mirror of
https://github.com/vercel-labs/fx.git
synced 2026-10-28 10:15:17 +08:00
50 lines
982 B
YAML
50 lines
982 B
YAML
# Manual local release eval for fx.
|
|
# FxAgent uses the Harbor model name below as FX_MODEL inside the task container.
|
|
jobs_dir: evals/jobs-out
|
|
n_attempts: 1
|
|
timeout_multiplier: 1.0
|
|
n_concurrent_trials: 1
|
|
quiet: false
|
|
|
|
environment:
|
|
type: docker
|
|
force_build: true
|
|
delete: true
|
|
|
|
agents:
|
|
- name: fx
|
|
import_path: evals.agents.fx_agent:FxAgent
|
|
model_name: anthropic/claude-sonnet-4.6
|
|
|
|
datasets:
|
|
- path: evals/datasets/fx-release
|
|
|
|
metrics:
|
|
- type: uv-script
|
|
kwargs:
|
|
script_path: evals/metrics/fx_metrics.py
|
|
|
|
artifacts:
|
|
- source: /logs/agent
|
|
destination: agent
|
|
- source: /logs/verifier
|
|
destination: verifier
|
|
- source: /logs/artifacts
|
|
destination: fx-run
|
|
- source: /tmp/fx-home/.fx
|
|
destination: fx-home
|
|
- source: /app
|
|
destination: workspace
|
|
exclude:
|
|
- node_modules
|
|
- .git
|
|
- .cache
|
|
- __pycache__
|
|
- .pytest_cache
|
|
- dist
|
|
- build
|
|
- coverage
|
|
- target
|
|
- .venv
|
|
- venv
|