"""Build fixed reference history for final-turn OmniVChat evaluation. The caller supplies an adapter that encodes a video's path, including its audio track, in the target model's native user-content format. Only media paths and intermediate reference replies enter the messages; rubrics remain with the scorer. """ from __future__ import annotations from collections.abc import Callable from typing import Any def build_messages( row: dict[str, Any], make_user_content: Callable[[str], Any], ) -> list[dict[str, Any]]: """Return alternating user/assistant messages ending with the final user turn. ``make_user_content`` receives each repository-relative video path. It must preserve the video's audio. Resolve paths against the downloaded dataset root. Missing history is an error: generating substitute replies changes the benchmark. """ turns = row.get("turns") if not isinstance(turns, list) or len(turns) < 2: raise ValueError("A multi-turn instance must contain at least two turns") if type(row.get("num_turns")) is not int or row["num_turns"] != len(turns): raise ValueError("num_turns must equal the number of turns") # Validate the whole record before invoking the model-specific media adapter. for index, turn in enumerate(turns, 1): if not isinstance(turn, dict) or type(turn.get("round")) is not int: raise ValueError(f"Turn {index} must have an integer round") if turn["round"] != index: raise ValueError("Rounds must be ordered consecutively starting at 1") if not isinstance(turn.get("video"), str) or not turn["video"].strip(): raise ValueError(f"Turn {index} is missing its video path") reference = turn.get("reference_reply") if index < len(turns): if not isinstance(reference, str) or not reference.strip(): raise ValueError(f"Turn {index} is missing its reference_reply") elif reference is not None: # Some dataset loaders materialize an absent struct field as null. raise ValueError("The final turn must not contain a reference_reply") messages = [] for index, turn in enumerate(turns): messages.append({ "role": "user", "content": make_user_content(turn["video"]), }) if index < len(turns) - 1: messages.append({ "role": "assistant", "content": turn["reference_reply"], }) return messages