microsoft / microsoft/foundry-dev-tools

Create an eval from traces

Open
#291 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature request
Dominant language
JavaScript
Stars
2.1k
Forks
260
Avg merge
42m
Merged PRs (30d)
29

Description

Add new features that make simpler to create evals by using traces as the dataset. Example:

Image

Why this matters

  • Evals are important to get the most out of AI systems but are incredibly time consuming and frustrating to create
  • It can be confusing on what shape the data needs to be in for different eval types

Key scenarios this enables

  1. Detecting prompt regressions in agents
  2. Bulk model and prompt experimentation

MVP

  • The Input and Output can be automatically mapped to eval variables {{query}} and {{response}}. This is an assumption to streamline the experience. The same may be true for mapping to tool_definitions and tool_calls.
  • Start with options that easily map to trace data like Relevance, Task Adherence, Coherence, Similarity, Intent Resolution, and Tool Call Accuracy.

Contributor guide

No contributing guide indexed for this repository

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.

Research direction

No files, tests, or entry points are named. Start by tracing how datasets, traces, and eval types are currently represented, then compare the MVP scenarios and mappings against those flows; done means creating an eval from trace data with the listed variable mappings and supported eval options.

Written by the indexing model from the issue text.

Assessment

Domain
ai, testing-qa
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
28/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.