[Model Request] Gemma 4 #810
Description
I would like to propose adding support for Gemma 4 as a built-in model in WebLLM.
Available Models
| Model | Total Params | Active Params | Architecture |
|---|---|---|---|
| Gemma 4 31B | 30.7B | 30.7B | Dense |
| Gemma 4 26B A4B | 25.2B | 3.8B | MoE( 128 experts, 8 active and 1 shared) |
| Gemma 4 E4B | 4.5B effective (8B with embeddings) | Dense | |
| Gemma 4 E2B | 2.3B effective (5.1B with embeddings) | Dense |
Why Gemma 4 for WebLLM?
- Native Multimodal Support (Vision/Audio)
- Superior Efficiency-to-Performance Ratio: Gemma 4 achieves state-of-the-art results while maintaining a smaller parameter count than competitors. This makes it stable for VRAM-constrained environments like Chrome/Edge.
- Apache 2.0 Licensing: The Apache 2.0 license removes commercial barriers. This will likely bring more web developers to the ecosystem.
- Optimized for Agentic Workflows: Gemma 4 is fine-tuned for "Chain of Thought" reasoning and tool-calling.
Weights are available on Hugging Face.