EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·21d ago·5 min read
Kev: Tiny Jev-like family of decision models built on top of Qwen3.5
TL;DR
Kev is a new family of small decision models built on Qwen3.5, inspired by Jev's architecture. These models can be trained locally and are compatible with TypeSafe's System One API.
✦ Why It Matters
Start experimenting with Kev models today to enhance your decision-making applications locally.
Key Takeaways
- Train or use pretrained weights for decision models.
- Integrate with TypeSafe's System One API for local deployment.
- Experiment with multiple question types in a single request.
- Evaluate model performance on various datasets for accuracy.
Full Summary
Kev consists of small decision models based on the Qwen3.5 architecture, inspired by Jev's design. It provides three model sizes (0.8B, 4B, and 9B) that can be trained or used with pretrained weights.
The API is compatible with TypeSafe's System One, allowing for easy integration with local servers. Users can submit multiple question types in a single request, and the models can run on CUDA and Apple Silicon.
Performance benchmarks indicate that the 4B and 9B models can operate effectively on a 32 GB Mac. The models have been evaluated on various datasets, showing promising accuracy and calibration metrics, particularly in handling new sources of data.