[Bug] When making minimax m2.7 hf checkpoint to torch_dist format, ran into error

[Lynnzake]

Bug Description

when i try to convert minimax m2.7's hf checkpoint to torch_dist format with the following command,

source /nfs-152/disk6/tujie/train_env_slime/slime/scripts/models/minimax-m2.7.sh

echo "开始将 HuggingFace 格式转换为 torch-dist 格式..."
echo "输入路径: ${HF_MODEL_PATH}"
echo "输出路径: ${TORCH_DIST_OUTPUT_PATH}"

# 执行转换（单机 8 卡）
PYTHONPATH=/root/Megatron-LM/:$(pwd) torchrun \
   --nproc-per-node 8 \
   --master-addr localhost \
   --master-port 12345 \
   --nnodes=1 \
   --node-rank 0 \
   tools/convert_hf_to_torch_dist.py \
   ${MODEL_ARGS[@]} \
   --hf-checkpoint ${HF_MODEL_PATH} \
   --save ${TORCH_DIST_OUTPUT_PATH}

I ran into the following error:

[rank3]: Traceback (most recent call last):
[rank3]:   File "/nfs-152/disk6/tujie/train_env_slime/slime/tools/convert_hf_to_torch_dist.py", line 146, in <module>
[rank3]:     main()
[rank3]:   File "/nfs-152/disk6/tujie/train_env_slime/slime/tools/convert_hf_to_torch_dist.py", line 119, in main
[rank3]:     bridge = AutoBridge.from_pretrained(hf_model_path, trust_remote_code=True)
[rank3]:              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.12/dist-packages/mbridge/core/auto_bridge.py", line 30, in from_pretrained
[rank3]:     return cls.from_config(config, **kwargs)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.12/dist-packages/mbridge/core/auto_bridge.py", line 48, in from_config
[rank3]:     return _MODEL_REGISTRY[model_type](hf_config, **kwargs)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.12/dist-packages/mbridge/core/bridge.py", line 51, in __init__
[rank3]:     self.config = self._build_config()
[rank3]:                   ^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/nfs-152/disk6/tujie/train_env_slime/slime/slime_plugins/mbridge/minimax_m2.py", line 42, in _build_config
[rank3]:     return self._build_base_config(
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.12/dist-packages/mbridge/core/llm_bridge.py", line 108, in _build_base_config
[rank3]:     return self.TransformerConfigClass(**base_config)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: TypeError: TransformerConfig.__init__() got an unexpected keyword argument 'rotary_percent'

i dock pull the latest version slimerl/slime:latest, and git clone the main branch of the repo, what's the possible cause of the problem, and how to solve? Appreciate the help!

Steps to Reproduce

1. docker pull

2. ran the conversion command

Expected Behavior

The conversion succeed

Actual Behavior

ran into error

Environment

• slime version:
• Python version:
• PyTorch version:
• CUDA/ROCm version:
• GPU type and count:
• OS:
• SGLang version (if relevant):
• Megatron-LM version (if relevant):

Logs

Additional Context

No response

Pre-submission Checklist

• I have read the CONTRIBUTING.md and understand the collaboration scope.
• I have read the documentation and my issue is not addressed there.
• I have searched for existing issues and this is not a duplicate.
• I have provided a minimal, reproducible example.

[Lynnzake]

I did't ran

# Path can be adjusted according to actual situation
cd /root/slime
git pull
pip install -e . --no-deps

instead, i git clone the repo, but did't pip install, maybe that's the possible cause?

[lilei199908]

I did't ran

Path can be adjusted according to actual situation

cd /root/slime
git pull
pip install -e . --no-deps
instead, i git clone the repo, but did't pip install, maybe that's the possible cause?

yes, you need git apply this patch for minimax2 serial:

diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/transformer/transformer_config.py
index 0eebfbcd5..951da3f39 100644
--- a/megatron/core/transformer/transformer_config.py
+++ b/megatron/core/transformer/transformer_config.py
@@ -164,6 +164,10 @@ class TransformerConfig(ModelParallelConfig):
     """True is rotate pairs of even and odd dimensions (RoFormer style), False is rotate pairs of
     first half and second half (LLaMa style). Default to False."""
 
+    rotary_percent: Optional[float] = None
+    """Optional percent of rotary dimension to use for rotary position embeddings. If unset,
+    model constructors use their rotary_percent argument."""
+
     window_size: Optional[Tuple[int, int]] = None
     """If not None, then will use sliding window attention. The size of the window is specified by
     the numbers inside the tuple; -1 is special value meaning "infinite window size"."""

we will update it in docker images later