export checkpoint error: too many values to unpack (expected 2)

[wenkaixiansheng]

When I tried to convert the Megatron checkpoint such as Qwen3-8B, 32B, and 30B to Huggingface using megatron-bridge, I encountered the following error. I am using the latest commit of the main branch of Megatron-LM, and in this version, converting from Huggingface to Megatron is correct.

"
Traceback (most recent call last):
File "/usr/local/lib/python3.12/dist-packages/megatron/bridge/models/decorators/torchrun.py", line 37, in wrapper
return_value = recorded_fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/init.py", line 357, in wrapper
return f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/mnt/cephfs/xianda/slime-internal/tools/megatron-bridge-beta/convert_torch_dist_to_hf_multi_GPU.py", line 110, in main
for name, param in bridge.export_hf_weights(megatron_model, show_progress=False):
^^^^^^^^^^^
ValueError: too many values to unpack (expected 2)
"

[yaoyu-33]

mind sharing the exact command you are running? The case you mentioned in your description is covered in cicd. might be env issues.

[wenkaixiansheng]

the command：
”
PYTHONPATH=/mnt/cephfs/xianda/env/Megatron-LM python examples/conversion/convert_checkpoints.py export
--hf-model /mnt/cephfs/xianda/Qwen/Qwen3-8B
--megatron-path /mnt/cephfs/xianda/Qwen/Qwen3-8B-dist-mb
--hf-path /mnt/cephfs/xianda/Qwen/Qwen3-8B-hf
“

[wenkaixiansheng]

@yaoyu-33 If it's an environmental issue, I would be extremely grateful if you could tell me exactly where the problem might be, such as an incorrect megatron-lm version.

[yaoyu-33]

can you check using our container or build a container?
also better to use uv run python xxx
check https://github.com/NVIDIA-NeMo/Megatron-Bridge/blob/9a708dd81d4ed1ebba7f55b53620743093487911/CONTRIBUTING.md#development-environment