Ornith-1.0 model family claims SOTA across four sizes
Four models from 9B to 397B parameters all report state-of-the-art results
Deep Dive
A new Hugging Face collection called "ornith-10" by deepreinforce-ai includes four models: 9B Dense, 31B Dense, 35B MoE, and 397B MoE — and claims state-of-the-art on various benchmarks, though the reporter adds "let's see if this holds."
Key Points
- Four model sizes: 9B, 31B Dense and 35B, 397B MoE architectures
- Claims state-of-the-art performance on multiple unknown benchmarks
- 397B MoE variant pushes open-weight model scale to new heights
Why It Matters
Ornith-1.0 could democratize access to GPT-4-class performance through open-weight MoE models.