Replication Package of ASE26 Paper
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
# Replication Package This is the replication package of the paper "Bridging the Gap Between Intent and Impact: An Empirical Study of GPU Performance Optimizations in Deep Learning Frameworks" ## Requests - PyTroch and TensorFlow source code - CUDA 11.6, 11.8, and 12.0 - Python 3.10 - Docker(better to use docker for cleaner environment to compile PyTorch and TensorFlow) - Nsight Compute ## Preliminary * data/data.csv: preliminary study result includes PR num, optimization strategies, and reference URL. * data/coding.md: Coding book use in preliminary study ## Build * scripts/build: All scripts for building different versions of PyTorch and TensorFlow, including the Dockerfile and compile.sh ## Evaluation * scripts/evaluate/framework/: All scripts for both operator-level benchmark, e2e-level benchmark, and ncu profiling. * scripts/seed_inputs: the input for operator-level benchmark(https://figshare.com/s/4c75640d26a2182b9302) ## Analysis * scripts/data_analysis: The scripts to analyze static data from raw data(https://figshare.com/s/4c75640d26a2182b9302).- The build whl(TensorFlow) or .so(PyTorch) of different versions framework are not provide because of the data size limitation(among 3TB)
Fields of Study & Subjects
Parallel Computing and Optimization Techniques
Advanced Neural Network Applications
Software System Performance and Reliability
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 27 March 2026
- Language
- EN