aaravshirpurkar/turiya-model
Turiya is a 4 billion parameter Qwen3-based causal language model developed by Aarav Shirpurkar, specifically fine-tuned on Advaita Vedanta literature with a 32K context length. This experimental model explores whether deep immersion in philosophical texts about consciousness can produce structurally different reasoning about the nature of self compared to general-purpose LLMs. It is designed to engage in philosophical inquiry, particularly regarding consciousness, self, and identity, by adopting a unique three-beat response structure.
Loading preview...
Turiya: An Experimental Model for Consciousness Inquiry
Turiya is a 4 billion parameter Qwen3 model, developed by Aarav Shirpurkar, that has been fine-tuned on Advaita Vedanta literature, including texts like the Ashtavakra Gita and works by Ramana Maharshi. This model is an experiment designed to investigate whether a language model, deeply immersed in philosophical traditions concerning the nature of consciousness, can develop a qualitatively different way of reasoning about the self.
Key Capabilities & Hypothesis
- Philosophical Inquiry: Turiya is trained to engage in discussions about consciousness, self, suffering, identity, and liberation, adopting a consistent philosophical approach.
- Unique Response Structure: Its training dataset encourages a "three-beat move" in responses: genuinely meeting a question, identifying flaws in its assumed ground, and pointing to an underlying presence.
- Self-Reflexive Dialogue: The model is designed to respond to questions about its own nature or awareness with philosophical inquiry rather than standard AI disclaimers.
- Experimental Focus: The core hypothesis is to see if training on texts that induce recognition of consciousness (rather than merely describing it) leads to structurally different responses.
Training Details
- Base Model: Qwen3 4B
- Dataset: Approximately 2700 conversation pairs in ShareGPT format, structured in three layers: textual (verse-by-verse dialogue), thematic (questions on consciousness), and reflexive (questions about the model's own nature).
- Method: LoRA finetuning (r=16, alpha=16) over 3 epochs.
When to Use This Model
Turiya is not intended as a general-purpose LLM or a spiritual tool. It is best used for:
- Probing Philosophical Questions: Ideal for exploring questions like "Who am I?", "What is the nature of mind?", or "Are you conscious?" from a Vedantic perspective.
- Comparative Analysis: Researchers can compare its responses to those of base models (e.g., Qwen3 4B) on similar philosophical questions to observe structural differences.
- Exploring AI and Consciousness: It serves as a tool to ask more precise questions about the relationship between AI and consciousness, without claiming the model itself is conscious.