Gemma Tuner Multimodal vs Instant

Gemma Tuner Multimodal is an open-source fine-tuning toolkit for Google's Gemma 4 and Gemma 3n models that runs entirely on Apple Silicon using PyTorch with Metal Performance Shaders (MPS) backend — no NVIDIA GPU or cloud infrastructure required. It supports LoRA training on multimodal inputs: audio, images, and text simultaneously, using local CSV files or streamed from Google Cloud Storage or BigQuery. The tool targets the growing segment of developers who own M-series Macs but have been locked out of fine-tuning workflows that assume CUDA availability. Gemma 4's architecture is particularly well-suited to this use case: its 4B multimodal variant (designed for on-device deployment) trains efficiently on M3 Max and M4 Pro hardware within the available unified memory constraints. Primary use cases include medical transcription fine-tuning (audio → text with clinical terminology), visual QA systems (image + text → structured response), and private on-device pipelines where cloud API calls are prohibited by compliance requirements. The project fills a specific niche that Google's own fine-tuning documentation doesn't cover well for Apple hardware.

Instant 1.0 is a backend-as-a-service specifically designed for the era of AI-coded applications. Instead of building REST APIs, developers (and the AI agents coding for them) get a real-time database directly in the frontend — with built-in auth, permissions, storage, and payments bundled in. The API surface is deliberately minimal enough for LLMs to understand without large context windows. The key differentiation is agent-friendliness: Instant is fully operable via CLI, supports undo for destructive actions (critical when LLM-generated code makes mistakes), and includes a Google Zanzibar-inspired permissions system out of the box. YC-backed and already in production at multiple startups including Eden, HeroUI, and Prism, it has validation beyond prototype use cases. With AI agents increasingly writing the first draft of every app, backends that LLMs can reliably reason about become a competitive moat. Instant's bet is that the next generation of infrastructure needs to be designed for machines to operate, not just humans to configure. The HN thread had strong positive response with nuanced debate on Firebase comparisons.