AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Mastering Sparse CUDA Generation through Pretrained Models and Deep Reinforcement Learning

Yaoyu Wang, Hankun Dai, Zhidong Yang, Junmin Xiao, Guangming Tan

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Joint pruning and quantization

Read the paper

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 CUDA-specific pretraining (ablation)SparseRL · SuiteSparse SpMVpass@1000: 40.75 -> 49.25Improved
add CUDA-specific pretraining (ablation)SparseRL · SuiteSparse SpMVaverage execution speed (GFLOPS): 95.22 -> 116.20Improved
add PPO stage after SFT (ablation)SparseRL · SuiteSparse SpMVpass@1000: 32.75 -> 49.25Improved
add SFT stage before PPO (ablation)SparseRL · SuiteSparse SpMVpass@1000: 15.25 -> 49.25Improved
use sinusoidal embedding instead of raw indices (ablation)SparseRL+CodeT5 · SuiteSparse SpMVpass@1000: 40.75 -> 48.75Improved
use sinusoidal embedding instead of Max-Min normalization (ablation)SparseRL+CodeT5 · SuiteSparse SpMVpass@1000: 43.25 -> 48.75Improved
use sinusoidal embedding instead of UniXcoder-style embedding (ablation)SparseRL+CodeT5 · SuiteSparse SpMVpass@1000: 43.00 -> 48.75Improved
add execution efficiency reward to correctness-only RL (ablation)SparseRL · SuiteSparse SpMVGFLOPSImproved
raise efficiency scale r_eff from 0.5 to 1.0 (ablation)SparseRL · SuiteSparse SpMVaverage GFLOPS: 102.3 -> 116.2Improved
raise efficiency scale r_eff from 1.0 to 1.5 (ablation)SparseRL · SuiteSparse SpMVpass@1000 and GFLOPS: 49.25 -> 48.50 pass@1000; 116.2 -> 118.7 GFLOPSMixed
raise memory penalty r_penalty from 0.3 to 0.5 (ablation)SparseRL · SuiteSparse SpMVpass@1000: 49.25 -> 46.80Hurt
weight correctness more (alpha 0.5 -> 0.9) (ablation)SparseRL · SuiteSparse SpMVpass@1000 and GFLOPS: 49.25 -> 51.2 pass@1000; 116.2 -> 98.7 GFLOPSMixed

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