AI Paper Atlas

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

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Gradient-free adaptation and merging · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
select jointly across domains under a partition matroid instead of independently per domain (ID)PartitionSel · MetaMathQA, 8-benchmark average, Qwen2.5-3Baverage accuracy: 0.6263 -> 0.6363Improved
use continuous prototype weights with independent domain-wise selection (IWD) instead of binary (ID) (ablation)independent domain-wise selection · MetaMathQA, 8-benchmark average, Qwen2.5-3Baverage accuracy: 0.6263 -> 0.6225No clear effect
PartitionSel vs GREATS single global budget selectionminibatch selection · Mol-Instructions average BLEU at 25% subsetBLEU: 0.62 -> 0.66Improved
PartitionSel vs random selection on Mol-Instructions at 12.5% subsetminibatch selection · Mol-Instructions average BLEU 12.5%BLEU: 0.62 -> 0.63No clear effect
PartitionSel vs DoReMi proxy-based mixtureMetaMathQA fine-tuning · MetaMathQA downstream averageaccuracy (%): 61.34 -> 65.99Improved
joint cross-domain selection instead of independent per-domain selection (ID) (ablation)PartitionSel · gradient conflicts per batchconflicting gradient pairsImproved
compute gradients from 3 layers instead of 1 layer (ablation)PartitionSel · Qwen2.5 validation log-pplxvalidation log-pplx / latency: 2682 -> 2822.3 msMixed
add randomized FFT (FJLT) projection to gradients (ablation)PartitionSel · Qwen2.5-3B, LoRA rank 64feasibility (OOM): 3 and 5 layers: OOM -> feasibleImproved
apply PartitionSel instead of random selection across Qwen2.5 and Llama3 modelsfine-tuning of Qwen2.5 (3B, 7B) and Llama3 (3B, 8B) · downstream accuracyaccuracyImproved
PartitionSel vs Random peak GPU memory and timePartitionSel · Qwen2.5-3B, 1K stepswall-clock time: 44.17 -> 51:08 minHurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy