microml 0.1.2

The easy, open source, and lightweight ML library for beginners.

microML

The easy, open source, and lightweight ML library for beginners.

Quick Start

from microml import LoadDS, SplitDataset, Tokenize, DefModel, Train, Infer, SaveChkp, LoadModel

# 1. Load data
data = LoadDS("dataset.txt")
train_data, val_data = SplitDataset(data, 0.8)

# 2. Tokenize
tokenized = Tokenize(train_data, "Text")
val_tokens = Tokenize(val_data, "Text")

# 3. Define model
model = DefModel("generative", "tiny", "Text", learning_rate=1e-3, batch_size=4)
model.summary()

# 4. Train
model, history = Train(
    model, steps=50, epochs=2, tokens=tokenized, device="CPU",
    validation_data=val_tokens, print_every=25
)

# 5. Inference
result = Infer(model, "Hello", temp=0.8, max_length=20, tokenizer=tokenized["tokenizer"])
print(result["response"])

# 6. Save
SaveChkp("my_model.mml")

Installation

pip install torch numpy tqdm
pip install -e .

For optional features:

# Image support
pip install -e ".[image]"

# Audio support
pip install -e ".[audio]"

# Video support
pip install -e ".[video]"

# Everything
pip install -e ".[all]"

# Development (tests, etc.)
pip install -e ".[dev]"

What Can microML Do?

microML makes it simple to train small models for many tasks:

• Text Generation -- Write a prompt, get generated text
• Chat Models -- Train conversational AI
• Image Classification -- Classify images with a CNN
• Image Autoencoders -- Compress and reconstruct images
• Audio Generation -- Generate audio waveforms
• Video Classification -- Classify video clips
• Time Series Prediction -- Predict future values from sequences
• Regression -- Predict continuous numbers
• Classification -- Predict categories from text or numbers

Examples

See the examples/ folder for ready-to-run scripts:

• examples/text_generation.py -- Train a tiny language model
• examples/image_classifier.py -- Define a CNN for image classification
• examples/time_series_prediction.py -- Predict sine wave values with an LSTM

Running Tests

pytest tests/

Documentation

See Docs.md for the full API reference and beginner-friendly machine learning explanations.

License

MIT