diff --git a/docs/DEVELOP.md b/docs/DEVELOP.md index f4e614823fc..5617ece18a0 100644 --- a/docs/DEVELOP.md +++ b/docs/DEVELOP.md @@ -16,6 +16,8 @@ This document explains how to configure the environment, run the LLM benchmark t Use this single command to quickly unblock CI by regenerating hashes and running only GPT-5 for the minimal Rust + C# passes. This is not the full benchmark suite. +**Note: You will need OpenAI API keys to run this locally**. Alternatively, any SpacetimeDB member can comment `/update-llm-benchmark` on a PR to start a CI job to do this. + `cargo llm ci-quickfix` What this does: 1. Runs Rust rustdoc_json pass for GPT-5 only. diff --git a/tools/xtask-llm-benchmark/src/bin/llm_benchmark.rs b/tools/xtask-llm-benchmark/src/bin/llm_benchmark.rs index 53e7eae176a..be7d49a8970 100644 --- a/tools/xtask-llm-benchmark/src/bin/llm_benchmark.rs +++ b/tools/xtask-llm-benchmark/src/bin/llm_benchmark.rs @@ -283,7 +283,7 @@ fn cmd_ci_check(args: CiCheckArgs) -> Result<()> { }; // Debug hint for how to (re)generate entries - let hint_for = |_lang: Lang| -> &'static str { "cargo llm ci-quickfix" }; + let hint_for = |_lang: Lang| -> &'static str { "Check DEVELOP.md for instructions on how to proceed." }; // Load docs-benchmark summary to compare hashes against let summary_path = docs_benchmark_summary(); @@ -298,13 +298,13 @@ fn cmd_ci_check(args: CiCheckArgs) -> Result<()> { match xtask_llm_benchmark::context::resolve_mode_paths(mode) { Ok(paths) if !paths.is_empty() => {} Ok(_) => bail!( - "CI check FAILED: {}/{} resolved to 0 paths.\n→ Try: {}", + "CI check FAILED: {}/{} resolved to 0 paths.\n→ {}", mode, lang_str, hint_for(lang) ), Err(e) => bail!( - "CI check FAILED: {}/{} not available: {}.\n→ Try: {}", + "CI check FAILED: {}/{} not available: {}.\n→ {}", mode, lang_str, e,