Ornith 1.0 35B heretic APEX GGUF APEX quantized GGUF builds of thanet s/Ornith 1.0 35B heretic — a decensored (Heretic v1.4.0) checkpoint of deepreinforce ai/Ornith 1.0 35B, a Qwen3.5 family MoE reasoning + coding model with vision input. Built with APEX quantization — a MoE aware, per layer tensor type layout that keeps edge layers high precision and compresses the middle experts — sized to run on a 24 GB GPU . Vision (mmproj) preserved. Files File Tier Size BPW imatrix Notes Ornith 1.0 35B heretic Q6 K APEX I Quality.gguf I Quality ~22.8 GB 5.26 ✅ Recommended — best accuracy that fits 24 GB Ornith 1.0 35B heretic Q6 K APEX Quality.gguf Quality ~22.8 GB 5.26 ❌ Plain APEX control Ornith 1.0 35B heretic Q4 K M APEX I Compact.gguf I Compact ~16.5 GB 3.81 ✅ Long context / 16 GB headroom Ornith 1.0 35B heretic Q4 K M APEX Compact.gguf Compact ~16.5 GB 3.81 ❌ Plain APEX control mmproj Ornith 1.0 35B heretic f16.gguf vision projector ~0.9 GB — — Required for image input ( mmproj ) The imatrix (I ) variants were calibrated on a multi domain corpus (chat + code + reasoning + tool calling, no Wikipedia) with full 256 expert coverage . Architecture Qwen3 5MoeForConditionalGeneration ( qwen3…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy