Zyphra/ZAYA1-reasoning-base

Full fine-tune vs Zyphra/ZAYA1-base. Fine-tune: 100% of the weights changed, across 80 of 80 layers.

Tensors changed: other 401/401, Embeddings 1/1, Norms 121/121, Attention 400/400, MLP 1560/1560

3 derivatives checked

• Zyphra/ZAYA1-8B: Different architecture
• Zyphra/ZAYA1-8B-legacy: Full fine-tune
• 1bit-MONSTER/ZAYA1-reasoning-base-GGUF: Not comparable yet

On the Hub · Downloads on Model Pulse