5ff4b8a2eb
Port LEM scoring/training pipeline into CoreGo as pkg/ml with: - Inference abstraction with HTTP, llama-server, and Ollama backends - 3-tier scoring engine (heuristic, exact, LLM judge) - Capability and content probes for model evaluation - GGUF/safetensors format converters, MLX to PEFT adapter conversion - DuckDB integration for training data pipeline - InfluxDB metrics for lab dashboard - Training data export (JSONL + Parquet) - Expansion generation pipeline with distributed workers - 10 CLI commands under 'core ml' (score, probe, export, expand, status, gguf, convert, agent, worker) - 5 MCP tools (ml_generate, ml_score, ml_probe, ml_status, ml_backends) All 37 ML tests passing. Binary builds at 138MB with all commands. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
63 lines
2.2 KiB
Go
63 lines
2.2 KiB
Go
// Package ml provides ML inference, scoring, and training pipeline commands.
|
|
//
|
|
// Commands:
|
|
// - core ml score: Score responses with heuristic and LLM judges
|
|
// - core ml probe: Run capability and content probes against a model
|
|
// - core ml export: Export golden set to training JSONL/Parquet
|
|
// - core ml expand: Generate expansion responses
|
|
// - core ml status: Show training and generation progress
|
|
// - core ml gguf: Convert MLX LoRA adapter to GGUF format
|
|
// - core ml convert: Convert MLX LoRA adapter to PEFT format
|
|
// - core ml agent: Run the scoring agent daemon
|
|
// - core ml worker: Run a distributed worker node
|
|
package ml
|
|
|
|
import (
|
|
"forge.lthn.ai/core/cli/pkg/cli"
|
|
)
|
|
|
|
func init() {
|
|
cli.RegisterCommands(AddMLCommands)
|
|
}
|
|
|
|
var mlCmd = &cli.Command{
|
|
Use: "ml",
|
|
Short: "ML inference, scoring, and training pipeline",
|
|
Long: "Commands for ML model scoring, probe evaluation, data export, and format conversion.",
|
|
}
|
|
|
|
// AddMLCommands registers the 'ml' command and all subcommands.
|
|
func AddMLCommands(root *cli.Command) {
|
|
initFlags()
|
|
mlCmd.AddCommand(scoreCmd)
|
|
mlCmd.AddCommand(probeCmd)
|
|
mlCmd.AddCommand(exportCmd)
|
|
mlCmd.AddCommand(expandCmd)
|
|
mlCmd.AddCommand(statusCmd)
|
|
mlCmd.AddCommand(ggufCmd)
|
|
mlCmd.AddCommand(convertCmd)
|
|
mlCmd.AddCommand(agentCmd)
|
|
mlCmd.AddCommand(workerCmd)
|
|
root.AddCommand(mlCmd)
|
|
}
|
|
|
|
// Shared persistent flags.
|
|
var (
|
|
apiURL string
|
|
judgeURL string
|
|
judgeModel string
|
|
influxURL string
|
|
influxDB string
|
|
dbPath string
|
|
modelName string
|
|
)
|
|
|
|
func initFlags() {
|
|
mlCmd.PersistentFlags().StringVar(&apiURL, "api-url", "http://10.69.69.108:8090", "OpenAI-compatible API URL")
|
|
mlCmd.PersistentFlags().StringVar(&judgeURL, "judge-url", "http://10.69.69.108:11434", "Judge model API URL (Ollama)")
|
|
mlCmd.PersistentFlags().StringVar(&judgeModel, "judge-model", "gemma3:27b", "Judge model name")
|
|
mlCmd.PersistentFlags().StringVar(&influxURL, "influx", "", "InfluxDB URL (default http://10.69.69.165:8181)")
|
|
mlCmd.PersistentFlags().StringVar(&influxDB, "influx-db", "", "InfluxDB database (default training)")
|
|
mlCmd.PersistentFlags().StringVar(&dbPath, "db", "", "DuckDB database path (or set LEM_DB env)")
|
|
mlCmd.PersistentFlags().StringVar(&modelName, "model", "", "Model name for API")
|
|
}
|