Training files

[h0jicha]

Thank you for your great work!
I couldn’t find the following files used to train the model. Could you provide them with the experimental settings in the paper?

The missing files are:

• grpo_vllm_sam_stage2_colocate.py (used in train_qwen2p5_3b_stage2.sh)
• train_qwen2p5_3b_reasonseg.sh (mentioned at the training section in README.md)

[catchcodes]

• grpo_vllm_sam_stage2_colocate.py (used in train_qwen2p5_3b_stage2.sh)
  This was a version-related typo, and it has now been fixed.

• train_qwen2p5_3b_reasonseg.sh (mentioned in the training section of README.md)
  Previously, we merged train_qwen2p5_3b_reasonseg.sh and train_qwen2p5_3b_cot.sh into a single script, but the README had not been updated accordingly.
  We have now separated the two scripts again.