samuellimabraz/Qwen3-VL-8B-rslora-r32-2

VISIONPricing:Input $0.727 / Output $5.405Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 9, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

samuellimabraz/Qwen3-VL-8B-rslora-r32-2 is an 8 billion parameter multimodal Vision-Language Model, fine-tuned from Qwen3-VL-8B-Instruct by samuellimabraz. Specialized for quantum computing with Qiskit 2.0, it excels at interpreting quantum circuit diagrams, Bloch spheres, and measurement histograms. The model generates Qiskit code from natural language and visual inputs, making it ideal for quantum code generation and conceptual explanations.

Loading preview...

Quantum Assistant: Multimodal Model for Quantum Computing

This model, samuellimabraz/Qwen3-VL-8B-rslora-r32-2, is a specialized 8 billion parameter multimodal Vision-Language Model based on Qwen3-VL-8B-Instruct. Developed by samuellimabraz, it is the first of its kind to be specifically fine-tuned for quantum computing tasks using Qiskit 2.0. The model demonstrates a unique ability to understand and process visual representations common in quantum mechanics, such as circuit diagrams, Bloch spheres, and measurement histograms.

Key Capabilities

  • Multimodal Understanding: Interprets quantum circuit diagrams, Bloch spheres, and histograms to generate code or answer questions.
  • Qiskit Code Generation: Generates complete Qiskit code from natural language descriptions and visual inputs.
  • Function Completion: Completes Qiskit function bodies based on signatures and docstrings.
  • Conceptual Explanations: Provides answers to theoretical questions about quantum computing.
  • Qiskit 2.0 Compliant: Utilizes modern Qiskit APIs like SamplerV2, EstimatorV2, and generate_preset_pass_manager.

Performance Highlights

The model was trained using Rank-Stabilized Low-Rank Adaptation (rsLoRA) for 2 epochs on the Quantum Assistant Dataset. It shows significant improvements over its baseline, Qwen3-VL-8B-Instruct, particularly in multimodal tasks:

  • +11.26 pp on Qiskit HumanEval (function completion).
  • +16.56 pp on Qiskit HumanEval Hard (code generation).
  • +25.89 pp on multimodal code generation, achieving 63.39% Pass@1.

This specialization makes it highly effective for tasks requiring both textual and visual understanding within the quantum computing domain.

Good For

  • Educational assistance in learning quantum computing with Qiskit.
  • Generating Qiskit code from descriptions or visual circuit diagrams.
  • Understanding and documenting quantum circuit visualizations.
  • Rapid prototyping of quantum algorithms in research.

Finetunes
2 models