Create 2nd version of translate_eqbench.py
Browse files
eqbench-ja/translate_eqbench_v2.py
ADDED
|
@@ -0,0 +1,693 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
translate_eqbench_v2.py — EQ-Bench3 日本語化スクリプト (Version 2)
|
| 3 |
+
|
| 4 |
+
フロー:
|
| 5 |
+
1. EQ-Bench3 の全データファイル (12ファイル) を読み込む
|
| 6 |
+
2. vLLM(Qwen3.5-9B)経由で各ファイルを日本語に翻訳
|
| 7 |
+
3. 翻訳結果を保存(チェックポイント対応・再開可能)
|
| 8 |
+
4. 完成した日本語版ファイルを output/ に書き出し
|
| 9 |
+
5. HF Hub にアップロード
|
| 10 |
+
|
| 11 |
+
実行例:
|
| 12 |
+
python translate_eqbench_v2.py
|
| 13 |
+
python translate_eqbench_v2.py --dry-run # 最初の2シナリオのみ
|
| 14 |
+
python translate_eqbench_v2.py --target-file scenario_notes.txt # 特定ファイルのみ
|
| 15 |
+
"""
|
| 16 |
+
|
| 17 |
+
from __future__ import annotations
|
| 18 |
+
|
| 19 |
+
import argparse
|
| 20 |
+
import asyncio
|
| 21 |
+
import json
|
| 22 |
+
import os
|
| 23 |
+
import sys
|
| 24 |
+
import time
|
| 25 |
+
import traceback
|
| 26 |
+
from datetime import datetime, timezone
|
| 27 |
+
from pathlib import Path
|
| 28 |
+
|
| 29 |
+
import httpx
|
| 30 |
+
from tqdm import tqdm
|
| 31 |
+
|
| 32 |
+
# ── 設定 ──────────────────────────────────────────────────────
|
| 33 |
+
VLLM_BASE_URL: str = os.environ.get("VLLM_BASE_URL", "http://localhost:8000/v1")
|
| 34 |
+
TRANSLATE_MODEL: str = os.environ.get("TRANSLATE_MODEL", "Qwen/Qwen3.5-9B")
|
| 35 |
+
HF_TOKEN: str = os.environ.get("HF_TOKEN", "")
|
| 36 |
+
HF_USERNAME: str = os.environ.get("HF_USERNAME", "YUGOROU")
|
| 37 |
+
HF_REPO: str = os.environ.get("HF_REPO", f"{HF_USERNAME}/teememo-eq-bench-ja")
|
| 38 |
+
|
| 39 |
+
EQBENCH_DIR: Path = Path(os.environ.get("EQBENCH_DIR", "/workspace/eqbench-ja/eqbench3"))
|
| 40 |
+
OUTPUT_DIR: Path = Path(os.environ.get("OUTPUT_DIR", "/workspace/eqbench-ja/output"))
|
| 41 |
+
CHECKPOINT_FILE: Path = Path(os.environ.get("CHECKPOINT_FILE", "/workspace/eqbench-ja/output/.checkpoint_translate_v2.json"))
|
| 42 |
+
|
| 43 |
+
MAX_CONCURRENT: int = int(os.environ.get("MAX_CONCURRENT", "4"))
|
| 44 |
+
MAX_RETRIES: int = int(os.environ.get("MAX_RETRIES", "3"))
|
| 45 |
+
REQUEST_TIMEOUT: float = float(os.environ.get("REQUEST_TIMEOUT", "180.0"))
