Documentation
¶
Overview ¶
Package finetune exports completed task I/O pairs as JSONL files suitable for LLM fine-tuning. It supports OpenAI and Anthropic fine-tuning formats, optional quality filtering via pkg/eval, and prompt anonymization.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type ExportConfig ¶
type ExportConfig struct {
WorkDir string
OutputPath string
Format Format
// MinQuality is a 0-100 quality threshold. Pairs scored below this are skipped.
// Set to 0 to disable quality filtering.
MinQuality int
// Anonymize strips absolute file paths and the project directory name from prompts and outputs.
Anonymize bool
// Eval provider settings — required when MinQuality > 0.
EvalProvider provider.Provider
EvalModel string
EvalTimeout time.Duration
}
ExportConfig holds all parameters for a fine-tune export run.
type ExportResult ¶
type ExportResult struct {
Total int // candidate pairs found
Exported int // pairs written to the JSONL file
Skipped int // pairs skipped (low quality or missing output)
AvgQuality float64 // average quality score (1-10) of exported pairs; 0 if no scoring
Tokens int // estimated total tokens across all exported pairs
}
ExportResult summarises the outcome of an export run.
func Export ¶
func Export(ctx context.Context, plan *pm.Plan, goal, instructions string, cfg ExportConfig) (*ExportResult, error)
Export scans the completed tasks in the plan, reconstructs each I/O pair, and writes them as JSONL to cfg.OutputPath. It returns a summary.
Source Files
Click to show internal directories.
Click to hide internal directories.