Quantum Calibration and Observable Processing Model (QCOP/M)

Model Summary

Total Parameters	~40B total, 5B active per token (MoE sparse activation)
Architecture	Mixture-of-Experts Vision-Language Model (MoE VLM)
Experts	384 experts, 16 active per token
Context Length	396,488 tokens
Precision	BF16
Input	Image (PNG, JPEG) + Text
Output	Text (technical analysis, conclusions, parameter extraction)
Training	Two-phase sequential SFT (72.5K entries)
Minimum vRAM	24GB (Custom runtime and Modded firmware)
Release Date	April 20, 2026
License	MIT

| Paper(-s): | A Unified Framework for Calibration, Decoding, and Resource-Aware Processing in Hybrid Quantum-Classical Machine Learning — Tunjay Akbarli (April 2026) | Publication Date | October 1, 2026 | | Publisher | The Royal Society | | Citation | Akbarli T. 2026 A Unified framework for calibration, decoding and resource-aware processing in hybrid quantum–classical machine learning. R. Soc. OpenSci. 13: 260812. | Article | R. Soc. Open Sci. (2026) 13 (10): 260812

Description

The QCOPM analyzes quantum computing calibration experiment plots and generates structured technical text across six analysis question categories. QCOPM was developed by Tunjay Akbarli as a quantum calibration vision-language model. This model is ready for commercial/non-commercial use.

Deployment Geography: Global

Quick Start

Suggested inference settings: temperature=0.2, max_tokens=16384.

Use Case

Quantum computing researchers, calibration engineers, and developers can use this model to analyze experiment plot images and generate technical descriptions, experimental conclusions, significance assessments, fit quality evaluations, parameter extractions, and experiment success classifications. Model outputs should be validated by domain experts before acting on experimental conclusions.

Benchmark Results

Question Type	Ising Cal-1	Qwen3.5-35B base	QCOPM
Q1 Technical Description	87.8	86.8	91.2
Q2 Experimental Conclusion	67.1	39.9	78.4
Q3 Experimental Significance	64.7	45.7	75.9
Q4 Fit Quality Assessment	90.5	52.7	94.1
Q5 Parameter Extraction	62.5	57.8	70.3
Q6 Experiment Success	75.3	50.6	84.6
Overall	74.7	55.5	82.4

Analysis

• State-of-the-Art Performance: QCOPM outperforms Ising Cal-1 by an average of 7.7 points overall.
• Experimental Reasoning: The most significant gains are observed in Experimental Conclusion (Q2) and Significance (Q3), where QCOPM demonstrates a deeper understanding of scientific context.
• Data Precision: In Parameter Extraction (Q5), QCOPM achieves a score of 70.3, setting a new baseline for high-accuracy numerical retrieval in technical documentation.

Methodology

Models were tested on a closed-set of 1,200 technical queries requiring multi-step reasoning, mathematical fit evaluation, and experimental validation. Each score represents the percentage of correctly processed tasks as verified by domain experts.

Model Architecture

Architecture Type: Mixture-of-Experts Vision-Language Model (MoE VLM)

Network Architecture: Integrated vision encoder for experiment plot images combined with the MoE language model for autoregressive text generation.

Number of model parameters: ~40B total parameters, ~5B active per token (256 experts, 8 active)

Training Methodology

Phase 1 — ICL-formatted SFT

• 23.8K in-context learning formatted entries
• Teaches the model to process multi-image demonstrations
• Learning rate: 1e-5, 1 epoch

Phase 2 — Zero-shot SFT

• 48.7K zero-shot entries, LLM-augmented
• Strengthens single-plot understanding across all question types
• Learning rate: 5e-6, 1 epoch

Total training data: 72.5K entries

Input

Input Type(s): Image, Text

Input Format(s):

• Image: PNG, JPEG
  
• Text: String
  

Input Parameters:

• Image: Two-Dimensional (2D)
  
• Text: One-Dimensional (1D)
  

Other Properties Related to Input: Single-image or multi-image quantum calibration experiment plots with text prompts delivered through an OpenAI-compatible API. Suggested inference settings: temperature=0.2 and max_tokens=16384. Context length: 262,144 tokens.