|
| 46 |
+
|
| 47 |
+
# ── 翻訳対象ファイル定義 ──────────────────────────────────────
|
| 48 |
+
# ファイルタイプ:
|
| 49 |
+
# "scenario_split" : シナリオIDで分割 (######## 1 | ...)
|
| 50 |
+
# "note_split" : ノート形式で分割 (# 1\n<note>)
|
| 51 |
+
# "single" : 単一ファイル(全体を翻訳)
|
| 52 |
+
# "analysis_split" : 分析用シナリオ分割形式
|
| 53 |
+
|
| 54 |
+
TARGET_FILES: dict[str, dict] = {
|
| 55 |
+
# シナリオプロンプト(シナリオID分割)
|
| 56 |
+
"scenario_prompts.txt": {
|
| 57 |
+
"type": "scenario_split",
|
| 58 |
+
"description": "メインシナリオプロンプト",
|
| 59 |
+
},
|
| 60 |
+
# シナリオノート(ノート形式分割)
|
| 61 |
+
"scenario_notes.txt": {
|
| 62 |
+
"type": "note_split",
|
| 63 |
+
"description": "シナリオ採点ノート",
|
| 64 |
+
},
|
| 65 |
+
# メッセージ作成用マスタープロンプト
|
| 66 |
+
"scenario_master_prompt_message_drafting.txt": {
|
| 67 |
+
"type": "single",
|
| 68 |
+
"description": "メッセージ作成用マスタープロンプト",
|
| 69 |
+
},
|
| 70 |
+
# 分析用マスタープロンプト
|
| 71 |
+
"scenario_master_prompt_analysis.txt": {
|
| 72 |
+
"type": "single",
|
| 73 |
+
"description": "分析用マスタープロンプト",
|
| 74 |
+
},
|
| 75 |
+
# 汎用マスタープロンプト
|
| 76 |
+
"scenario_master_prompt.txt": {
|
| 77 |
+
"type": "single",
|
| 78 |
+
"description": "汎用マスタープロンプト",
|
| 79 |
+
},
|
| 80 |
+
# ルーブリック採点プロンプト(分析用)
|
| 81 |
+
"rubric_scoring_prompt_analysis.txt": {
|
| 82 |
+
"type": "single",
|
| 83 |
+
"description": "分析用ルーブリック採点プロンプト",
|
| 84 |
+
},
|
| 85 |
+
# ルーブリック採点プロンプト
|
| 86 |
+
"rubric_scoring_prompt.txt": {
|
| 87 |
+
"type": "single",
|
| 88 |
+
"description": "ルーブリック採点プロンプト",
|
| 89 |
+
},
|
| 90 |
+
# ルーブリック採点基準(分析用)
|
| 91 |
+
"rubric_scoring_criteria_analysis.txt": {
|
| 92 |
+
"type": "single",
|
| 93 |
+
"description": "分析用ルーブリック採点基準",
|
| 94 |
+
},
|
| 95 |
+
# ルーブリック採点基準
|
| 96 |
+
"rubric_scoring_criteria.txt": {
|
| 97 |
+
"type": "single",
|
| 98 |
+
"description": "ルーブリック採点基準",
|
| 99 |
+
},
|
| 100 |
+
# ペアワイズ比較プロンプト(分析用)
|
| 101 |
+
"pairwise_prompt_eqbench3_analysis.txt": {
|
| 102 |
+
"type": "single",
|
| 103 |
+
"description": "分析用ペアワイズ比較プロンプト",
|
| 104 |
+
},
|
| 105 |
+
# ペアワイズ比較プロンプト
|
| 106 |
+
"pairwise_prompt_eqbench3.txt": {
|
| 107 |
+
"type": "single",
|
| 108 |
+
"description": "ペアワイズ比較プロンプト",
|
| 109 |
+
},
|
| 110 |
+
# デブリーフプロンプト
|
| 111 |
+
"debrief_prompt.txt": {
|
| 112 |
+
"type": "single",
|
| 113 |
+
"description": "デブリーフプロンプト",
|
| 114 |
+
},
|
| 115 |
+
}
|
| 116 |
+
|
| 117 |
+
# ── 翻訳システムプロンプト ────────────────────────────────────
|
| 118 |
+
TRANSLATE_SYSTEM = """You are a professional translator specializing in Japanese localization of psychological and emotional intelligence assessment materials.
|
| 119 |
+
|
| 120 |
+
Your task is to translate English text into natural, fluent Japanese while:
|
| 121 |
+
1. Preserving the original meaning, nuance, and emotional tone precisely
|
| 122 |
+
2. Maintaining all formatting markers (######## , ####### , {{placeholder}}, etc.) exactly as-is
|
| 123 |
+
3. Keeping scenario numbers, category names, and structural markers unchanged
|
| 124 |
+
4. Using natural conversational Japanese appropriate for the social context described
|
| 125 |
+
5. Preserving any special instructions in brackets [like this] translated into Japanese
|
| 126 |
+
6. For role-play scenarios involving interpersonal conflict, use natural Japanese speech patterns including appropriate keigo or casual speech as the context demands
|
| 127 |
+
7. Translating all proper nouns contextually (names can be kept or given Japanese equivalents)
|
| 128 |
+
8. Keeping JSON keys, variable names, and code-like placeholders unchanged
|
| 129 |
+
|
| 130 |
+
Output ONLY the translated text with no explanations or commentary."""
|
| 131 |
+
|
| 132 |
+
|
| 133 |
+
# ── vLLM クライアント ─────────────────────────────────────────
|
| 134 |
+
|
| 135 |
+
class VLLMClient:
|
| 136 |
+
def __init__(self) -> None:
|
| 137 |
+
self._client = httpx.AsyncClient(
|
| 138 |
+
timeout=httpx.Timeout(REQUEST_TIMEOUT),
|
| 139 |
+
)
|
| 140 |
+
self._semaphore = asyncio.Semaphore(MAX_CONCURRENT)
|
| 141 |
+
|
| 142 |
+
async def close(self) -> None:
|
| 143 |
+
await self._client.aclose()
|
| 144 |
+
|
| 145 |
+
async def __aenter__(self) -> "VLLMClient":
|
| 146 |
+
return self
|
| 147 |
+
|
| 148 |
+
async def __aexit__(self, *_) -> None:
|
| 149 |
+
await self.close()
|
| 150 |
+
|
| 151 |
+
async def wait_for_server(self, max_wait: int = 300, interval: int = 5) -> None:
|
| 152 |
+
print(f"[client] vLLM サーバーの起動を待機中 (最大 {max_wait}s)...")
