squ11z1/Hypnos-i1-8B

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Nov 22, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The squ11z1/Hypnos-i1-8B is an 8 billion parameter reasoning model based on Nous Hermes 3 (Llama 3.1 8B) architecture. It is uniquely fine-tuned using a novel "Quantum Noise Injection" method, incorporating real entropy data from IBM Quantum Heron processors. This model excels in complex logic, chain-of-thought reasoning, and mathematical problem-solving, aiming to improve creativity and break deterministic patterns in generation. It is optimized for long-context reasoning and suitable for edge deployments and rapid prototyping.

Loading preview...

Hypnos i1-8B: Quantum-Informed Reasoning Model

Hypnos i1-8B is an 8 billion parameter model built upon the Nous Hermes 3 (Llama 3.1 8B) architecture, specifically engineered for advanced reasoning tasks. Its core differentiator is a unique training methodology involving Hybrid Quantum-Classical Machine Learning. The model was fine-tuned on a dataset enriched with real entropy data generated by IBM Quantum Heron processors, a process termed "Quantum Noise Injection." This stochastic regularization aims to enhance the model's creativity and prevent deterministic generation patterns.

Key Capabilities

  • S-Tier Reasoning: Demonstrates strong performance in logic and mathematics, rivaling larger 70B class models in specific complex tasks like multi-step logic puzzles and causal inference.
  • Quantum-Informed: Represents the first known LLM fine-tuned using raw measurement data from 100+ qubit GHZ states generated on IBM's latest quantum hardware.
  • Uncensored & Compliant: Based on Nous Hermes 3, it follows instructions without refusal while maintaining general safety.
  • Deep Thinker: Optimized for long-context reasoning (4096+ tokens), often employing a "think out loud" approach for higher accuracy on intricate queries.

Training Methodology

The model's unique training involved exposing it to raw bitstring measurements from entangled quantum states (GHZ) during the Supervised Fine-Tuning (SFT) stage. This data, containing true quantum randomness and hardware noise from IBM Quantum Heron r1 and r2 processors, forces the model's attention mechanism to adapt to non-linguistic, high-entropy patterns. This approach is theorized to reduce "mode collapse" and introduce a distinct "temperature" in creative outputs.

Good For

  • Complex Logic & Math: Ideal for tasks requiring multi-step reasoning, problem-solving, and causal inference.
  • Edge & Experimental Use: Its 8B parameter size makes it suitable for deployment on consumer hardware and for rapid prototyping.
  • Exploring Novel AI: For researchers and developers interested in the intersection of quantum computing and large language models.