Instructions to use Yingyaeliae/Hypnos-i1-8B-heretic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Yingyaeliae/Hypnos-i1-8B-heretic with Transformers:

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

pipe = pipeline("text-generation", model="Yingyaeliae/Hypnos-i1-8B-heretic")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Yingyaeliae/Hypnos-i1-8B-heretic")
model = AutoModelForCausalLM.from_pretrained("Yingyaeliae/Hypnos-i1-8B-heretic", 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 Yingyaeliae/Hypnos-i1-8B-heretic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Yingyaeliae/Hypnos-i1-8B-heretic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Yingyaeliae/Hypnos-i1-8B-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Yingyaeliae/Hypnos-i1-8B-heretic

SGLang

How to use Yingyaeliae/Hypnos-i1-8B-heretic 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 "Yingyaeliae/Hypnos-i1-8B-heretic" \
    --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": "Yingyaeliae/Hypnos-i1-8B-heretic",
		"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 "Yingyaeliae/Hypnos-i1-8B-heretic" \
        --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": "Yingyaeliae/Hypnos-i1-8B-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Yingyaeliae/Hypnos-i1-8B-heretic with Docker Model Runner:
```
docker model run hf.co/Yingyaeliae/Hypnos-i1-8B-heretic
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

This is a decensored version of adamm-hf/Hypnos-i1-8B, made using Heretic v1.4.0

This model is reproducible!

See the README in the reproduce directory for more information.

Abliteration parameters

Parameter	Value
direction_index	19.33
attn.o_proj.max_weight	1.48
attn.o_proj.max_weight_position	20.62
attn.o_proj.min_weight	1.39
attn.o_proj.min_weight_distance	15.11
mlp.down_proj.max_weight	1.22
mlp.down_proj.max_weight_position	21.39
mlp.down_proj.min_weight	1.16
mlp.down_proj.min_weight_distance	13.56

Performance

Metric	This model	Original model (adamm-hf/Hypnos-i1-8B)
KL divergence	0.0431	0 (by definition)
Refusals	3/100	15/100

Hypnos i1-8B (Quantum-Informed Reasoning Model)

🌌 Model Overview

Hypnos i1 8B is a specialized reasoning model based on Nous Hermes 3 (Llama 3.1 8B), designed to excel in complex logic, chain-of-thought (CoT) reasoning, and mathematical problem-solving.

It represents a unique experiment in Hybrid Quantum-Classical Machine Learning. Unlike standard fine-tunes, Hypnos i1 was trained on a dataset enriched with real entropy data generated by IBM Quantum Heron processors (133/156-qubit architecture). This "Quantum Noise Injection" serves as a stochastic regularizer, aiming to improve the model's creativity and break deterministic patterns in generation.

⚡ Key Features

S-Tier Reasoning: Outperforms standard 8B models in logic and math, rivaling 70B class models in specific, narrow tasks (e.g., multi-step logic puzzles, causal inference).
Quantum-Informed: The first known LLM fine-tuned on raw measurement data from 100+ qubit GHZ states generated on IBM's latest quantum hardware.
Uncensored & Compliant: Built on the robust Nous Hermes 3 base, it follows instructions without refusal or moralizing lectures, while maintaining safety for general use.
Deep Thinker: Optimized for long-context reasoning (4096+ tokens). It tends to "think out loud" before answering, ensuring higher accuracy on complex queries.

📊 Performance Benchmarks

🧬 The Hypnos Family

Model	Parameters	Quantum Sources	Best For	Status
Hypnos-Colossus-1T	1T (MoE)	3 (IBM + IQM + Cosmic)	Deep Simulation, Grand Challenges	🌌 Flagship
Hypnos-i2-32B	32B	3 (Matter + Light + Nucleus)	Production, Research	✅ Stable
Hypnos-i1-8B	8B	1 (Matter only)	Edge, Experiments	✅ 10k+ Downloads

Which one to choose?

Colossus 1T: For when you need maximum reasoning depth.
i2-32B: The "Giant Killer" - best balance of logic and efficiency for consumer GPUs.
i1-8B: Perfect for laptops and rapid prototyping.

⚛️ The Quantum Experiment (Training Methodology)

Hypnos i1 introduces a novel concept: Data-Driven Stochastic Regularization via Quantum Entropy.

During the Supervised Fine-Tuning (SFT) stage, the model was exposed to raw bitstring measurements from entangled quantum states (GHZ). These patterns contain true quantum randomness and specific hardware noise that cannot be simulated algorithmically.

Hardware Used for Data Generation:

IBM Quantum Heron r2 (ibm_fez): 156 Qubits
IBM Quantum Heron r1 (ibm_torino): 133 Qubits

Verified Quantum Job IDs (IBM Quantum Platform):

d4gcir92bisc73a3d29g (Torino - High Entropy Run)
d4gcoqscdebc73f10g3g (Fez - Domain Wall Phenomena)
d4go61olslhc73d0u1ig (Fez - Baseline)

Theoretical Impact: This injection of "Out-of-Distribution" quantum data forces the model's attention mechanism to adapt to non-linguistic, high-entropy patterns. In practice, this results in a model that is less prone to "mode collapse" (repetitive loops) and exhibits a unique "temperature" in creative writing tasks.

Downloads last month: 157

Safetensors

Model size

8B params

Tensor type

F16

Model tree for Yingyaeliae/Hypnos-i1-8B-heretic

Base model

meta-llama/Llama-3.1-8B

Finetuned

NousResearch/Hermes-3-Llama-3.1-8B

Finetuned

(35)

this model

Quantizations

2 models

Yingyaeliae
/

Hypnos-i1-8B-heretic

This is a decensored version of adamm-hf/Hypnos-i1-8B, made using Heretic v1.4.0

Abliteration parameters

Performance

Hypnos i1-8B (Quantum-Informed Reasoning Model)

🌌 Model Overview

⚡ Key Features

📊 Performance Benchmarks

🧬 The Hypnos Family

⚛️ The Quantum Experiment (Training Methodology)

Hardware Used for Data Generation:

Verified Quantum Job IDs (IBM Quantum Platform):

Model tree for Yingyaeliae/Hypnos-i1-8B-heretic

Datasets used to train Yingyaeliae/Hypnos-i1-8B-heretic