|
| 153 |
+
start = time.time()
|
| 154 |
+
while time.time() - start < max_wait:
|
| 155 |
+
try:
|
| 156 |
+
response = await self._client.get(f"{VLLM_BASE_URL}/models")
|
| 157 |
+
if response.status_code == 200:
|
| 158 |
+
print("[client] vLLM サーバーが起動しました。")
|
| 159 |
+
return
|
| 160 |
+
except Exception:
|
| 161 |
+
pass
|
| 162 |
+
await asyncio.sleep(interval)
|
| 163 |
+
raise TimeoutError(f"[client] vLLM サーバーが {max_wait}s 以内に起動しませんでした。")
|
| 164 |
+
|
| 165 |
+
async def translate(self, text: str) -> str:
|
| 166 |
+
"""テキストを日本語に翻訳する。"""
|
| 167 |
+
payload = {
|
| 168 |
+
"model": TRANSLATE_MODEL,
|
| 169 |
+
"messages": [
|
| 170 |
+
{"role": "system", "content": TRANSLATE_SYSTEM},
|
| 171 |
+
{"role": "user", "content": f"Translate the following to Japanese:\n\n{text}"},
|
| 172 |
+
],
|
| 173 |
+
"temperature": 0.1, # 翻訳は低温で安定させる
|
| 174 |
+
"top_p": 0.9,
|
| 175 |
+
"max_tokens": 8192, # 長いファイル対応
|
| 176 |
+
"chat_template_kwargs": {"enable_thinking": False},
|
| 177 |
+
}
|
| 178 |
+
|
| 179 |
+
last_exc = None
|
| 180 |
+
for attempt in range(MAX_RETRIES):
|
| 181 |
+
try:
|
| 182 |
+
async with self._semaphore:
|
| 183 |
+
response = await self._client.post(
|
| 184 |
+
f"{VLLM_BASE_URL}/chat/completions",
|
| 185 |
+
json=payload,
|
| 186 |
+
headers={"Content-Type": "application/json"},
|
| 187 |
+
)
|
| 188 |
+
response.raise_for_status()
|
| 189 |
+
result = response.json()
|
| 190 |
+
return result["choices"][0]["message"].get("content") or ""
|
| 191 |
+
except Exception as exc:
|
| 192 |
+
last_exc = exc
|
| 193 |
+
wait = 2 ** attempt
|
| 194 |
+
print(f"[client] リトライ {attempt+1}/{MAX_RETRIES} ({type(exc).__name__}) — {wait}s 待機")
|
| 195 |
+
await asyncio.sleep(wait)
|
| 196 |
+
raise RuntimeError(f"[client] 翻訳失敗: {last_exc}")
|
| 197 |
+
|
| 198 |
+
|
| 199 |
+
# ── パーサー関数群 ─────────────────────────────────────────────
|
| 200 |
+
|
| 201 |
+
def parse_scenario_prompts(filepath: Path) -> list[dict]:
|
| 202 |
+
"""
|
| 203 |
+
scenario_prompts.txt を解析してシナリオリストを返す。
|
| 204 |
+
|
| 205 |
+
Returns:
|
| 206 |
+
[{"id": 1, "header": "...", "prompts": {"Prompt1": "...", "Prompt2": "..."}}, ...]
|
| 207 |
+
"""
|
| 208 |
+
scenarios = []
|
| 209 |
+
current: dict | None = None
|
| 210 |
+
current_prompt_key: str | None = None
|
| 211 |
+
current_prompt_lines: list[str] = []
|
| 212 |
+
|
| 213 |
+
def _flush_prompt():
|
| 214 |
+
if current and current_prompt_key and current_prompt_lines:
|
| 215 |
+
current["prompts"][current_prompt_key] = "\n".join(current_prompt_lines).strip()
|
| 216 |
+
current_prompt_lines.clear()
|
| 217 |
+
|
| 218 |
+
with open(filepath, encoding="utf-8") as f:
|
| 219 |
+
for line in f:
|
| 220 |
+
line = line.rstrip("\n")
