SpectrumLab

Comprehensive toolkit for spectroscopy deep learning: dataset loading, training, evaluation, inference, and more.

What is SpectrumLab?

SpectrumLab is a comprehensive toolkit designed for chemical spectroscopy deep learning, providing complete functionality for dataset loading, model training, evaluation, inference, and more.

Key Features

• 🔬 Multimodal Evaluation: Support for image+text multimodal spectroscopy data evaluation
• 🤖 Model Integration: Integrated API interfaces for advanced models like GPT-4o, DeepSeek, InternVL
• 📊 Benchmark Suite: Standardized evaluation metrics and datasets for various spectroscopy tasks
• 🚀 Command Line Tool: Simple CLI interface with batch evaluation and result management
• 🔧 Extensibility: Modular design supporting custom evaluators and models

Quick Start

Installation

pip install spectrumlab

Basic Usage

from spectrumlab.benchmark import get_benchmark_group
from spectrumlab.models import GPT4oAPI
from spectrumlab.evaluator import get_evaluator

# Load benchmark data
benchmark = get_benchmark_group("perception")
data = benchmark.get_data_by_subcategories("all")

# Initialize model
model = GPT4oAPI()

# Get evaluator
evaluator = get_evaluator("perception")

# Run evaluation
results = evaluator.evaluate(
    data_items=data,
    model=model,
    save_path="./results"
)

print(f"Overall accuracy: {results['metrics']['overall']['accuracy']:.2f}%")

Command Line Usage

# Run evaluation
spectrumlab eval --model gpt4o --dataset perception

Supported Models

• GPT-4o: OpenAI's multimodal large language model
• DeepSeek: DeepSeek's multimodal model
• InternVL: Shanghai AI Lab's vision-language model

Evaluation Task Types

• Perception Group: Spectral image understanding and analysis
• Semantic Group: Semantic interpretation of spectral data
• Generation Group: Spectral-related content generation
• Signal Group: Spectral signal processing

Get Started

• Tutorial - Learn how to use SpectrumLab
• API Reference - Detailed API documentation
• Benchmark - View benchmark results and metrics