Lightning Talk: Tensor Query Processing
summary
Tensor Query Processor (TQP) maps SQL physical plans into PyTorch tensor programs, implements custom tensor operators for joins and group-by, and applies operator fusion plus code generation to target accelerators. It encodes columns (numbers, timestamps, padded UTF‑8 strings) as tensors, runs the full TPC-H SF50 in 33s, and reports about 4x faster than SQL Server on the tested CPU-to-GPU configuration while remaining portable to other hardware supported by the runtime.
AI-written summary · about
chapters (7)
- 0:00Tensor query processing
- 1:49ML to SQL with Tensor Abstraction
- 3:35Tabular data as tensors
- 5:01SQL queries to PyTorch
- 7:09SQL on hardware accelerators
- 7:32TPC-H benchmarks
- 8:55Q&A
related talks
16mqTPU: Hybrid Tensor Networks for Quantum-Classical Acceleration
26mThunder: Distribute and Optimize Your PyTorch Models With Zero...
17mLightning Thunder: Supercharged PyTorch for Modern Hardware
23mThe Race to the Bottom: Low Latency in the age of the Transformer
14mLightning Talk: Every Millisecond Counts: The Fine-tuning Journey of an Ultra-Eff...
19mJosh Patterson, Voltron Data | Nvidia GTC 2024
28mSupercharge your training on TPUs
18mQ&A Session