|
| 221 |
+
|
| 222 |
+
# シナリオヘッダー: ######## 1 | Work Dilemma | ...
|
| 223 |
+
if line.startswith("######## "):
|
| 224 |
+
_flush_prompt()
|
| 225 |
+
if current:
|
| 226 |
+
scenarios.append(current)
|
| 227 |
+
parts = line.split("|")
|
| 228 |
+
scenario_id = parts[0].replace("#", "").strip()
|
| 229 |
+
current = {
|
| 230 |
+
"id": int(scenario_id) if scenario_id.isdigit() else scenario_id,
|
| 231 |
+
"header": line,
|
| 232 |
+
"prompts": {},
|
| 233 |
+
}
|
| 234 |
+
current_prompt_key = None
|
| 235 |
+
current_prompt_lines = []
|
| 236 |
+
|
| 237 |
+
# プロンプトキー: ####### Prompt1
|
| 238 |
+
elif line.startswith("####### "):
|
| 239 |
+
_flush_prompt()
|
| 240 |
+
current_prompt_key = line.replace("#", "").strip()
|
| 241 |
+
current_prompt_lines = []
|
| 242 |
+
|
| 243 |
+
# プロンプト本文
|
| 244 |
+
else:
|
| 245 |
+
if current_prompt_key is not None:
|
| 246 |
+
current_prompt_lines.append(line)
|
| 247 |
+
|
| 248 |
+
_flush_prompt()
|
| 249 |
+
if current:
|
| 250 |
+
scenarios.append(current)
|
| 251 |
+
|
| 252 |
+
return scenarios
|
| 253 |
+
|
| 254 |
+
|
| 255 |
+
def parse_scenario_notes(filepath: Path) -> dict[str, str]:
|
| 256 |
+
"""
|
| 257 |
+
scenario_notes.txt を解析する。
|
| 258 |
+
# 1\n<note>\n# 2\n<note> の形式。
|
| 259 |
+
|
| 260 |
+
Returns:
|
| 261 |
+
{"1": "<note>", "2": "<note>", ...}
|
| 262 |
+
"""
|
| 263 |
+
notes = {}
|
| 264 |
+
current_key: str | None = None
|
| 265 |
+
current_lines: list[str] = []
|
| 266 |
+
|
| 267 |
+
def _flush():
|
| 268 |
+
if current_key and current_lines:
|
| 269 |
+
notes[current_key] = "\n".join(current_lines).strip()
|
| 270 |
+
current_lines.clear()
|
| 271 |
+
|
| 272 |
+
with open(filepath, encoding="utf-8") as f:
|
| 273 |
+
for line in f:
|
| 274 |
+
line = line.rstrip("\n")
|
| 275 |
+
if line.startswith("# ") and line[2:].strip().isdigit():
|
| 276 |
+
_flush()
|
| 277 |
+
current_key = line[2:].strip()
|
| 278 |
+
current_lines = []
|
| 279 |
+
else:
|
| 280 |
+
if current_key is not None:
|
| 281 |
+
current_lines.append(line)
|
| 282 |
+
|
| 283 |
+
_flush()
|
| 284 |
+
return notes
|
| 285 |
+
|
| 286 |
+
|
| 287 |
+
def parse_single_file(filepath: Path) -> str:
|
| 288 |
+
"""単一ファイル全体を読み込む。"""
|
| 289 |
+
return filepath.read_text(encoding="utf-8")
|
| 290 |
+
|
| 291 |
+
|
| 292 |
+
# ── ビルダー関数群 ─────────────────────────────────────────────
|
| 293 |
+
|
| 294 |
+
def build_output_prompts(
|
| 295 |
+
scenarios: list[dict],
|
| 296 |
+
translated: dict[str, dict],
|
| 297 |
+
) -> str:
|
| 298 |
+
"""翻訳済みシナリオを scenario_prompts.txt 形式に再構築する。"""
|
| 299 |
+
lines = []
|
| 300 |
+
for scenario in scenarios:
|
| 301 |
+
sid = str(scenario["id"])
|
| 302 |
+
if sid in translated:
|
| 303 |
+
trans = translated[sid]
|
| 304 |
+
lines.append(trans.get("header", scenario["header"]))
|
| 305 |
+
for prompt_key, prompt_text in scenario["prompts"].items():
|
| 306 |
+
lines.append(f"####### {prompt_key}")
|
| 307 |
+
translated_text = trans.get("prompts", {}).get(prompt_key, prompt_text)
|
| 308 |
+
lines.append(translated_text)
|
| 309 |
+
lines.append("")
|
| 310 |
+
else:
|
| 311 |
+
# 未翻訳はオリジナルをそのまま
|
| 312 |
+
lines.append(scenario["header"])
|
| 313 |
+
for prompt_key, prompt_text in scenario["prompts"].items():
|
| 314 |
+
lines.append(f"####### {prompt_key}")
|
| 315 |
+
lines.append(prompt_text)
|
| 316 |
+
lines.append("")
|
| 317 |
+
return "\n".join(lines)
|
| 318 |
+
|
| 319 |
+
|
| 320 |
+
def build_output_notes(
|
| 321 |
+
original_notes: dict[str, str],
|
| 322 |
+
translated_notes: dict[str, str],
|
| 323 |
+
) -> str:
|
| 324 |
+
"""翻訳済みノートを scenario_notes.txt 形式に再構築する。"""
|
| 325 |
+
lines = []
|
| 326 |
+
for key, note in original_notes.items():
|
| 327 |
+
lines.append(f"# {key}")
|
| 328 |
+
lines.append(translated_notes.get(key, note))
|
| 329 |
+
lines.append("")
|
| 330 |
+
return "\n".join(lines)
|
| 331 |
+
|
| 332 |
+
|
| 333 |
+
# ── チェックポイント ──────────────────────────────────────────
|
| 334 |
+
|
| 335 |
+
def load_checkpoint() -> dict:
|
| 336 |
+
"""
|
| 337 |
+
チェックポイント構造:
|
| 338 |
+
{
|
| 339 |
+
"files": {
|
| 340 |
+
"scenario_prompts.txt": {
|
| 341 |
+
"translated_scenarios": {...},
|
| 342 |
+
"status": "completed" | "in_progress"
|
| 343 |
+
},
|
| 344 |
+
"scenario_notes.txt": {
|
| 345 |
+
"translated_notes": {...},
|
| 346 |
+
"status": "completed" | "in_progress"
|
| 347 |
+
},
|
| 348 |
+
"single_file_name.txt": {
|
| 349 |
+
"translated": "...",
|
| 350 |
+
"status": "completed" | "in_progress"
|
| 351 |
+
},
|
| 352 |
+
...
