[Bug] qwen3.5 model init error.

Bug Description

based on env: slimerl/slime:nightly-dev-20260307a

Status message: Job entrypoint command failed with exit code 1, last available logs (truncated to 20,000 chars):
^^^^^^^^^^^^
File "/root/slime/slime/utils/arguments.py", line 1453, in parse_args
args = megatron_parse_args(
^^^^^^^^^^^^^^^^^^^^
File "/root/slime/slime/backends/megatron_utils/arguments.py", line 118, in megatron_parse_args
_hf_validate_args(args, hf_config)
File "/root/slime/slime/backends/megatron_utils/arguments.py", line 76, in _hf_validate_args
raise AssertionError("hf_validate_args failed: " + "; ".join(errors))
AssertionError: hf_validate_args failed: tie_word_embeddings in hf config True is not equal to untie_embeddings_and_output_weights True, please check the config.

problem highlight：tranformer read config uncorrectly

Steps to Reproduce

bash -x scripts/run-qwen3.5-35B-A3B.sh

Expected Behavior

work correctly

Actual Behavior

tie_word_embeddings in hf_config equal untie_embeddings_and_output_weights of mcore config

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.

Start with scripts/run-qwen3.5-35B-A3B.sh and trace argument parsing through slime/utils/arguments.py, slime/backends/megatron_utils/arguments.py, and _hf_validate_args. Reproduce the command and inspect how the Hugging Face and Megatron configuration values are read. Done means the Qwen3.5 initialization no longer raises this validation assertion.