NotebookMLX is an open-source implementation of NotebookLM, designed for users who want to transform PDF documents into engaging podcast-style audio. It leverages a pipeline of MLX-compatible models to process PDFs, generate transcripts, rewrite them dramatically, and finally convert the text to speech.

How It Works

The project orchestrates a four-step MLX model pipeline. First, a Qwen2.5-1.5B model preprocesses PDF content into text. Second, a Qwen2.5-14B model generates a podcast transcript from this text. Third, a Qwen2.5-7B model enhances the transcript with a more dramatic tone. Finally, the f5-tts-mlx model converts the rewritten transcript into conversational audio. This modular approach allows for distinct processing stages, each handled by specialized models.

Quick Start & Requirements

• Install: pip install mlx
• Prerequisites: MLX framework, Python 3.x. Requires specific MLX-compatible models from mlx-community and lucasnewman.
• Resources: MLX is designed for Apple Silicon (M1/M2/M3), implying hardware requirements for optimal performance. Model sizes suggest significant RAM and VRAM usage.
• Docs: NotebookMLX GitHub

Highlighted Details

• Leverages MLX for efficient on-device AI processing.
• Utilizes a chain of Qwen2.5 models for specialized text generation tasks.
• Integrates a Text-to-Speech model for audio output.
• Aims to replicate the functionality of NotebookLM.

Maintenance & Community

The project is maintained by johnmai-dev. Community engagement and roadmap details are not explicitly provided in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility is primarily with macOS on Apple Silicon due to the MLX framework.

Limitations & Caveats

The project appears to be a direct port and may be in early development. Specific performance benchmarks, error handling, and comprehensive documentation are not detailed. The reliance on specific MLX model versions could lead to compatibility issues with future MLX updates.