|
| 353 |
+
}
|
| 354 |
+
}
|
| 355 |
+
"""
|
| 356 |
+
if CHECKPOINT_FILE.exists():
|
| 357 |
+
try:
|
| 358 |
+
return json.loads(CHECKPOINT_FILE.read_text(encoding="utf-8"))
|
| 359 |
+
except Exception:
|
| 360 |
+
pass
|
| 361 |
+
return {"files": {}}
|
| 362 |
+
|
| 363 |
+
|
| 364 |
+
def save_checkpoint(state: dict) -> None:
|
| 365 |
+
CHECKPOINT_FILE.parent.mkdir(parents=True, exist_ok=True)
|
| 366 |
+
CHECKPOINT_FILE.write_text(json.dumps(state, ensure_ascii=False, indent=2), encoding="utf-8")
|
| 367 |
+
|
| 368 |
+
|
| 369 |
+
# ── 翻訳処理関数群 ────────────────────────────────────────────
|
| 370 |
+
|
| 371 |
+
async def translate_scenario_file(
|
| 372 |
+
client: VLLMClient,
|
| 373 |
+
filepath: Path,
|
| 374 |
+
file_key: str,
|
| 375 |
+
state: dict,
|
| 376 |
+
dry_run: bool,
|
| 377 |
+
lock: asyncio.Lock,
|
| 378 |
+
) -> dict:
|
| 379 |
+
"""シナリオ分割ファイルを翻訳する。"""
|
| 380 |
+
scenarios = parse_scenario_prompts(filepath)
|
| 381 |
+
|
| 382 |
+
if file_key not in state["files"]:
|
| 383 |
+
state["files"][file_key] = {"translated_scenarios": {}, "status": "in_progress"}
|
| 384 |
+
|
| 385 |
+
file_state = state["files"][file_key]
|
| 386 |
+
translated = file_state.get("translated_scenarios", {})
|
| 387 |
+
|
| 388 |
+
target_scenarios = scenarios[:2] if dry_run else scenarios
|
| 389 |
+
pending = [s for s in target_scenarios if str(s["id"]) not in translated]
|
| 390 |
+
|
| 391 |
+
print(f"[translate] {file_key}: シナリオ翻訳開始: {len(target_scenarios)}件 (未翻訳: {len(pending)}件)")
|
| 392 |
+
|
| 393 |
+
async def _translate_one(scenario: dict) -> None:
|
| 394 |
+
sid = str(scenario["id"])
|
| 395 |
+
trans_scenario: dict = {"header": scenario["header"], "prompts": {}}
|
| 396 |
+
|
| 397 |
+
for prompt_key, prompt_text in scenario["prompts"].items():
|
| 398 |
+
if not prompt_text.strip():
|
| 399 |
+
trans_scenario["prompts"][prompt_key] = prompt_text
|
| 400 |
+
continue
|
| 401 |
+
try:
|
| 402 |
+
translated_text = await client.translate(prompt_text)
|
| 403 |
+
trans_scenario["prompts"][prompt_key] = translated_text
|
| 404 |
+
except Exception as e:
|
| 405 |
+
print(f"\n[translate] WARNING: {file_key} シナリオ{sid}/{prompt_key} 翻訳失敗: {e}")
|
| 406 |
+
trans_scenario["prompts"][prompt_key] = prompt_text
|
| 407 |
+
|
| 408 |
+
async with lock:
|
| 409 |
+
translated[sid] = trans_scenario
|
| 410 |
+
file_state["translated_scenarios"] = translated
|
| 411 |
+
save_checkpoint(state)
|
| 412 |
+
print(f"[translate] {file_key} シナリオ {sid} 完了")
|
| 413 |
+
|
| 414 |
+
await asyncio.gather(*[_translate_one(s) for s in pending])
|
| 415 |
+
|
| 416 |
+
if not dry_run and len(translated) >= len(scenarios):
|
| 417 |
+
file_state["status"] = "completed"
|
| 418 |
+
save_checkpoint(state)
|
| 419 |
+
|
| 420 |
+
print(f"[translate] {file_key}: シナリオ翻訳完了: {len(translated)}件")
|
| 421 |
+
return {"scenarios": scenarios, "translated": translated}
|
| 422 |
+
|
| 423 |
+
|
| 424 |
+
async def translate_note_file(
|
| 425 |
+
client: VLLMClient,
|
| 426 |
+
filepath: Path,
|
| 427 |
+
file_key: str,
|
| 428 |
+
state: dict,
|
| 429 |
+
dry_run: bool,
|
| 430 |
+
lock: asyncio.Lock,
|
| 431 |
+
) -> dict:
|
| 432 |
+
"""ノート分割ファイルを翻訳する。"""
|
| 433 |
+
notes = parse_scenario_notes(filepath)
|
| 434 |
+
|
| 435 |
+
if file_key not in state["files"]:
|
| 436 |
+
state["files"][file_key] = {"translated_notes": {}, "status": "in_progress"}
|
| 437 |
+
|
| 438 |
+
file_state = state["files"][file_key]
|
| 439 |
+
translated_notes = file_state.get("translated_notes", {})
|
| 440 |
+
|
| 441 |
+
target_notes = dict(list(notes.items())[:2]) if dry_run else notes
|
