AI Paper Atlas

Research map / Large language models / Agents, code and math

Regression Language Models for Code

Yash Akhauri, Xingyou Song, Arissa Wongpanich, Bryan Lewandowski, Mohamed S. Abdelfattah

ICML 2026 · Large language models › Agents, code and math · approach: Execution-based code verification · 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
use decoder numeric head instead of normalized MSE regression head (ablation)RLM · NASBench-101Spearman rho: 0.717 -> 0.800Improved
normalize y-values for MSE regression head (ablation)Regression Head · NAS spacesSpearman rho: 0.478 -> 0.717Improved
use larger T5Gemma encoder (300M to 600M) (ablation)RLM · CodeNetSpearman rho: 0.744 -> 0.782Improved
use learned 8K ONNX tokenizer instead of T5 32K (ablation)RLM · NASBench-101Spearman rho: 0.533 -> 0.723Improved
increase encoder context from 1K to 4K (ablation)RLM · NASBench-101Spearman rho: 0.819 -> 0.838Improved
use explicit digit tokenization instead of T5Gemma tokenizer (ablation)RLM · CodeNet (random decoder init)Spearman rho: 0.654 -> 0.747Improved
initialize from language-pretrained T5Gemma encoder (ablation)RLM · Triton KernelSpearman rho: 0.504 -> 0.532Improved
pretrain on synthetic FLOPS before accuracy training (ablation)RLM · NASBench-101Spearman rho: 0.83 -> 0.85Improved
add NAS graph data to code training (ablation)RLM · coding benchmarks avgSpearman rho: 0.701 -> 0.706No clear effect
fine-tune on 1024 language-specific samplesRLM · CodeNet C++Spearman rho: 0.730 -> 0.595Hurt
fine-tune on 1K target NAS samplesRLM · NASBench-201Spearman rho: 0.681 -> 0.738Improved
fine-tune on 1K target NAS samplesRLM · AmoebaSpearman rho: 0.045 -> 0.501Improved

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