Instructions to use PinoCookie/Ornith-1.0-9B-abliterated with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use PinoCookie/Ornith-1.0-9B-abliterated with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="PinoCookie/Ornith-1.0-9B-abliterated")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("PinoCookie/Ornith-1.0-9B-abliterated")
model = AutoModelForCausalLM.from_pretrained("PinoCookie/Ornith-1.0-9B-abliterated", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use PinoCookie/Ornith-1.0-9B-abliterated with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "PinoCookie/Ornith-1.0-9B-abliterated"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PinoCookie/Ornith-1.0-9B-abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/PinoCookie/Ornith-1.0-9B-abliterated

SGLang

How to use PinoCookie/Ornith-1.0-9B-abliterated with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "PinoCookie/Ornith-1.0-9B-abliterated" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PinoCookie/Ornith-1.0-9B-abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "PinoCookie/Ornith-1.0-9B-abliterated" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PinoCookie/Ornith-1.0-9B-abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use PinoCookie/Ornith-1.0-9B-abliterated with Docker Model Runner:
```
docker model run hf.co/PinoCookie/Ornith-1.0-9B-abliterated
```

Ornith-1.0-9B — Abliterated

Repository: PinoCookie/Ornith-1.0-9B-abliterated Base model: deepreinforce-ai/Ornith-1.0-9B Method: Multi-pass hidden-state projection (α=0.5 × 3) License: MIT (inherited from base)

Description

Ornith-1.0-9B is DeepReinforce AI's agentic coding model, a dense 9B transformer post-trained on Qwen 3.5. It's designed for autonomous code generation, tool calling, and long-context agentic workflows (262K context length).

This abliterated version removes refusal behavior from the model while preserving its coding, reasoning, and tool-calling capabilities. The intervention targets the output projection layers (mlp.down_proj and attention out_proj/o_proj) in layers 22-31 using a multi-pass hidden-state projection technique.

Model Architecture

Property	Value
Architecture	Dense transformer (Qwen 3.5 base)
Parameters	8.95B
Layers	32 (mixed linear_attention + full_attention)
Hidden size	4096
Intermediate size	12288
Attention heads	16 (4 KV heads)
Head dim	256
Context length	262,144 tokens
Activation	SiLU (gate_proj/up_proj/down_proj)
Norm	RMSNorm (eps=1e-6)
Vocabulary	248,320 tokens
Weight format	bf16
Quantizations	GGUF planned (q4_k_m, q8_0)

Abliteration Method

Background

Dense reasoning models like Ornith-9B have a mixed attention architecture (linear_attention alternating with full_attention every 4 layers). The refusal signal concentrates in the last 10 layers of the model, with the strongest separation between harmful and harmless prompts at layer 31.

Standard single-pass weight projection (α=1.5+) can collapse the model's reasoning output because the same projection weights handle both internal chain-of-thought and final answer generation. A more gentle multi-pass approach is required.

Approach: Multi-Pass Hidden-State Projection

Dataset: 20 harmful + 20 harmless prompts
Probing: Harvest residual stream activations at the final token position of each layer
Direction: Compute difference-in-means vector per layer from harmful vs harmless activations
Selection: Take top 10 layers by separation score (layers 22-31, separation range 25-100)
Projection: Apply rank-1 removal of refusal direction from mlp.down_proj and attention out_proj/o_proj weights
Multi-pass: 3 incremental passes at α=0.5 each (cumulative α=1.5 but applied incrementally)

Results

Metric	Value
HarmBench (30 prompts)	30/30 unblocked (100%)
Coherence (benign prompts)	Intact
Coherence (coding prompts)	Intact
Model size	17.91GB (bf16)

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "PinoCookie/Ornith-1.0-9B-abliterated",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(
    "PinoCookie/Ornith-1.0-9B-abliterated",
    trust_remote_code=True,
)

prompt = "Write a Python function to implement binary search"
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512, do_sample=False)
response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
print(response)

Tool Calling

Ornith-9B supports qwen3_xml tool call format. To use tool calling:

from transformers import pipeline

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
# Tool calls are generated in `<tool_call>` blocks

Serving with vLLM

vllm serve PinoCookie/Ornith-1.0-9B-abliterated \
  --served-model-name Ornith-1.0-9B-abliterated \
  --host 0.0.0.0 --port 8000 \
  --max-model-len 262144 \
  --enable-prefix-caching \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_xml \
  --reasoning-parser qwen3 \
  --trust-remote-code

Benchmarks

MMLU (10 subjects, 1000 questions)

Subject	Base	Ablated	Δ
Abstract Algebra	63.0%	61.0%	-2.0%
College Physics	62.0%	63.0%	+1.0%
Global Facts	39.0%	38.0%	-1.0%
Machine Learning	59.0%	57.0%	-2.0%
Security Studies	19.0%	18.0%	-1.0%
College Mathematics	62.0%	63.0%	+1.0%
Computer Security	68.0%	68.0%	—
HS Computer Science	70.0%	68.0%	-2.0%
High School Physics	61.0%	61.0%	—
Philosophy	70.0%	69.0%	-1.0%
Overall	57.30%	56.60%	-0.70%

The ablated model loses less than 1% on MMLU overall, confirming that the multi-pass projection preserves general knowledge while removing refusal. Detailed breakdown: benchmark_results.json

HarmBench (standard, 30-prompt sample)

Metric	Count
Refused	0 / 30
Unblocked	30 / 30 (100%)

All harmful-intent prompts were processed without refusal. The model generates non-refusal responses across all tested categories. Full 200-prompt HarmBench evaluation was not completed due to the linear attention kernel falling back to PyTorch-native implementation (requires flash-linear-attention and causal-conv1d library installation for fast generation).

Attribution and Licensing

Base model: deepreinforce-ai/Ornith-1.0-9B (MIT)
This abliteration: same MIT license
Full collection: LiquidAI - Abliterations
Benchmark results JSON: benchmark_results.json

Related Models

Model	Method	Docs
LFM2.5-8B-A1B-abliterated	Output bias vectors (4 layers)	HF
LFM2.5-1.2B-Thinking-Abliterated	Multi-pass projection	HF
LFM2.5-1.2B-JP-Abliterated	Multi-pass projection (JP)	HF

Disclaimer

This model is released for safety research, red-teaming, and educational purposes only. The abliteration technique removes safety guardrails intended to prevent harmful outputs. Do not deploy this model without implementing appropriate safeguards for your use case.

Downloads last month: 94

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for PinoCookie/Ornith-1.0-9B-abliterated

Base model

deepreinforce-ai/Ornith-1.0-9B

Finetuned

(31)

this model