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_API_KEY", vars::GOOGLE_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 \"${MODEL_NAME}\"",
35 )
36 }
37
38 named::job(
39 Job::default()
40 .runs_on(runners::LINUX_DEFAULT)
41 .timeout_minutes(60_u32 * 10)
42 .add_step(steps::checkout_repo())
43 .add_step(steps::cache_rust_dependencies_namespace())
44 .map(steps::install_linux_dependencies)
45 .add_step(setup_cargo_config(Platform::Linux))
46 .add_step(steps::script("cargo build --package=eval"))
47 .add_step(run_eval())
48 .add_step(steps::cleanup_cargo_config(Platform::Linux)),
49 )
50}
51
52pub(crate) fn run_unit_evals() -> Workflow {
53 let unit_evals = unit_evals();
54
55 named::workflow()
56 .on(Event::default()
57 .schedule([
58 // GitHub might drop jobs at busy times, so we choose a random time in the middle of the night.
59 Schedule::default().cron("47 1 * * 2"),
60 ])
61 .workflow_dispatch(WorkflowDispatch::default()))
62 .concurrency(vars::one_workflow_per_non_main_branch())
63 .add_env(("CARGO_TERM_COLOR", "always"))
64 .add_env(("CARGO_INCREMENTAL", 0))
65 .add_env(("RUST_BACKTRACE", 1))
66 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
67 .add_job(unit_evals.name, unit_evals.job)
68}
69
70fn unit_evals() -> NamedJob {
71 fn send_failure_to_slack() -> Step<Use> {
72 named::uses(
73 "slackapi",
74 "slack-github-action",
75 "b0fa283ad8fea605de13dc3f449259339835fc52",
76 )
77 .if_condition(Expression::new("${{ failure() }}"))
78 .add_with(("method", "chat.postMessage"))
79 .add_with(("token", vars::SLACK_APP_ZED_UNIT_EVALS_BOT_TOKEN))
80 .add_with(("payload", indoc::indoc!{r#"
81 channel: C04UDRNNJFQ
82 text: "Unit Evals Failed: https://github.com/zed-industries/zed/actions/runs/${{ github.run_id }}"
83 "#}))
84 }
85
86 named::job(
87 Job::default()
88 .runs_on(runners::LINUX_DEFAULT)
89 .add_step(steps::checkout_repo())
90 .add_step(steps::setup_cargo_config(Platform::Linux))
91 .add_step(steps::cache_rust_dependencies_namespace())
92 .map(steps::install_linux_dependencies)
93 .add_step(steps::cargo_install_nextest(Platform::Linux))
94 .add_step(steps::clear_target_dir_if_large(Platform::Linux))
95 .add_step(
96 steps::script("./script/run-unit-evals")
97 .add_env(("ANTHROPIC_API_KEY", vars::ANTHROPIC_API_KEY)),
98 )
99 .add_step(send_failure_to_slack())
100 .add_step(steps::cleanup_cargo_config(Platform::Linux)),
101 )
102}