| 216 | } |
| 217 | |
| 218 | export function printUsage(): string { |
| 219 | return [ |
| 220 | "Usage: bun run scripts/longmemeval/runner.ts --dataset <path> [options]", |
| 221 | "", |
| 222 | "Options:", |
| 223 | " --dataset <path> Required dataset JSON path", |
| 224 | " --start <index> Inclusive start index", |
| 225 | " --end <index> Exclusive end index", |
| 226 | " --types <a,b> Filter by question types", |
| 227 | " --question-ids <a,b> Filter by explicit question IDs", |
| 228 | " --resume Resume from saved runner-state.json", |
| 229 | " --fast Use minimal delays", |
| 230 | " --parallel <n> Number of questions to process in parallel", |
| 231 | " --cleanup Delete OpenCode sessions after judging", |
| 232 | " --output-dir <path> Override output directory", |
| 233 | " --opencode-url <url> OpenCode API base URL", |
| 234 | " --turn-delay-ms <ms> Delay between replayed user turns", |
| 235 | " --session-delay-ms <ms> Delay between haystack sessions", |
| 236 | " --final-delay-ms <ms> Delay before asking final benchmark question", |
| 237 | " --max-attempts <n> Retry attempts for network/API failures", |
| 238 | " --retry-base-delay-ms <ms> Base delay for exponential backoff", |
| 239 | " --opencode-*-usd-per-million Optional estimated pricing inputs for OpenCode usage", |
| 240 | " --judge-model <model> Judge model name (default gpt-4o-2024-08-06)", |
| 241 | ].join("\n"); |
| 242 | } |