expand Down