Skip to content

remove hf_grouped lora error #75

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

Merged
Jintao-Huang merged 3 commits into from
May 13, 2026
Merged

remove hf_grouped lora error #75

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
06894da
support hf_grouped lora
Jintao-Huang May 13, 2026
ed1f3cd
fix
Jintao-Huang May 13, 2026
674a48b
update
Jintao-Huang May 13, 2026
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: 11 additions & 6 deletions src/mcore_bridge/bridge/gpt_bridge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,7 @@ class GPTBridge:
hf_expert_bias_key = 'gate.e_score_correction_bias'
additional_dim0_keys = set()
additional_dim1_keys = set()
_support_hf_grouped_lora = True

def __init__(self, config: ModelConfig):
self.config = config
Expand Down Expand Up @@ -938,9 +939,11 @@ def _set_mlp_state(
dist.all_reduce(is_lora, group=self.pp_group)
if is_lora:
if hf_grouped:
raise ValueError('Since this model\'s transformers and megatron have different expert '
'weight organization methods, LoRA weight conversion is not supported. '
'You can solve this issue by setting `--merge_lora true`.')
logger.warning_once(
'Since this model\'s transformers and megatron have different expert weight organization '
'methods, LoRA weights may not be available for inference. It is recommended to set '
Copy link
Copy Markdown

@gemini-code-assist gemini-code-assist Bot May 13, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

There's a minor formatting issue in the warning message. The period for --merge_lora true should be outside the backticks, and a space is needed before the next sentence for better readability.

Suggested change
'methods, LoRA weights may not be available for inference. It is recommended to set '
'methods, LoRA weights may not be available for inference. It is recommended to set `--merge_lora true`. '

'`--merge_lora true`. You can also manually merge LoRA weights using the '
'`megatron export` command.')
if mg_mlp is None:
lora_A = None
lora_B = None
Expand Down Expand Up @@ -1166,9 +1169,11 @@ def _set_mlp_state(
dist.all_reduce(is_lora, group=self.pp_group)
if is_lora:
if hf_grouped:
raise ValueError('Since this model\'s transformers and megatron have different expert '
'weight organization methods, LoRA weight conversion is not supported. '
'You can solve this issue by setting `--merge_lora true`.')
logger.warning_once(
'Since this model\'s transformers and megatron have different expert weight organization '
Copy link
Copy Markdown

@gemini-code-assist gemini-code-assist Bot May 13, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

This warning message is duplicated from earlier in the function (around line 944). To improve maintainability, consider extracting the message into a shared constant.

Additionally, this instance has the same formatting issue as the first one: the period should be outside the backticks for --merge_lora true, and a space is needed before the next sentence.

Suggested change
'Since this model\'s transformers and megatron have different expert weight organization '
'methods, LoRA weights may not be available for inference. It is recommended to set `--merge_lora true`. '

'methods, LoRA weights may not be available for inference. It is recommended to set '
'`--merge_lora true`. You can also manually merge LoRA weights using the '
'`megatron export` command.')
if mg_mlp is None:
lora_A = None
lora_B = None
Expand Down
Loading