Support uncertainty-aware performance modeling in TraceLens

Open
  • Is something wrong?
#893 0 comments 0 reactions 1 assignee View on GitHub

Maintainers usually reply within 1 day

@ajassani is already working on this.

Since Aug 4, 2026.

Assessment

This issue has not been assessed yet.

Description

TraceLens performance models currently return point estimates for metrics such as FLOPs and bytes. Some workload properties, however, are inherently variable or unknown from the available trace metadata.

MoE routing is one example: the number of active experts—and therefore estimated weight traffic—depends on the routing distribution. A single expected value hides this uncertainty.

This issue tracks evolving the TraceLens performance-model abstraction to support:

  • Point estimates when inputs are known
  • Distributions or uncertainty ranges when inputs are variable
  • Explicit modeling assumptions
  • Measured distributions when runtime metadata is available
  • Backward-compatible scalar outputs for existing reports

The design should be general enough for future probabilistic models beyond MoE routing. Implementation details and report presentation require a separate design discussion.

Dominant language
Python
Stars
101
Forks
34
Avg merge
2d 1h
Merged PRs (30d)
78

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from AMD-AGI/TraceLens

All issues in AMD-AGI/TraceLens

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.