Paper Conclusion RL Training: Qwen3-VL-8B-Thinking with External Judge

Paper Conclusion RL Training is a dataset for reinforcement learning training based on the EasyR1 (verl) framework. The training model is Qwen3-VL-8B-Thinking, using an external judge model (Qwen3-4B-Instruct-2507) to score predicted conclusions against a 235B teacher model's reference conclusions. The dataset was authored by SII-ChengqiLi and last updated on 2026-04-10.

Use Cases

• Fine-tuning language models for academic conclusion generation based on the described RL training framework.
• Benchmarking model-generated text against a teacher model's output based on the described scoring methodology.
• Studying the effectiveness of reinforcement learning with human feedback (RLHF) variants for text quality improvement.
• Developing reward models for text generation tasks using the described judge model architecture.

Strengths

• Training framework (EasyR1/verl) and configuration files are provided, offering a reproducible setup.
• Uses a specific, large teacher model (235B) and a defined judge model (Qwen3-4B-Instruct-2507) for scoring.
• Includes prompt templates (paper_conclusion_json.jinja) for structured input formatting.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and sample data are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Likely generated for reinforcement learning training of a language model on academic text tasks.

Time Range

null

Freshness

Last updated 2026-04-10 10:28:24; freshness should be verified.

Geography

null