2 commits

Author	SHA1	Message	Date
Claude	fcd1758b7d	feat(ml): add format converters, data pipeline, and scoring agent ... Port remaining lem-repo components into pkg/ml/: - convert.go: safetensors reader/writer, MLX→PEFT converter - gguf.go: GGUF v3 writer, MLX→GGUF LoRA converter - export.go: training data JSONL export with split/filter - parquet.go: Parquet export with snappy compression - db.go: DuckDB wrapper for golden set and expansion prompts - influx.go: InfluxDB v3 client for metrics/status - ollama.go: Ollama model management (create/delete with adapters) - status.go: training and generation status display - expand.go: expansion generation pipeline (Backend interface) - agent.go: scoring agent with probe running and InfluxDB push - worker.go: distributed worker for LEM API task processing Adds parquet-go and go-duckdb dependencies. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-15 23:46:24 +00:00
Claude	6f52e4e3ae	feat(ml): add ML inference and scoring engine from lem-repo ... Port LEM scoring pipeline into CoreGo pkg/ml/: - Backend interface abstracting HTTP, llama-server, and future backends - HTTPBackend for OpenAI-compatible APIs with retry logic - LlamaBackend managing llama-server via pkg/process - Scoring engine with heuristic, semantic, content, and exact suites - Judge for LLM-based multi-dimensional scoring - 23 capability probes (math, logic, reasoning, code) - 6 sovereignty content probes - GGUF/PEFT format helpers, safetensors reader - 37 tests passing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-15 23:34:54 +00:00