Osaurus — All your AI. One app.
Chat with GPT-5, Claude, Llama, and more — or download local MLX models from Hugging Face. Supercharge Cursor with powerful tools. Open source and bui...
Summary
mlx-retrieval is a toolkit for training embedding and reranker models for retrieval tasks on Apple Silicon using MLX, supporting LoRA training, gradient accumulation, and integration with evaluation and logging tools.
It features efficient data loading from JSONL or Elasticsearch, and runs the gemma-3-270m model at 4000-5000 tokens per second on M3 Ultra hardware. The project is primarily educational, focuses solely on single-task LoRA configurations, and uses mean pooling for final embedding generation.