Siyuc/INFUSER-Qwen3-8B-base-artifacts

INFUSER Qwen3-8B-base Artifacts

This dataset contains raw INFUSER Qwen3-8B-base generator output artifacts for steps 0, 30, 60, and 90, as presented in the paper INFUSER: Influence-Guided Self-Evolution Improves Reasoning.

Paper | Code | Blog

Artifact Information

• Source run path: qwen3_8b_base/FW-Alr_2e-6-Glr_4e-6-DrGRPO-TIS_token-dev_800-precond_cos-seed456
• Source model/run context: INFUSER Qwen3-8B-base generator outputs from training config FW-Alr_2e-6-Glr_4e-6-DrGRPO-TIS_token-dev_800-precond_cos-seed456.
• Related model repo: Siyuc/INFUSER-Qwen3-8B-base

Files

The repository contains the following .pt files at the root:

• gen_output_step000.pt
• gen_output_step030.pt
• gen_output_step060.pt
• gen_output_step090.pt

Usage

These .pt files are raw DataProto torch artifacts. They can be converted to question JSON using the project code's derive_gen_questions() function after loading the DataProto artifact. In the official codebase, see verl_inf_evolve/utils/data_utils.py for implementation details.

Citation

@misc{chen2026infuser,
  title        = {INFUSER: Influence-Guided Self-Evolution Improves Reasoning},
  author       = {Siyu Chen and Miao Lu and Beining Wu and Heejune Sheen and Fengzhuo Zhang and Shuangning Li and Zhiyuan Li and Jose Blanchet and Tianhao Wang and Zhuoran Yang},
  year         = {2026},
  eprint       = {2606.09052},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url          = {https://arxiv.org/abs/2606.09052}
}