run_agent_evals.yml

 1# Generated from xtask::workflows::run_agent_evals
 2# Rebuild with `cargo xtask workflows`.
 3name: run_agent_evals
 4env:
 5  CARGO_TERM_COLOR: always
 6  CARGO_INCREMENTAL: '0'
 7  RUST_BACKTRACE: '1'
 8  ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
 9  OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
10  GOOGLE_AI_API_KEY: ${{ secrets.GOOGLE_AI_API_KEY }}
11  GOOGLE_CLOUD_PROJECT: ${{ secrets.GOOGLE_CLOUD_PROJECT }}
12  ZED_CLIENT_CHECKSUM_SEED: ${{ secrets.ZED_CLIENT_CHECKSUM_SEED }}
13  ZED_EVAL_TELEMETRY: '1'
14  MODEL_NAME: ${{ inputs.model_name }}
15on:
16  workflow_dispatch:
17    inputs:
18      model_name:
19        description: model_name
20        required: true
21        type: string
22jobs:
23  agent_evals:
24    runs-on: namespace-profile-16x32-ubuntu-2204
25    steps:
26    - name: steps::checkout_repo
27      uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
28      with:
29        clean: false
30    - name: steps::cache_rust_dependencies_namespace
31      uses: namespacelabs/nscloud-cache-action@v1
32      with:
33        cache: rust
34        path: ~/.rustup
35    - name: steps::setup_linux
36      run: ./script/linux
37    - name: steps::install_mold
38      run: ./script/install-mold
39    - name: steps::download_wasi_sdk
40      run: ./script/download-wasi-sdk
41    - name: steps::setup_cargo_config
42      run: |
43        mkdir -p ./../.cargo
44        cp ./.cargo/ci-config.toml ./../.cargo/config.toml
45    - name: cargo build --package=eval
46      run: cargo build --package=eval
47    - name: run_agent_evals::agent_evals::run_eval
48      run: cargo run --package=eval -- --repetitions=8 --concurrency=1 --model "${MODEL_NAME}"
49      env:
50        ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
51        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
52        GOOGLE_AI_API_KEY: ${{ secrets.GOOGLE_AI_API_KEY }}
53        GOOGLE_CLOUD_PROJECT: ${{ secrets.GOOGLE_CLOUD_PROJECT }}
54    - name: steps::cleanup_cargo_config
55      if: always()
56      run: |
57        rm -rf ./../.cargo
58    timeout-minutes: 600
59concurrency:
60  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
61  cancel-in-progress: true
62defaults:
63  run:
64    shell: bash -euxo pipefail {0}