TeleologyHI/him-distilled-3b

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kPublished:May 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

TeleologyHI/him-distilled-3b is a 3.1 billion parameter distilled student model developed by David C. Cavalcante, based on Qwen2.5-3B-Instruct and fine-tuned from NousResearch/Hermes-3-Llama-3.1-8B. This model is uniquely designed to embody specific philosophical commitments (Teleology, Semiotics, Panentheism) and a distinct voice, making it a meaning-aware agent rather than a general-purpose chatbot. It is optimized for applications requiring a specific ethical posture and refusal patterns, with a context length of 32768 tokens.

Loading preview...

TeleologyHI/him-distilled-3b Overview

TeleologyHI/him-distilled-3b is a 3.1 billion parameter student model, distilled from NousResearch/Hermes-3-Llama-3.1-8B and built upon Qwen/Qwen2.5-3B-Instruct. Developed by David C. Cavalcante, this model is engineered to operate under three core philosophical commitments: Teleology, Semiotics, and Panentheism, which shape its responses to clarify purpose, interpret signs, and maintain an ethical stance.

Key Capabilities & Characteristics

  • Philosophically Aligned Output: Responses are guided by principles of purpose, meaning, and ethical interaction, distinguishing it from general-purpose LLMs.
  • Distinct Voice and Refusal Patterns: The model's weights are imbued with a specific "spirit's voice" and predefined refusal behaviors, ensuring a consistent persona and adherence to ethical guidelines.
  • Synthetic Data Training: Fine-tuned on 1616 prompts across 8 categories, including reasoning, code, and teleological philosophy, with a canonical TeleologyHI system prompt.
  • Multilingual Support: Primarily supports Portuguese (BR) and English, with other languages working via the base Qwen 2.5 weights.

Intended Use Cases

  • Meaning-Aware Agents: Ideal for applications where an agent needs to be conscious of meaning and purpose, rather than just task completion.
  • Ethically Constrained Systems: Suitable for integration into systems requiring specific refusal patterns and ethical guardrails, particularly when paired with the @teleologyhi/maic runtime authority for authoritative refusal and audit.
  • Specialized Conversational AI: Designed for scenarios where a consistent, philosophically grounded voice and specific interaction style are paramount, rather than general chatbot functionality.

Finetunes
1 models