EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 14·5 min read
Qwen 3.8 27B is out: open weights, best local dense model yet
TL;DR
Qwen 3.8 27B has been released with open weights and fine-grained FP8 quantization, achieving performance metrics nearly identical to the original model. It offers a hosted API service for scalable inference with enhanced production features.
✦ Why It Matters
Explore the new Qwen 3.8 27B model for your AI projects and leverage its hosted API for scalable solutions.
Key Takeaways
- Utilize Qwen 3.8 for advanced coding tasks and research.
- Leverage the API for scalable inference without infrastructure concerns.
- Explore multimodal capabilities for image and video understanding.
- Adjust reasoning depth with the reasoning_effort parameter for optimal performance.
How It Works
Qwen 3.8 employs fine-grained FP8 quantization, which reduces the model's memory footprint while maintaining performance. This method allows for efficient deployment in various environments, making it suitable for high-throughput applications.