SPB Git

Repository Public

baselines.py 935 B 24 lines python

1"""Baselines. B0 identity (copy the rewrite); B1 generic paraphrase (rule-based mock paraphraser applied to the2rewrite — a paraphrase that does not know the target distribution); B2 pretrained encoder–decoder fine-tuning is3implemented in ``htr.training.hf_finetune`` (optional, needs weights downloaded under their own licence)."""4from __future__ import annotations56from htr.rewrite.mock_provider import MockProvider7from htr.rewrite.prompts import build_messages8910def b0_identity(source_text: str, metadata: dict | None = None) -> str:11    return source_text121314_mock = MockProvider(seed=7)151617def b1_generic_paraphrase(source_text: str, metadata: dict | None = None) -> str:18    sys_, msgs = build_messages(source_text, "A")19    r = _mock.rewrite(sys_, msgs, 1000, {"variant": "A", "sample_index": 1})20    return r.text or source_text212223BASELINES = {"B0_identity": b0_identity, "B1_generic_paraphrase": b1_generic_paraphrase}24
Files main
Menu

Theme