paper2sw

device: cpu precision: bf16 enable_cache: true selection_keep_ratio: 0.5 backend: dummy

Configuration

You can customize Paper2SW using a YAML or JSON config file. Pass it to the CLI with --config or use it in Python.

Example config file

model_id: paper2sw/paper2sw-diff-base   # Model to use
device: cpu                            # 'cpu' or 'cuda' for GPU
precision: bf16                        # Model precision: bf16, fp16, or fp32
enable_cache: true                     # Speed up repeated runs
selection_keep_ratio: 0.5              # Keep top fraction of relevant text
backend: dummy                         # Backend (for future extensions)

Using the config file

Command Line

uv run paper2sw predict --paper ./README.md --config ./predict.yaml --out sw.jsonl

Python

from paper2sw import Predictor, load_config

cfg = load_config("./predict.yaml")
predictor = Predictor.from_config(cfg)

You can adjust any option in the config file to fit your hardware or workflow.

This site is open source. Improve this page.