Research map / Large language models / Instruction and fine-tuning
Theoretical Perspectives on Data Quality and Synergistic Effects in Pre- and Post-Training Reasoning Models
Adel Javanmard, Baharan Mirzasoleiman, Vahab Mirrokni
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| post-train SFT on examples aligned with the adaptation shift (smaller theta, harder for pretrained model) (ablation) | LSA transformer SFT · in-context weight prediction (d=400, m=200) | post-test error | Improved |
| increase number of SFT prompts B or prompt length n beyond an intermediate value, with interference (r>0) (ablation) | LSA transformer SFT · in-context weight prediction (d=400, m=200) | post-test error | Hurt |
| increase SFT prompts B with no interference (r=0) (ablation) | LSA transformer SFT · in-context weight prediction | post-test error | Mixed |
| increase SFT sample size B or context length n (ablation) | GPT-2 (9.5M) SFT · in-context weight prediction | test loss | Mixed |
| increase post-training CoT length k (ablation) | GPT-2 (9.5M) SFT · in-context weight prediction | test loss | No clear effect |
| increase outcome-supervision sample size B and context length n (ablation) | GPT-2 (9.5M) OS · in-context weight prediction | test loss | Improved |
| increase post-training CoT length k (ablation) | GPT-2 (9.5M) OS · in-context weight prediction | test loss | Hurt |
| weaken interference by raising r from 0.01 to 0.1 (ablation) | LSA transformer SFT · in-context weight prediction | post-test error | Improved |
Closest papers by meaning
- Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead (ICLR 2026 · similarity 0.92)
- On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models (ICML 2026 · similarity 0.90)
- UFT: Unifying Supervised and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- Behavior Injection: Preparing Language Models for Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs (ICLR 2026 · similarity 0.89)
- An Emulator for Fine-tuning Large Language Models using Small Language Models (ICLR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
