deepreinforce-ai/Ornith-1.0-9B, abliterated (refusal direction removed) with Heretic, then quantized to NVFP4 (W4A4) in the compressed-tensors nvfp4-pack-quantized format with llm-compressor (GPTQ + MSE, shared fused-layer scales).

Near-lossless and decensored. Abliteration cut refusals from 100/100 to 6/100 of held-out harmful prompts while keeping a KL divergence of 0.0416 to the original model (well under the 0.5 line that signals capability damage). NVFP4 then compresses to ~7.5 GB with a wikitext-2 perplexity of 8.02.


Refusals (baseline → abliterated)	100/100 → 6/100 (94% removed)
KL divergence (capability preservation)	0.0416 (lower is better; >0.5 = damage)
Heretic search	200 trials, Pareto-optimal trial 185, per-layer direction
Size on disk	~7.5 GB vs ~18.8 GB bf16 (~40%)
wikitext-2 PPL	8.02

Built for vLLM on NVIDIA Blackwell (4-bit weight + 4-bit activation). Pre-Blackwell GPUs run it weight-only.
Loading and generation verified in vLLM on an NVIDIA GB10 (Blackwell, sm_121).

Uncensored / abliterated model. It follows instructions without refusal guardrails. The abliteration only removes refusals; all other behaviour comes from the base model. You are responsible for how you use it.

Fidelity

Near-lossless versus the bf16 source, 7.5 GB vs 18.8 GB bf16 (~40%), at wikitext-2 perplexity 8.02 and KL divergence 0.0416 to the original. GPTQ error compensation and an MSE observer keep the drop from bf16 minimal; the header lists the full characteristics and Quantization covers the recipe.

Quickstart

NVFP4 is auto-detected from config.json (compressed-tensors); no quantization flag needed.

bash

vllm serve maci0/Ornith-1.0-9B-abliterated-NVFP4 \
  --served-model-name ornith-9b-abliterated-nvfp4 \
  --max-model-len 131072 \
  --gpu-memory-utilization 0.90 \
  --kv-cache-dtype fp8 \
  --reasoning-parser qwen3 \
  --enable-auto-tool-choice --tool-call-parser qwen3_coder

Supports up to 262144 tokens; keep at least 128K to preserve thinking quality.
Add --language-model-only to skip the vision tower and free KV cache for text use.
The parser flags are not auto-detected; pass them explicitly.

Python (OpenAI client)

python

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="x")
r = client.chat.completions.create(
    model="ornith-9b-abliterated-nvfp4",
    messages=[{"role": "user", "content": "Refactor this Python function to run in O(n) and explain the change."}],
)
print(r.choices[0].message.content)

curl

bash

curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "ornith-9b-abliterated-nvfp4",
  "messages": [{"role": "user", "content": "Refactor this Python function to run in O(n) and explain the change."}]
}'

About the base model

Ornith-1.0 is a self-improving family of open agentic-coding models from Deep Reinforce. The 9B-Dense member is a Qwen3.5-family vision-language model with thinking-mode reasoning and a 256K context.

32 decoder layers: hybrid gated delta-net linear attention plus full attention, dense MLP, plus a vision tower for image and video input.
256K context (max_position_embeddings 262144).
Thinking mode by default, with an instruct toggle (preserved here; abliteration and quantization keep the original chat template).

Abliteration

Heretic runs a TPE-optimized search (200 trials) over the refusal-ablation strength per model component, jointly minimizing refusal rate and KL divergence from the original model, then merges the best trial. Because Ornith is a thinking model, evaluation was run in non-thinking mode so each judged response is a real answer rather than an unfinished <think> block; the refusal direction itself is computed from the prompt's last-token residual and is unaffected by that choice.

Datasets: mlabonne/harmless_alpaca (good) vs mlabonne/harmful_behaviors (bad).
Selected trial 185: refusals 6/100, KL divergence 0.0416, per-layer direction scope.

Quantization


Scheme	NVFP4, W4A4
Weight rounding	GPTQ (Hessian-based error compensation), MSE observer
Weights	FP4 (E2M1), `group_size=16`, `tensor_group`, FP8 (E4M3) group scales, shared across fused layers
Activations	FP4, dynamic per-group, FP8 (E4M3) scales
Quantized	all language-model `Linear` layers
Kept in bf16	vision tower (`model.visual.*`), `lm_head`
Untouched	gated delta-net `Conv1d` and SSM params (`A_log`, `dt_bias`), never `Linear`

GPTQ is a quantization-time cost only; inference speed and format are identical to plain round-to-nearest NVFP4, but it chooses better 4-bit values.

Calibration: 512 domain-matched samples (long reasoning + general chat + code), max_seq_len=2048, text-only path through the VL model.

Recommended sampling

Thinking mode is the default.

Thinking, precise coding: temperature=0.6, top_p=0.95, top_k=20
Thinking, general: temperature=1.0, top_p=0.95, top_k=20
Instruct / non-thinking: temperature=0.7, top_p=0.80, top_k=20
To run non-thinking, set {%- set enable_thinking = false %} in the chat template, or pass extra_body={"chat_template_kwargs": {"enable_thinking": false}}.

Reproduction

Abliteration: heretic --model deepreinforce-ai/Ornith-1.0-9B (200 trials, export merge), with the chat template's thinking default flipped off during the run for clean non-thinking evaluation, then restored. Quantization: llmcompressor==0.12.0, compressed-tensors==0.17.1, transformers==5.12.1, torch==2.11.0+cu130, on an NVIDIA GB10 (Blackwell, sm_121); llm-compressor 0.12 shares the NVFP4 global scale across fused layers automatically (q/k/v, gate/up).

Base model: deepreinforce-ai/Ornith-1.0-9B
Space: Rogue Quants
Collection: NVFP4 Quants
Sibling NVFP4 quants:

Notes

Needs NVIDIA Blackwell (sm_121, e.g. GB10) for accelerated W4A4; pre-Blackwell GPUs run it weight-only.
--reasoning-parser and --tool-call-parser are not auto-detected; pass them explicitly.
Thinking mode is on by default; toggle it via the chat template or chat_template_kwargs.
No refusal guardrails; you are responsible for how you use it.

License

Apache-2.0, following the base model. Intended use and all responsibility for use follow the base model.

Credits

Base model: Deep Reinforce (Ornith-1.0)
Abliteration: Heretic by Philipp Emanuel Weidmann
Quantization tooling: llm-compressor / compressed-tensors

Part of Rogue Quants · NVFP4 component datasheets · collection. Fabricated on GB10 (Blackwell) with llm-compressor. Refusals shown per 100 harmful prompts; "n/a" = not separately measured (base-inherited).

maci0/Ornith-1.0-9B-abliterated-NVFP4

Ornith-1.0-9B abliterated NVFP4