DeepSpeed team to host virtual office hours on October 27
The office hours are a direct channel to influence DeepSpeed's Q4 priorities, which is rare for a framework of this scale. Practitioners benchmarking training frameworks should attend to question the parallelism results against their own workloads, since the claims of matching state-of-the-art frameworks were made only on selected workloads. The benchmark comparisons against rival frameworks set the baseline for distributed training choices going into 2027.
The DeepSpeed team announced virtual office hours on 27 October at 9 AM PT / 12 PM ET, open to everyone. The team presented its latest progress at the PyTorch Conference, including AutoEP (expert parallelism), AutoTP (tensor parallelism), AutoSP (sequence parallelism) and Ray-based pipeline parallelism. Benchmarks shown at the conference indicate DeepSpeed matches or outperforms other state-of-the-art frameworks on the evaluated workloads. The office hours will cover those features and benchmarks in more depth, plus DeepSpeed's proposed Q4 roadmap.
Join the @DeepSpeedAI team for virtual office hours on October 27!
At last week’s PyTorch Conference, the DeepSpeed team shared the project’s latest progress during the PyTorch Foundation keynote and a technical session. They highlighted advances including AutoEP (expert parallelism), AutoTP (tensor parallelism), AutoSP (sequence parallelism), and Ray-based pipeline parallelism, with benchmarks showing that DeepSpeed matches or outperforms other state-of-the-art frameworks on the workloads evaluated.
These virtual office hours will offer a deeper dive into the features and benchmark results presented at the conference, along with a discussion of DeepSpeed’s proposed Q4 roadmap.
Interested in learning more, asking the team questions, or contributing to these projects? Join the conversation and share your ideas for what DeepSpeed should prioritize next. Everyone is welcome!
📅 Tuesday, October 27 🕘 9 AM PT / 12 PM ET 🔗 Join: https://t.co/Vm6LFc9qLb
