Task

Given a paper's context and a goal, the model produces a detailed, controlled ablation experiment design plan (objective, setup, variants, fixed protocols and metrics).

Training data

SFT on train/sft_task2_37019.jsonl, from SlowGuess/abforge-data (derived from CC-licensed research papers). Evaluation uses the held-out AblationBench split (eval/ablationbench_200.jsonl) of the same dataset.

Related models (Task 2)

• SlowGuess/ABForge-Qwen3-8B-Task2
• SlowGuess/ABForge-Qwen3-8B-Task2-SFT (this model)
• SlowGuess/ABForge-Qwen3-8B-Task2-RL

Evaluation

Reproduce AblationBench evaluation with the SlowGuess/Abforge_1 code:

bash
Copy code
git clone https://github.com/SlowGuess/Abforge_1 && cd Abforge_1
huggingface-cli download SlowGuess/abforge-data --repo-type dataset --local-dir data

export MODEL_PATH=SlowGuess/ABForge-Qwen3-8B-Task2-SFT

# 1. Generate predictions on AblationBench
python run_inference_local.py --task 2 \
  --input data/eval/ablationbench_200.jsonl \
  --output preds.jsonl \
  --model-path "$MODEL_PATH" --dtype bf16 --max-new-tokens 4096

# 2. Score against the fixed AblationBench rubric (Claude judge)
export ANTHROPIC_API_KEY=<your-key>
python scripts/eval_task2_claude_rubric_v2.py --input preds.jsonl --output scored.jsonl

Links

• Dataset: SlowGuess/abforge-data
• Code: SlowGuess/Abforge_1

Citation

bibtex
Copy code
@misc{abforge,
  title  = {ABForge: A Post-Training Pipeline for Paper-Grounded Ablation Design},
  author = {ABForge authors},
  year   = {2026},
  howpublished = {\url{https://github.com/SlowGuess/Abforge_1}}
}