paper2sw predict --paper <URL|PATH> --out <FILE|-> [--top_k K] [--keep_ratio R] [--seed S] [--no_cache]
[--backend NAME] [--model_id ID] [--device DEV] [--precision P]
[--cache_dir DIR] [--format {jsonl,csv}]
paper2sw batch --papers <P1 P2 ...> --out_dir <DIR> [--top_k K] [--keep_ratio R] [--seed S] [--no_cache]
[--backend NAME] [--model_id ID] [--device DEV] [--precision P]
[--cache_dir DIR] [--format {jsonl,csv}]
paper2sw schema
paper2sw version
!!! note “Output formats”
Use --format csv to produce CSV instead of JSONL. Use --out - to stream results to stdout.
--paper: input URL or path--out: output path, or - for stdout--format: output format (jsonl default, or csv)--top_k: number of predictions--keep_ratio: fraction of text to keep for long-context selection (0..1)--no_cache: disable cache for this run--cache_dir: override cache directory (default: ~/.cache/paper2sw)--backend: backend id (reserved for future models)--model_id: model identifier (default: paper2sw/paper2sw-diff-base)--device: compute device (e.g., cpu, cuda:0)--precision: numeric precision (e.g., bf16, fp16, fp32)Examples ```bash title=”Predict and write JSONL” paper2sw predict –paper ./README.md –out sw.jsonl –top_k 5
```bash title="Predict and write CSV"
paper2sw predict --paper ./README.md --out sw.csv --top_k 5 --format csv
```bash title=”Stream JSONL to stdout” paper2sw predict –paper ./README.md –out - –top_k 3 | head
```bash title="Batch mode to CSV files"
paper2sw batch --papers ./README.md ./LICENSE --out_dir outs --top_k 2 --format csv
```bash title=”Print JSON schema” paper2sw schema
```bash title="Print version"
paper2sw version