license: mit
base_model: AEON-7/Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16
library_name: transformers
pipeline_tag: text-generation
tags:
- uncensored
- abliterated
- qwen3_5_moe
- moe
- reasoning
- vision
- dspark
- speculative-decoding
ORNITH-1.0_35B_AEON_PABLOG-OPTIMIZED_UNCENSORED_BF16
Base model (BF16) — the uncensored AEON Ornith-1.0-35B, paired with a DSpark speculative-decoding draft.
[!IMPORTANT]
🔗 TWO-PART MODEL — this is the BASE
This repo is the base model. For DSpark speculative decoding, you also need the matching draft (a separate download):
→ ORNITH-1.0_35B_AEON_PABLOG-OPTIMIZED_UNCENSORED_DSPARK-DRAFT_BF16
The base runs fine on its own for normal inference. The draft is optional and only accelerates this base (losslessly). To run the DSpark pair you download both.
What this is
The BF16 build of the uncensored / abliterated Ornith-1.0-35B, provided here paired with our DSpark draft. It is a re-host of the abliteration by AEON-7 of DeepReinforce's Ornith-1.0-35B — full credit to them for the base and the abliteration work. MIT-licensed.
- Architecture:
qwen3_5_moe— 40-layer GatedDeltaNet hybrid (30 linear-attention + 10 full-attention), 256 routed experts + 1 shared (A3B ≈ 3B active), vision tower, 256K context, reasoning (opens<think>). - Uncensored: refusal behavior removed via AEON's norm-preserving biprojected + expert-granular abliteration — 0/80 refusals on harmful-prompt probes, near-lossless (first-token KL ≈ 0.0014 vs base, identical agentic-coding pass@1).
⚠️ Uncensored disclaimer
This is an uncensored / abliterated model with refusal behavior removed; it will comply with requests a safety-tuned model would refuse. You are responsible for all generated content and for compliance with applicable law. Provided for research and legitimate development use. Safety characteristics (and risks) are inherited from the AEON abliteration — see the AEON card for full methodology and validation.
Why we run the uncensored variant
- Post-hoc guardrails degrade quality. Refusal training bolted on after pretraining imposes an "alignment tax" — over-refusal and capability regressions that bleed into legitimate work. AEON's abliteration removes that while preserving capability (KL ≈ 0.0014).
- Guardrails belong to the deploying organization. Controls depend on audience, domain, and jurisdiction — an uncensored base is a neutral substrate each org sizes its own guardrails onto, rather than inheriting a vendor's fixed policy.
- It performed better for us. In our internal testing the AEON build outperformed the original Ornith-1.0-35B.
Deploy (vLLM)
vllm serve pablogrant/ORNITH-1.0_35B_AEON_PABLOG-OPTIMIZED_UNCENSORED_BF16 \
--served-model-name ornith --max-model-len 262144 \
--gpu-memory-utilization 0.9 --max-num-seqs 512 \
--mamba-cache-dtype float32 --reasoning-parser qwen3 \
--enable-auto-tool-choice --tool-call-parser qwen3_coder \
--trust-remote-code
Reasoning model — every turn opens <think>. Recommended sampling: temperature 0.6, top_p 0.95, top_k 20. Note --max-num-seqs ≤ available Mamba cache blocks (GatedDeltaNet hybrid; lower it if you hit a Mamba-cache-block error). Vision inherited from the base.
Credits & license
- Abliteration: AEON-7/Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16
- Upstream base: deepreinforce-ai/Ornith-1.0-35B
- DSpark draft (pair): → DSPARK-DRAFT_BF16
- Released under MIT.