SlowGuess/ABForge-Qwen3-8B-Task1

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 11, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

SlowGuess/ABForge-Qwen3-8B-Task1 is an 8 billion parameter Qwen3-based model developed by SlowGuess, specifically designed for ablation objective generation. This model is post-trained using the ABForge pipeline, which includes supervised fine-tuning and rubric-guided GRPO, to propose candidate ablation objectives for research papers. It excels at identifying a Target Module and a corresponding Research Question given an ablation-free context. The model has a context length of 32768 tokens, making it suitable for processing extensive research paper content.

Loading preview...

ABForge-Qwen3-8B-Task1 Overview

This model, ABForge-Qwen3-8B-Task1, is an 8 billion parameter language model based on the Qwen3 architecture, developed by SlowGuess. It is specifically engineered for Task 1: Ablation Objective Generation within the ABForge framework, a post-training pipeline for paper-grounded ablation design.

Key Capabilities

  • Ablation Objective Generation: Given the ablation-free context of a research paper, the model proposes candidate ablation objectives.
  • Structured Output: Each proposed objective is expressed as a Target Module (the component to ablate) paired with a Research Question it aims to answer.
  • Advanced Post-Training: The model undergoes a full ABForge pipeline, starting with supervised fine-tuning (SFT) from Qwen/Qwen3-8B, followed by rubric-guided GRPO (SFT → GRPO).
  • Extensive Context: Supports a context length of 32768 tokens, enabling it to process detailed research paper content.

Training and Evaluation

The model was fine-tuned on sft_task1_45961.jsonl and then further trained with GRPO on RL_task1_30K.jsonl from the SlowGuess/abforge-data dataset. Evaluation is performed using the held-out AblationBench split (ablationbench_200.jsonl) of the same dataset. Users can reproduce the AblationBench evaluation using the provided SlowGuess/Abforge_1 code.

Good For

  • Researchers and developers needing automated assistance in identifying potential ablation studies for their papers.
  • Generating structured ablation objectives (Target Module + Research Question) from research paper text.
  • Applications requiring deep understanding of research paper components for experimental design.

Finetunes
3 models