darjyo-agri-1b-quantum

Model Description

This project integrates IBM Quantum hardware-generated measurement outcomes into a prompt conditioning pipeline for a classical 1B parameter language model, enabling quantum-state-dependent inference variability. The quantum subsystem does not modify model weights or architecture. It operates as an external stochastic control signal.

Technical Classification

What this is: QPU-driven stochastic prompt modulation layer over classical LLM inference.

What this is NOT:

• A quantum-trained model
• A quantum neural network
• A quantum attention mechanism

Architecture: Quantum circuit (ibm_kingston) → measurement collapse → deterministic prompt transformation → classical LLM inference

The quantum subsystem operates as an external stochastic control signal, not as part of model weights or training.

Quantum Enhancement Results

Tested on 8 diverse agricultural prompts:

Metric	Result
Success Rate	100% (8/8 prompts)
Quantum Confidence	33.6% avg (range 8-91%)
Output Change	Always different from baseline
Content Change	12 chars shorter avg (more concise)

Quantum State Distribution

• 1101: 37.5% of prompts
• 0101: 25.0% of prompts
• 1001: 25.0% of prompts
• 1000: 12.5% of prompts

Example Enhancement

Prompt: What is the best practice for crop rotation?

Version	Output Focus
Baseline	Generic CEC (22.1 cmolc/kg), lime application
Quantum-Enhanced	Specific CEC (34.9 cmolc/kg), profile intelligence system

Quantum state used: 0101 (91.2% confidence)

Usage

⚠️ Requires real IBM Quantum hardware access

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("persadian/darjyo-agri-1b")
tokenizer = AutoTokenizer.from_pretrained("persadian/darjyo-agri-1b")

*Run quantum-enhanced inference*

Statistical Significance Test

To verify quantum states causally influence LLM outputs, we performed Fisher's exact test on 8 prompts:

Metric	Value
Quantum runs with changed output	8/8 (100%)
Baseline runs with changed output	0/8 (0%)
Fisher's exact p-value	0.000078
Odds ratio	Infinite (perfect separation)
Cramér's V (effect size)	0.875 (Large)

Interpretation: P-value < 0.05 indicates statistically significant association between quantum state and output change. The probability of observing this pattern by chance is less than 0.01%.

📊 Full statistical test results

Citation

If you use this model or architecture in your research, please cite:

@misc{darjyo-quantum-2026,
  author = {Darshani Persadh},
  title = {darjyo-agri-1b-quantum: Quantum-Enhanced Agricultural LLM},
  year = {2026},
  publisher = {Hugging Face},
  doi = {10.57967/hf/9080},
  url = {https://doi.org/10.57967/hf/9080}
}

In Academic Papers

Persadh, D.R. (2026). darjyo-agri-1b-quantum: Quantum-Enhanced Agricultural LLM. Hugging Face. https://doi.org/10.57967/hf/9080

Acknowledgments

• IBM Quantum for ibm_kingston access
• Qiskit developers for quantum computing framework
• Hugging Face for model hosting