NeuroBrix 0.6.2 benchmark.18 models, Triton engine vs PyTorch engine, same container, same request, one run alone on the whole machine, median of 3, Tesla V100 32 GB and 16 GB.The Triton engine (no PyTorch) is faster in 30 of 36 pairs: up to 5.24x on Janus-Pro-7B, 3.64x on DeepSeek-Coder-V2-Lite. Slower in 6 (Kokoro, Flex.1, Sana 4K), on the chart too.Every number is a cell of one dated register: https://github.com/NeuroBrix/neurobrix/blob/main/docs/BENCHMARKS.md#opensource #AI
NeuroBrix 0.6.2 benchmark.18 models, Triton engine vs PyTorch engine, same container, same request, one run alone on the...
