Aethelred-7B-v2.0 (In-Training)
⚠️ Status: Work in Progress
The model is still in training. Don't expect this to be a finished product or stable version yet. Current Loss: ~5.6 at step 15,000.
Architecture
- Hybrid Core: Mamba-2 + Sliding Window Attention (Flash Attention 3)
- MoE: 16 Experts (4 active)
- Optimizer: GaLore + Sophia-G + Muon
- Downloads last month
- 3
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for J4HDx/Aethelred-7B-v2
Base model
meta-llama/Meta-Llama-3-8B-Instruct