genlm-bytes 0.2.0

Algorithms for byte-level language modeling

GenLM Bytes is a Python library for byte-level language modeling. It contains algorithms for turning token-level language models into byte-level language models.

See the docs for details and basic usage.

Note: This project is under active development — expect bugs, missing features, and breaking changes. Please report any issues or suggestions in the issue tracker.

Quick Start

This library requires python>=3.11 and can be installed using pip:

pip install genlm-bytes

For faster and less error-prone installs, consider using uv:

uv pip install genlm-bytes

See DEVELOPING.md for details on how to install the project for development.

Usage

from genlm.bytes import ByteBeamState, BeamParams
from genlm.backend import load_model_by_name

# Load a token-level language model from a huggingface model name
# (Note: for fast GPU inference, specify `backend="vllm"`)
llm = load_model_by_name("gpt2-medium")

# Initialize a beam state with a maximum beam width of 5 and a prune threshold of 0.05 (higher threshold values lead to more aggressive pruning).
beam = await ByteBeamState.initial(llm, BeamParams(K=5, prune_threshold=0.05))

# Populate the beam state with byte context.
beam = await beam.prefill(b"An apple a day keeps the ")

# Get the log probability distribution over the next byte.
logp_next = await beam.logp_next()
logp_next.pretty().top(5)
# Example output:
# b'd' -0.5766762743944795
# b'b' -2.8732729803080233
# b's' -2.9816068063730867
# b'w' -3.3758250127787264
# b'm' -3.528177345847574

# Prune the beam and extend it with a new byte
new_beam = await (beam.prune() << 100) # 100 is the byte value of 'd'

See basic usage for a more detailed example.