Research map / Large language models / Instruction and fine-tuning
SPICE: Submodular Penalized Information–Conflict Selection for Efficient Large Language Model Training
Powei Chang, Jinpeng Zhang, Bowen Chen, Chenyu Wang, Chenlu Guo, Yixing Zhang, Yukang Gao, JianXiang Xiang, Yue Gao, Chaoqun Sun, Yiyi Chen, Dongying kong
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 |
|---|---|---|---|
| select 10% data with SPICE instead of training on full data | Qwen2-7B LoRA instruction tuning · 8-benchmark average | average score: 56.4 -> 58.0 | Improved |
| select 10% data with SPICE instead of training on full data | LLaMA2-7B LoRA instruction tuning · 8-benchmark average | average score: 30.8 -> 31.1 | Improved |
| use SPICE instead of LESS | Qwen2-7B · 8-benchmark average | average score: 55.2 -> 58.0 | Improved |
| use SPICE instead of DPP | Qwen2-7B · 8-benchmark average | average score: 57.0 -> 58.0 | Improved |
| use SPICE instead of random 10% selection | Qwen2-7B · IFEval | score: 28.3 -> 38.6 | Improved |
| use SPICE instead of full data | LLaMA2-7B · TruthfulQA | score: 43.5 -> 40.3 | Hurt |
| use SPICE instead of full data | Qwen2-7B · BBH | score: 61.3 -> 61.0 | No clear effect |
| add conflict penalty lambda greater than 0 to greedy log-det FIM selection (ablation) | SPICE · 8 benchmarks | performance | Improved |
| use larger 7B proxy model instead of 0.5B or 1.5B Qwen2 proxy (ablation) | SPICE · 8 benchmarks | average score: +0.1 average | No clear effect |
| use LLaMA2-style proxy instead of Qwen2 proxy to select data for Qwen2-7B (ablation) | SPICE · Qwen2-7B | accuracy | Hurt |
| use adaptive early stopping SPICE+ instead of fixed budget (ablation) | SPICE · benchmarks | accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SelectIT — SelectIT: Selective Instruction Tuning for LLMs via Uncertainty-Aware Self-Reflection (NeurIPS 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DELIFT: Data Efficient Language model Instruction Fine-Tuning (ICLR 2025 · similarity 0.89)
- Boosting Multi-Domain Fine-Tuning of Large Language Models through Evolving Interactions between Samples (ICML 2025 · similarity 0.88)
- One Algorithm, Two Goals: Dual Scoring for Parameter and Data Selection in LLM Fine-Tuning (NeurIPS 2026 · similarity 0.87)
- LESS: Selecting Influential Data for Targeted Instruction Tuning (ICML 2024 · similarity 0.87)
- MGUP: A Momentum-Gradient Alignment Update Policy for Stochastic Optimization (NeurIPS 2025 · similarity 0.87)
- APT: Adaptive Pruning and Tuning Pretrained Language Models for Efficient Training and Inference (ICML 2024 · similarity 0.87)
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.
