Research map / Large language models / Instruction and fine-tuning
Minibatch selection for Language Models via Partition Matroid Constrained Gradient Matching
Prayas Agrawal, Prateek Chanda, Ishita Khatri, Ganesh Ramakrishnan, Bamdev Mishra, Pratik Jawanpuria
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 jointly across domains under a partition matroid instead of independently per domain (ID) | PartitionSel · MetaMathQA, 8-benchmark average, Qwen2.5-3B | average accuracy: 0.6263 -> 0.6363 | Improved |
| use continuous prototype weights with independent domain-wise selection (IWD) instead of binary (ID) (ablation) | independent domain-wise selection · MetaMathQA, 8-benchmark average, Qwen2.5-3B | average accuracy: 0.6263 -> 0.6225 | No clear effect |
| PartitionSel vs GREATS single global budget selection | minibatch selection · Mol-Instructions average BLEU at 25% subset | BLEU: 0.62 -> 0.66 | Improved |
| PartitionSel vs random selection on Mol-Instructions at 12.5% subset | minibatch selection · Mol-Instructions average BLEU 12.5% | BLEU: 0.62 -> 0.63 | No clear effect |
| PartitionSel vs DoReMi proxy-based mixture | MetaMathQA fine-tuning · MetaMathQA downstream average | accuracy (%): 61.34 -> 65.99 | Improved |
| joint cross-domain selection instead of independent per-domain selection (ID) (ablation) | PartitionSel · gradient conflicts per batch | conflicting gradient pairs | Improved |
| compute gradients from 3 layers instead of 1 layer (ablation) | PartitionSel · Qwen2.5 validation log-pplx | validation log-pplx / latency: 2682 -> 2822.3 ms | Mixed |
| add randomized FFT (FJLT) projection to gradients (ablation) | PartitionSel · Qwen2.5-3B, LoRA rank 64 | feasibility (OOM): 3 and 5 layers: OOM -> feasible | Improved |
| apply PartitionSel instead of random selection across Qwen2.5 and Llama3 models | fine-tuning of Qwen2.5 (3B, 7B) and Llama3 (3B, 8B) · downstream accuracy | accuracy | Improved |
| PartitionSel vs Random peak GPU memory and time | PartitionSel · Qwen2.5-3B, 1K steps | wall-clock time: 44.17 -> 51:08 min | Hurt |
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.
- GREATS — GREATS: Online Selection of High-Quality Data for LLM Training in Every Iteration (NeurIPS 2024) · better on 5 of 7 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
- BLADE: Scalable Bi-level Adaptive Data Selection for LLM Training (NeurIPS 2026 · similarity 0.88)
- SmallToLarge (S2L): Scalable Data Selection for Fine-tuning Large Language Models by Summarizing Training Trajectories of Small Models (NeurIPS 2024 · similarity 0.87)
- Alignment of Large Language Models with Constrained Learning (NeurIPS 2025 · similarity 0.86)
- Boosting Multi-Domain Fine-Tuning of Large Language Models through Evolving Interactions between Samples (ICML 2025 · similarity 0.86)
- Data Selection via Optimal Control for Language Models (ICLR 2025 · similarity 0.86)
- Enhancing Large Language Model Performance with Gradient-Based Parameter Selection (AAAI 2025 · similarity 0.86)
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.