| 442 |
+
pending = {k: v for k, v in target_notes.items() if k not in translated_notes}
|
| 443 |
+
|
| 444 |
+
print(f"[translate] {file_key}: ノート翻訳開始: {len(target_notes)}件 (未翻訳: {len(pending)}件)")
|
| 445 |
+
|
| 446 |
+
async def _translate_one(key: str, note: str) -> None:
|
| 447 |
+
try:
|
| 448 |
+
translated_text = await client.translate(note)
|
| 449 |
+
except Exception as e:
|
| 450 |
+
print(f"\n[translate] WARNING: {file_key} ノート#{key} 翻訳失敗: {e}")
|
| 451 |
+
translated_text = note
|
| 452 |
+
async with lock:
|
| 453 |
+
translated_notes[key] = translated_text
|
| 454 |
+
file_state["translated_notes"] = translated_notes
|
| 455 |
+
save_checkpoint(state)
|
| 456 |
+
print(f"[translate] {file_key} ノート #{key} 完了")
|
| 457 |
+
|
| 458 |
+
await asyncio.gather(*[_translate_one(k, v) for k, v in pending.items()])
|
| 459 |
+
|
| 460 |
+
if not dry_run and len(translated_notes) >= len(notes):
|
| 461 |
+
file_state["status"] = "completed"
|
| 462 |
+
save_checkpoint(state)
|
| 463 |
+
|
| 464 |
+
print(f"[translate] {file_key}: ノート翻訳完了: {len(translated_notes)}件")
|
| 465 |
+
return {"notes": notes, "translated": translated_notes}
|
| 466 |
+
|
| 467 |
+
|
| 468 |
+
async def translate_single_file(
|
| 469 |
+
client: VLLMClient,
|
| 470 |
+
filepath: Path,
|
| 471 |
+
file_key: str,
|
| 472 |
+
state: dict,
|
| 473 |
+
dry_run: bool,
|
| 474 |
+
lock: asyncio.Lock,
|
| 475 |
+
) -> dict:
|
| 476 |
+
"""単一ファイル全体を翻訳する。"""
|
| 477 |
+
content = parse_single_file(filepath)
|
| 478 |
+
|
| 479 |
+
if file_key not in state["files"]:
|
| 480 |
+
state["files"][file_key] = {"translated": None, "status": "in_progress"}
|
| 481 |
+
|
| 482 |
+
file_state = state["files"][file_key]
|
| 483 |
+
|
| 484 |
+
# 既に翻訳済みならスキップ
|
| 485 |
+
if file_state.get("translated"):
|
| 486 |
+
print(f"[translate] {file_key}: 既に翻訳済み、スキップ")
|
| 487 |
+
return {"original": content, "translated": file_state["translated"]}
|
| 488 |
+
|
| 489 |
+
print(f"[translate] {file_key}: 翻訳開始 ({len(content)} chars)")
|
| 490 |
+
|
| 491 |
+
try:
|
| 492 |
+
translated_text = await client.translate(content)
|
| 493 |
+
except Exception as e:
|
| 494 |
+
print(f"\n[translate] ERROR: {file_key} 翻訳失敗: {e}")
|
| 495 |
+
translated_text = content # 失敗時はオリジナルをそのまま
|
| 496 |
+
|
| 497 |
+
async with lock:
|
| 498 |
+
file_state["translated"] = translated_text
|
| 499 |
+
if not dry_run:
|
| 500 |
+
file_state["status"] = "completed"
|
| 501 |
+
save_checkpoint(state)
|
| 502 |
+
|
| 503 |
+
print(f"[translate] {file_key}: 翻訳完了")
|
| 504 |
+
return {"original": content, "translated": translated_text}
|
| 505 |
+
|
| 506 |
+
|
| 507 |
+
# ── HF アップロード ───────────────────────────────────────────
|
| 508 |
+
|
| 509 |
+
async def upload_to_hf(output_dir: Path) -> None:
|
| 510 |
+
if not HF_TOKEN:
|
| 511 |
+
print("[hub] HF_TOKEN 未設定のためアップロードをスキップ")
|
| 512 |
+
return
|
| 513 |
+
try:
|
| 514 |
+
from huggingface_hub import HfApi
|
| 515 |
+
api = HfApi(token=HF_TOKEN)
|
| 516 |
+
api.create_repo(repo_id=HF_REPO, repo_type="dataset", private=True, exist_ok=True)
|
| 517 |
+
for f in output_dir.glob("*.txt"):
|
| 518 |
+
api.upload_file(
|
| 519 |
+
path_or_fileobj=str(f),
|
| 520 |
+
path_in_repo=f"data/{f.name}",
|
| 521 |
+
repo_id=HF_REPO,
|
| 522 |
+
repo_type="dataset",
|
| 523 |
+
commit_message=f"update: {f.name}",
|
| 524 |
+
)
|
| 525 |
+
print(f"[hub] アップロード完了: {f.name}")
|
| 526 |
+
print(f"[hub] ✅ https://huggingface.co/datasets/{HF_REPO}")
|
| 527 |
+
except Exception as e:
|
| 528 |
+
print(f"[hub] アップロードエラー: {e}")
|
| 529 |
+
print(traceback.format_exc())
|
| 530 |
+
|
| 531 |
+
|
