AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

SemRep : Generative Code Representation Learning with Code Transformations

Weichen Li, Jiamin Song, Bogdan Alexandru Stoica, Arav Dhoot, Gabriel Ryan, Shengyu Fu, Kexin Pei

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural-symbolic reasoning · 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
add generative code representation learning (mid-training) (ablation)finetuned QwQ-32B · KernelBenchCorrectness avg@16: 12.56 -> 25.75Improved
add generative code representation learning (mid-training) (ablation)finetuned QwQ-32B · EditBench CorePass@16 (%): 55.56 -> 62.96Improved
add SemRep test-time scaling to finetuned baseline (ablation)finetuned QwQ-32B · EditBench CorePass@1 (%): 53.70 -> 56.48Improved
add SemRep test-time scaling to finetuned baseline (ablation)finetuned QwQ-32B · KernelBenchCorrectness best@16: 74 -> 59Hurt
add SemRep test-time scaling to finetuned baseline (ablation)finetuned QwQ-32B · KernelBenchSpeedup best@16: 1.33 -> 1.47Improved
combine SemRep training and TTS (ablation)finetuned QwQ-32B · KernelBenchCorrectness best@16: 74 -> 93Improved
combine SemRep training and TTS (ablation)finetuned QwQ-32B · KernelBenchSpeedup best@16: 1.33 -> 2.87Improved
combine SemRep training and TTS (ablation)finetuned QwQ-32B · EditBench CorePass@16 (%): 55.56 -> 66.67Improved
add TTS on top of SemRep-trained model (ablation)SemRep-trained QwQ-32B · KernelBenchSpeedup avg@16: 1.25 -> 1.21Hurt
use SemRep instead of finetuning onlyQwQ-32B · KernelBench on unseen H200Correctness avg@16: 12.63 -> 20.38Improved
use SemRep instead of finetuning onlyQwQ-32B · EditBench Core (perturbed)Consistency (%): 75.00 -> 88.89Improved
use SemRep instead of Kevin-32BQwQ-32B · KernelBench (100 test tasks)Correctness best@16: 65 -> 93Improved

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