The model processes text through three stages, left to right.
BPE tokenizer maps text to token IDs (vocab 248,320). Embedding lookup table maps each ID to a 2048-dim vector. mRoPE adds 3D positional encoding (25% partial rotary, sections [11,11,10]).
Each block = RMSNorm → Attention → +residual → RMSNorm → MoE FFN → +residual. Two residual connections preserve gradient flow through all 40 layers.
Final RMSNorm, then LM Head (2048 → 248,320, untied) produces logits. Optional MTP head predicts a second token for speculative decoding.
| Total params | ~35B |
| Active/token | ~3B |
| Hidden dim | 2048 |
| Layers | 40 (30 GDN + 10 Full) |
| Experts | 256 (8 active) |
| Heads Q/KV | 16 / 2 (GQA 8:1) |
| Head dim | 256 |
| Context | 262,144 tokens |