| 532 |
+
# ── メイン処理 ────────────────────────────────────────────────
|
| 533 |
+
|
| 534 |
+
async def process_file(
|
| 535 |
+
client: VLLMClient,
|
| 536 |
+
filename: str,
|
| 537 |
+
file_info: dict,
|
| 538 |
+
data_dir: Path,
|
| 539 |
+
state: dict,
|
| 540 |
+
dry_run: bool,
|
| 541 |
+
lock: asyncio.Lock,
|
| 542 |
+
) -> tuple[str, dict] | None:
|
| 543 |
+
"""単一ファイルの処理を行う。"""
|
| 544 |
+
filepath = data_dir / filename
|
| 545 |
+
|
| 546 |
+
if not filepath.exists():
|
| 547 |
+
print(f"[skip] {filename} が見つかりません: {filepath}")
|
| 548 |
+
return None
|
| 549 |
+
|
| 550 |
+
file_type = file_info["type"]
|
| 551 |
+
description = file_info["description"]
|
| 552 |
+
|
| 553 |
+
print(f"\n{'='*60}")
|
| 554 |
+
print(f"[file] {filename} ({description})")
|
| 555 |
+
print(f"[file] タイプ: {file_type}")
|
| 556 |
+
print(f"{'='*60}")
|
| 557 |
+
|
| 558 |
+
if file_type == "scenario_split":
|
| 559 |
+
result = await translate_scenario_file(client, filepath, filename, state, dry_run, lock)
|
| 560 |
+
return (filename, {"type": "scenario_split", **result})
|
| 561 |
+
|
| 562 |
+
elif file_type == "note_split":
|
| 563 |
+
result = await translate_note_file(client, filepath, filename, state, dry_run, lock)
|
| 564 |
+
return (filename, {"type": "note_split", **result})
|
| 565 |
+
|
| 566 |
+
elif file_type == "single":
|
| 567 |
+
result = await translate_single_file(client, filepath, filename, state, dry_run, lock)
|
| 568 |
+
return (filename, {"type": "single", **result})
|
| 569 |
+
|
| 570 |
+
else:
|
| 571 |
+
print(f"[warning] 不明なファイルタイプ: {file_type}")
|
| 572 |
+
return None
|
| 573 |
+
|
| 574 |
+
|
| 575 |
+
async def main(dry_run: bool, target_file: str | None) -> None:
|
| 576 |
+
start_time = datetime.now(timezone.utc)
|
| 577 |
+
print(f"{'='*60}")
|
| 578 |
+
print(f"EQ-Bench3 日本語化スクリプト v2")
|
| 579 |
+
print(f"開始: [{start_time.isoformat()}]")
|
| 580 |
+
print(f"{'='*60}")
|
| 581 |
+
print(f" EQ-Bench3 ディレクトリ: {EQBENCH_DIR}")
|
| 582 |
+
print(f" 出力ディレクトリ : {OUTPUT_DIR}")
|
| 583 |
+
print(f" 翻訳モデル : {TRANSLATE_MODEL}")
|
| 584 |
+
print(f" dry-run : {dry_run}")
|
| 585 |
+
print(f" 対象ファイル数 : {len(TARGET_FILES)}")
|
| 586 |
+
print()
|
| 587 |
+
|
| 588 |
+
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
|
| 589 |
+
data_dir = EQBENCH_DIR / "data"
|
| 590 |
+
|
| 591 |
+
# ── チェックポイント読み込み ──────────────────────────────
|
| 592 |
+
state = load_checkpoint()
|
| 593 |
+
completed_files = sum(1 for f in state.get("files", {}).values() if f.get("status") == "completed")
|
| 594 |
+
if completed_files > 0:
|
| 595 |
+
print(f"[checkpoint] 再開: {completed_files}/{len(TARGET_FILES)}ファイル処理済み")
|
| 596 |
+
|
| 597 |
+
# ── vLLM 接続 ─────────────────────────────────────────────
|
| 598 |
+
async with VLLMClient() as client:
|
| 599 |
+
await client.wait_for_server()
|
| 600 |
+
|
| 601 |
+
lock = asyncio.Lock()
|
| 602 |
+
results: dict[str, dict] = {}
|
| 603 |
+
|
| 604 |
+
# ── 各ファイルの翻訳実行 ──────────────────────────────
|
| 605 |
+
target_files_to_process = (
|
| 606 |
+
{target_file: TARGET_FILES[target_file]}
|
| 607 |
+
if target_file and target_file in TARGET_FILES
|
| 608 |
+
else TARGET_FILES
|
| 609 |
+
)
|
| 610 |
+
|
| 611 |
+
for filename, file_info in target_files_to_process.items():
|
| 612 |
+
result = await process_file(
|
| 613 |
+
client, filename, file_info, data_dir, state, dry_run, lock
|
| 614 |
+
)
|
| 615 |
+
if result:
|
| 616 |
+
results[result[0]] = result[1]
|
| 617 |
+
|
| 618 |
+
# ── 出力ファイル生成 ─────────────────────────────────────
|
| 619 |
+
print("\n" + "=" * 60)
|
| 620 |
+
print("[output] 出力ファイル生成中...")
