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

VISIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Dec 9, 2025License:apache-2.0Architecture:Transformer Open Weights Cold

samuellimabraz/Qwen3-VL-8B-rslora-r32-2 is an 8 billion parameter multimodal vision-language model, fine-tuned by Samuel Lima Braz from Qwen3-VL-8B-Instruct. Specialized for quantum computing tasks using Qiskit 2.0, it excels at interpreting visual representations like circuit diagrams and generating Qiskit code. This model was trained with rsLoRA for 2 epochs on the Quantum Assistant Dataset, demonstrating significant improvements in multimodal quantum code generation and conceptual explanations.

Loading preview...

Model Overview

This model, samuellimabraz/Qwen3-VL-8B-rslora-r32-2, is an 8 billion parameter multimodal vision-language model. It is a fine-tuned version of Qwen3-VL-8B-Instruct, specifically specialized for quantum computing tasks using Qiskit 2.0. The model can interpret visual representations of quantum computing, such as circuit diagrams, Bloch spheres, and measurement histograms, and generate Qiskit code.

Key Capabilities

  • Code Generation: Generates complete Qiskit code from natural language descriptions.
  • Function Completion: Completes function bodies from signatures and docstrings.
  • Visual Understanding: Interprets quantum circuit diagrams, Bloch spheres, and histograms.
  • Conceptual Explanations: Answers questions about quantum computing theory.
  • Qiskit 2.0 Compliant: Utilizes modern Qiskit APIs.

Performance Highlights

Fine-tuned using Rank-Stabilized Low-Rank Adaptation (rsLoRA) with rank 32 for 2 epochs on the Quantum Assistant Dataset, this model shows substantial improvements over its baseline. It achieved a +11.26 pp increase on Qiskit HumanEval Pass@1 and a +16.56 pp increase on Qiskit HumanEval Hard Pass@1. Notably, it demonstrates a +25.89 pp improvement in multimodal code generation (63.39% Pass@1) compared to the baseline, validating its domain-specific visual understanding.

Good For

  • Educational assistance in quantum computing with Qiskit.
  • Generating Qiskit circuits from descriptions or diagrams.
  • Understanding quantum circuit visualizations.
  • Rapid prototyping of quantum algorithms in research.

Finetunes
1 models