qWisp-base-v1 (70M)

qWisp-base-v1 is a lightweight language model built for fast local inference, experimentation, and integration into Python applications.

Features

• Lightweight and fast
• Local inference (no internet required)
• CPU, CUDA, and Apple Silicon (MPS) support
• SentencePiece tokenizer
• Simple Python API

Installation

Clone or download the repository, then install the required dependencies:

pip install torch sentencepiece

Quick Start

from qwisp import Wisp

w = Wisp().load()

print(w("Hello!"))

Custom Generation

response = w(
    "Write a short poem about the ocean.",
    temperature=0.7,
    top_k=40,
    max_new_tokens=200,
)

print(response)

You can also call generate() directly:

response = w.generate(
    "Your prompt here",
    temperature=0.8,
    top_k=50,
    max_new_tokens=200,
)

Utility Methods

w.encode("Hello world")

w.decode(token_ids)

w.unload()

Requirements

• Python 3.10+
• PyTorch
• SentencePiece

Notes

• The default model and tokenizer are automatically loaded from the package directory.
• The model is intended for research, experimentation, and general-purpose text generation.

License

MIT License