~/dasomel
한국어

MLOps Pipelines

Local LLM fine-tuning (LoRA/QLoRA), MLX quantization, and low-latency inference.

MLOps Pipelines

Guide to authoring fine-tuning and inference pipelines in KubeMetal.

Supported Workflows

  1. One-Click Model Downloads: HuggingFace and Ollama weight ingestion
  2. LoRA / QLoRA Fine-Tuning: Supervised fine-tuning on local custom datasets
  3. 4-bit / 8-bit Quantization: 70% memory reduction via Apple MLX quantization
  4. High-Speed Local Serving: OpenAI-compatible /v1/chat/completions endpoints