traceopt-ai/traceml
View on GitHub ↗ ★ 189 Python
Find why PyTorch training is slow, compare runs, and catch performance regressions in CI.
6 mentions on Hacker News by 3 people: 6 stories, 0 comments · first 2025-10-01 · latest 2026-03-12
20082026
-
▲ 1
Ask HN: Should training bottleneck detection be a product or just a feature?story by traceopt-ai · 2026-03-12 · 0 comments
-
▲ 2
Ask HN: Why does single-node DDP sometimes get slower with more GPUs?story by traceopt-ai · 2026-02-17 · 0 comments
-
▲ 1
Show HN: Finding stragglers in multi-GPU PyTorch (DDP) trainingstory by traceopt-ai · 2026-02-09 · 1 comment
-
▲ 3
Show HN: Distributed Training Observability for PyTorch (TraceML)story by traceml-ai · 2026-01-27 · 0 comments
-
▲ 3
Show HN: "htop" for PyTorch training, see stalls, memory and step time livestory by traceopt · 2026-01-19 · 0 comments
-
▲ 1
Show HN: TraceML, a tool to trace live memory usage in PyTorch trainingstory by traceopt-ai · 2025-10-01 · 1 comment