Output

Output Type(s): Text

Output Format(s):

• Text: String
  

Output Parameters:

• Text: One-Dimensional (1D)
  

Other Properties Related to Output: Natural language technical analysis, experimental conclusions, significance assessments, fit quality evaluations, parameter extractions, and experiment success classifications.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA hardware and software frameworks, the model achieves faster inference times compared to CPU-only solutions.

Software Integration

Runtime Engine(s):

• Custom runtime, BF16 serving precision
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Blackwell (sm100, B200/B300/GB200)
  
• NVIDIA Hopper (sm90, H100/H200/GH200)
  

Supported Operating System(s):

• Linux (Ubuntu 22.04+)
  
• Windows (26200.8246)
  

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks before deployment.

Model Version(s)

v1.0.0

Training and Evaluation Datasets

Training

Data Modality

• Image
  
• Text
  

Training Data Size

72.5K total entries (Phase 1: 23.8K ICL-formatted entries; Phase 2: 48.7K zero-shot entries).

Data Collection Method by dataset

• Synthetic (LLM-augmented)
  

Labeling Method by dataset

• Synthetic
  

Properties: Synthetically generated quantum calibration experiment plots with paired analytical text.

Evaluation Dataset

Data Collection Method by dataset:

• Synthetic
  

Labeling Method by dataset:

• Synthetic
  

Properties: Curated quantum calibration experiments with ground-truth labels derived from simulation parameters.

Inference

Acceleration Engine: Modded version of Sglang, BF16 precision
Test Hardware:

• 1x NVIDIA RTX 4080 (12GB)
  
• 2x NVIDIA A10 (24GB)
  

Citation

@article{10.1098/rsos.260812,
    author = {Akbarli, Tunjay},
    title = {A unified framework for calibration, decoding and resource-aware processing in hybrid quantum–classical machine learning},
    journal = {Royal Society Open Science},
    volume = {13},
    number = {10},
    pages = {260812},
    year = {2026},
    month = {10},
    abstract = {Hybrid quantum–classical machine learning on noisy intermediate-scale quantum (NISQ) devices suffers from three coupled overheads: hardware miscalibration, classical syndrome-decoding latency and shot noise on gradient estimators. We present quantum calibration and observable processing (QCOP), a unified closed-loop control layer integrating a spectral calibration engine (SCE), a hierarchical syndrome decoder (HSD) and an observable compression layer (OCL) within a single GPU-accelerated pipeline. Three theoretical bounds characterize the joint effect on logical error rate (LER), convergence rate and observable variance. On device-calibrated noise-model simulators (n=20 seeds), QCOP achieves up to 2× wall-clock training speed-up and approximately 40\% variational quantum eigensolver (VQE) energy error reduction relative to the unaugmented baseline. A hardware pilot on ibm\_nairobi at surface-code distance d=5 (2000 syndrome volumes, 8192 shots each, single calibration window) yields a 1.6× latency reduction and 1.9× LER reduction for the HSD over minimum-weight perfect matching (MWPM)—directionally consistent with simulation but quantitatively smaller, indicating noise sources beyond the model. This pilot exercises the HSD alone: the SCE and the OCL have not been run on physical hardware, and the three modules have never been executed jointly on a device. The end-to-end QCOP benefits—speed-up, energy error reduction and variance reduction—are therefore not hardware-validated, and end-to-end hardware validation of all three modules remains future work. Code and model weights are available at https://github.com/hekimoghlu/QCOP.},
    issn = {2054-5703},
    doi = {10.1098/rsos.260812},
    url = {https://doi.org/10.1098/rsos.260812},
    eprint = {https://royalsocietypublishing.org/rsos/article-pdf/doi/10.1098/rsos.260812/6172520/rsos.260812.pdf},
}