|
| 621 |
+
print("=" * 60)
|
| 622 |
+
|
| 623 |
+
for filename, result in results.items():
|
| 624 |
+
output_path = OUTPUT_DIR / filename.replace(".txt", "_ja.txt")
|
| 625 |
+
|
| 626 |
+
if result["type"] == "scenario_split":
|
| 627 |
+
output_content = build_output_prompts(result["scenarios"], result["translated"])
|
| 628 |
+
elif result["type"] == "note_split":
|
| 629 |
+
output_content = build_output_notes(result["notes"], result["translated"])
|
| 630 |
+
elif result["type"] == "single":
|
| 631 |
+
output_content = result["translated"]
|
| 632 |
+
else:
|
| 633 |
+
continue
|
| 634 |
+
|
| 635 |
+
output_path.write_text(output_content, encoding="utf-8")
|
| 636 |
+
print(f"[output] ✅ {output_path.name}")
|
| 637 |
+
|
| 638 |
+
# オリジナルもコピー
|
| 639 |
+
original_path = data_dir / filename
|
| 640 |
+
if original_path.exists():
|
| 641 |
+
import shutil
|
| 642 |
+
shutil.copy(original_path, OUTPUT_DIR / filename.replace(".txt", "_en.txt"))
|
| 643 |
+
|
| 644 |
+
# ── HF アップロード ───────────────────────────────────────
|
| 645 |
+
if not dry_run:
|
| 646 |
+
await upload_to_hf(OUTPUT_DIR)
|
| 647 |
+
|
| 648 |
+
# チェックポイント削除(全完了時のみ)
|
| 649 |
+
if not dry_run and all(
|
| 650 |
+
state.get("files", {}).get(f, {}).get("status") == "completed"
|
| 651 |
+
for f in TARGET_FILES
|
| 652 |
+
):
|
| 653 |
+
if CHECKPOINT_FILE.exists():
|
| 654 |
+
CHECKPOINT_FILE.unlink()
|
| 655 |
+
print("[checkpoint] チェックポイント削除(全ファイル完了)")
|
| 656 |
+
|
| 657 |
+
elapsed = (datetime.now(timezone.utc) - start_time).total_seconds()
|
| 658 |
+
print()
|
| 659 |
+
print("=" * 60)
|
| 660 |
+
print(f"翻訳完了 (所要時間: {elapsed/60:.1f}分)")
|
| 661 |
+
print(f"出力先: {OUTPUT_DIR}/")
|
| 662 |
+
print("=" * 60)
|
| 663 |
+
|
| 664 |
+
|
| 665 |
+
if __name__ == "__main__":
|
| 666 |
+
parser = argparse.ArgumentParser(description="EQ-Bench3 日本語化スクリプト v2")
|
| 667 |
+
parser.add_argument("--dry-run", action="store_true", help="最初の2シナリオのみ処理して動作確認")
|
| 668 |
+
parser.add_argument(
|
| 669 |
+
"--target-file",
|
| 670 |
+
choices=list(TARGET_FILES.keys()),
|
| 671 |
+
default=None,
|
| 672 |
+
help="翻訳対象ファイルを指定(デフォルト: 全ファイル)",
|
| 673 |
+
)
|
| 674 |
+
parser.add_argument(
|
| 675 |
+
"--list-files",
|
| 676 |
+
action="store_true",
|
| 677 |
+
help="対象ファイル一覧を表示して終了",
|
| 678 |
+
)
|
| 679 |
+
args = parser.parse_args()
|
| 680 |
+
|
| 681 |
+
if args.list_files:
|
| 682 |
+
print("翻訳対象ファイル一覧:")
|
| 683 |
+
print("-" * 60)
|
| 684 |
+
for filename, info in TARGET_FILES.items():
|
| 685 |
+
print(f" {filename}")
|
| 686 |
+
print(f" タイプ: {info['type']}")
|
| 687 |
+
print(f" 説明 : {info['description']}")
|
| 688 |
+
sys.exit(0)
|
| 689 |
+
|
| 690 |
+
if not os.environ.get("HF_TOKEN"):
|
| 691 |
+
print("[WARNING] HF_TOKEN が未設定です。HF アップロードはスキップされます。")
|
| 692 |
+
|
| 693 |
+
asyncio.run(main(dry_run=args.dry_run, target_file=args.target_file))
|