#llm/training + #open-source

Because vision, audio, and text inputs share the exact same weights, you no longer need to co-tune separate frozen encoders. Downstream adapter fine-tuning (such as LoRA) or full fine-tuning naturally updates the entire multimodal token loop in a single pass.

While many AI open source model providers are pursuing larger and more powerful models, Google is still giving attention to the smaller, more local side of the market. Today, the tech giant released Gemma 4 12B, an 11.95-billion-parameter open-weights model with permissive Apache 2.0 license optimized to execute locally on a standard enterprise laptop using just 16GB of VRAM or unified memory.

Traditional multimodal systems typically utilize discrete, separate encoders to translate audio waveforms and visual data into representations that the core language model can process.

This conventional approach inherently increases both inference latency and total memory consumption.

Gemma 4 12B radically alters this pipeline by functioning entirely without these secondary encoders. Instead, visual patches and raw audio waveforms are projected directly into the core large language model's embedding space through lightweight linear layers.

The vision encoder is replaced by a 35-million-parameter module utilizing a single matrix multiplication, while the audio encoder is eliminated entirely.

For enterprise engineering teams, this unified architecture delivers distinct operational advantages: lower latency for multimodal tasks, reduced VRAM requirements (down to 16GB — typical for laptops), and the ability to fine-tune the entire multimodal system in a single, cohesive pass.

Google has ensured that Gemma 4 12B is not an isolated experiment; it is ready for production. Weights are available on Hugging Face and Kaggle, and the model integrates seamlessly with industry-standard deployment frameworks such as vLLM, SGLang, MLX, and llama.cpp.

Claude’s constitution is the foundational document that both expresses and shapes who Claude is. It contains detailed explanations of the values we would like Claude to embody and the reasons why. In it, we explain what we think it means for Claude to be helpful while remaining broadly safe, ethical, and compliant with our guidelines. The constitution gives Claude information about its situation and offers advice for how to deal with difficult situations and tradeoffs, like balancing honesty with compassion and the protection of sensitive information. Although it might sound surprising, the constitution is written primarily for Claude. It is intended to give Claude the knowledge and understanding it needs to act well in the world.

Claude itself also uses the constitution to construct many kinds of synthetic training data, including data that helps it learn and understand the constitution, conversations where the constitution might be relevant, responses that are in line with its values, and rankings of possible responses. All of these can be used to train future versions of Claude to become the kind of entity the constitution describes. This practical function has shaped how we’ve written the constitution: it needs to work both as a statement of abstract ideals and a useful artifact for training.