read-eval-logs

List, dump, convert, and view Inspect evaluation logs via Python APIs and CLI.

617|397|Updated Oct 2, 2024
One-click install
npx skills add https://github.com/UKGovernmentBEIS/inspect_evals --skill read-eval-logs
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: read-eval-logs
Source: https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/.claude/skills/read-eval-logs
Command: npx skills add https://github.com/UKGovernmentBEIS/inspect_evals --skill read-eval-logs

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill enables teams and developers to view and analyze Inspect evaluation log files using the Python API and CLI, reducing time spent on manual log inspection and enabling faster data-driven decisions.

Core Features & Use Cases

  • List logs, filter by status, and view summaries to quickly gauge model performance across runs.
  • Dump logs or convert formats (JSON or Eval) for offline analysis, debugging, or sharing results.
  • Open an interactive viewer to explore logs and samples in real time, including per-sample details and summaries.

Quick Start

uv run inspect log list --json uv run inspect log dump <log_file_path> uv run inspect view uv run inspect log schema

Frequently Asked Questions about read-eval-logs

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I view and analyze Inspect evaluation logs using Python?

You can list Inspect evaluation logs using the command `uv run inspect log list --json`. This command outputs a JSON list of logs, which you can filter by status to quickly gauge model performance across multiple runs.

How do I convert .eval log files to JSON format for offline analysis?

You can convert .eval log files to JSON format using the `uv run inspect log dump <log_file_path>` command. This dumps the log content, enabling offline analysis, debugging, or sharing results in a standard JSON format.

Can I explore per-sample details from evaluation logs in real time?

Yes, you can explore per-sample details from evaluation logs in real time by opening the interactive viewer with the `uv run inspect view` command. This allows you to explore logs and samples dynamically.

Do I need the uv command interface to extract the log schema?

Yes, you need the uv command interface to extract the log schema. Running `uv run inspect log schema` relies on the uv command interface to provide programmatic access and robust log schema extraction capabilities.

What is the best way to filter evaluation logs by status across multiple runs?

The best way to filter evaluation logs by status across multiple runs is using the CLI list command with status filtering. This approach provides programmatic access to summarize and view logs efficiently.