Cleven12/goshawk-r1-nano

Goshawk R1 is an edge-optimized reasoning language model combining DeepSeek Multi-Head Latent Attention (MLA), DeepSeekMoE, and Group Relative Policy Optimization (GRPO).

Architecture Configuration

  • Hidden Size (d_model): 192
  • Layers (n_layers): 4
  • Attention Heads (n_heads): 4
  • MLA KV Latent Rank (kv_lora_rank): 32
  • Routed Experts: 4 (Top-2 active) + 1 Shared
  • Multi-Token Prediction Depth: 1
Downloads last month
39
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Space using Cleven12/goshawk-r1-nano 1