Ornith-Qwopus-KAT-Coder-35B-Merged

This model is a tensor-streamed SLERP merge ($Alpha=0.5$) of two leading 35B Mixture-of-Experts (MoE) models:

1. OliviaRossi/Qwopus-KAT-Coder-35B-Merged: Fusion of Qwopus 3.6 reasoning with KAT-Coder V2.5 autonomous SWE tool-calling and MTP speculative decoding.
2. ornith-ai/Ornith-1.5-35B-A3B: Advanced reasoning, complex logic planning, and general instruction following.

Architecture Highlights

• Backbone: Hybrid Sparse MoE (40 layers, 256 routed experts, 8 active per token + 1 shared expert).
• Attention: Hybrid GatedDeltaNet linear recurrence interleaved with standard self-attention.
• Active Parameters: ~3B active parameters per token (out of 35B total).
• Speculative Decoding: Inherits Multi-Token Prediction (MTP) head for accelerated generation.