1use gh_workflow::{Event, Expression, Job, Run, Schedule, Step, Use, Workflow, WorkflowDispatch};
2
3use crate::tasks::workflows::{
4 runners::{self, Platform},
5 steps::{self, FluentBuilder as _, NamedJob, named, setup_cargo_config},
6 vars::{self, Input},
7};
8
9pub(crate) fn run_agent_evals() -> Workflow {
10 let agent_evals = agent_evals();
11 let model_name = Input::string("model_name", None);
12
13 named::workflow()
14 .on(Event::default().workflow_dispatch(
15 WorkflowDispatch::default().add_input(model_name.name, model_name.input()),
16 ))
17 .concurrency(vars::one_workflow_per_non_main_branch())
18 .add_env(("CARGO_TERM_COLOR", "always"))
19 .add_env(("CARGO_INCREMENTAL", 0))
20 .add_env(("RUST_BACKTRACE", 1))
21 .add_env(("ANTHROPIC_API_KEY", vars::ANTHROPIC_API_KEY))
22 .add_env(("OPENAI_API_KEY", vars::OPENAI_API_KEY))
23 .add_env(("GOOGLE_AI_API_KEY", vars::GOOGLE_AI_API_KEY))
24 .add_env(("GOOGLE_CLOUD_PROJECT", vars::GOOGLE_CLOUD_PROJECT))
25 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
26 .add_env(("ZED_EVAL_TELEMETRY", 1))
27 .add_env(("MODEL_NAME", model_name.to_string()))
28 .add_job(agent_evals.name, agent_evals.job)
29}
30
31fn agent_evals() -> NamedJob {
32 fn run_eval() -> Step<Run> {
33 named::bash(
34 "cargo run --package=eval -- --repetitions=8 --concurrency=1 --model \"${MODEL_NAME}\"",
35 )
36 }
37
38 named::job(
39 Job::default()
40 .runs_on(runners::LINUX_DEFAULT)
41 .timeout_minutes(60_u32 * 10)
42 .add_step(steps::checkout_repo())
43 .add_step(steps::cache_rust_dependencies_namespace())
44 .map(steps::install_linux_dependencies)
45 .add_step(setup_cargo_config(Platform::Linux))
46 .add_step(steps::script("cargo build --package=eval"))
47 .add_step(run_eval())
48 .add_step(steps::cleanup_cargo_config(Platform::Linux)),
49 )
50}
51
52pub(crate) fn run_unit_evals() -> Workflow {
53 let unit_evals = unit_evals();
54
55 named::workflow()
56 .name("run_unit_evals")
57 .on(Event::default()
58 .schedule([
59 // GitHub might drop jobs at busy times, so we choose a random time in the middle of the night.
60 Schedule::default().cron("47 1 * * 2"),
61 ])
62 .workflow_dispatch(WorkflowDispatch::default()))
63 .concurrency(vars::one_workflow_per_non_main_branch())
64 .add_env(("CARGO_TERM_COLOR", "always"))
65 .add_env(("CARGO_INCREMENTAL", 0))
66 .add_env(("RUST_BACKTRACE", 1))
67 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
68 .add_job(unit_evals.name, unit_evals.job)
69}
70
71fn unit_evals() -> NamedJob {
72 fn send_failure_to_slack() -> Step<Use> {
73 named::uses(
74 "slackapi",
75 "slack-github-action",
76 "b0fa283ad8fea605de13dc3f449259339835fc52",
77 )
78 .if_condition(Expression::new("${{ failure() }}"))
79 .add_with(("method", "chat.postMessage"))
80 .add_with(("token", vars::SLACK_APP_ZED_UNIT_EVALS_BOT_TOKEN))
81 .add_with(("payload", indoc::indoc!{r#"
82 channel: C04UDRNNJFQ
83 text: "Unit Evals Failed: https://github.com/zed-industries/zed/actions/runs/${{ github.run_id }}"
84 "#}))
85 }
86
87 named::job(
88 Job::default()
89 .runs_on(runners::LINUX_DEFAULT)
90 .add_step(steps::checkout_repo())
91 .add_step(steps::setup_cargo_config(Platform::Linux))
92 .add_step(steps::cache_rust_dependencies_namespace())
93 .map(steps::install_linux_dependencies)
94 .add_step(steps::cargo_install_nextest(Platform::Linux))
95 .add_step(steps::clear_target_dir_if_large(Platform::Linux))
96 .add_step(
97 steps::script("./script/run-unit-evals")
98 .add_env(("ANTHROPIC_API_KEY", vars::ANTHROPIC_API_KEY)),
99 )
100 .add_step(send_failure_to_slack())
101 .add_step(steps::cleanup_cargo_config(Platform::Linux)),
102 )
103}