Gemma Tuner Multimodal vs MemPalace

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.

MemPalace is a free, MIT-licensed AI memory framework that stores LLM conversation data verbatim locally — no AI summarization step, no per-query API costs. It integrates with Claude Code, ChatGPT, and Cursor via MCP, and claims the highest LongMemEval benchmark score among free memory frameworks at 96.6% (initially claimed 100% before community pressure forced a correction after GitHub issue #29 exposed test-set tuning). The project went viral on GitHub with 23,000+ stars in under 48 hours, partly because it was built by actress Milla Jovovich and developer Ben Sigman — an unusual origin story that dominated early coverage. But the technical pitch is real: competing paid solutions (Mem0 at $19–249/month, Zep at $25+/month) do similar things and charge for the privilege. MemPalace runs fully local, connects to any POSIX filesystem, and the verbatim storage approach avoids hallucination artifacts introduced by AI-summarized memory. The catch: verbatim storage means much higher storage overhead than summarization-based approaches, retrieval latency grows with context size, and the benchmark controversy raised questions about the team's methodology. For personal projects and small teams, the zero-cost angle is hard to argue with. For production systems where memory quality is critical, wait for independent benchmarking.