Gemma Tuner Multimodal vs Vera

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.

Vera is a programming language built from the ground up for LLMs to write — not humans. Named after the Latin word for truth, it compiles to WebAssembly and runs in both the CLI and browser. Its most radical design choice: it eliminates variable names entirely, replacing them with typed De Bruijn structural references (like `@Int.0` for the most recent integer binding). Research suggests naming confusion is one of the biggest failure modes in AI-generated code — Vera removes the problem at the language level. Every function in Vera must declare `requires()` preconditions, `ensures()` postconditions, and `effects()` side-effect declarations. The compiler uses Z3 formal verification to check contracts at every call site, meaning the AI can't ship code that violates its own preconditions. Error messages are structured JSON with stable codes — written as instructions for AI systems to parse and fix, not human developers to read. Benchmark results are striking: on VeraBench, Kimi K2.5 achieves 100% correctness writing Vera code, outperforming both Python (86%) and TypeScript (91%) implementations. At v0.0.127 with 810+ commits, 127 releases, 3,638 tests, and a 13-chapter spec, this is a serious project — not a weekend experiment. If AI is going to write most of our code, perhaps the code should be designed for AI to write.