层1/T2: max_tokens 放大至 16384（防截断，上限不计费）；并发提到 16；进度行加速度与预计剩余时间

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

ars_opd/teacher.py

@@ -16,6 +16,7 @@ from __future__ import annotations
 import json
 import os
 import re
+import time
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from pathlib import Path
 
@@ -140,6 +141,7 @@ def generate_completions(
 
     failures: list[tuple[str, str]] = []
     finished = 0
+    start = time.monotonic()
     # 写盘收口在主线程（as_completed 消费端），工作线程只跑网络请求——
     # 多线程同写一个文件句柄会交错损坏 JSONL
     with open(path, "a", encoding="utf-8") as f:
@@ -155,7 +157,15 @@ def generate_completions(
                 finally:
                     finished += 1
                     if finished % 20 == 0 or finished == len(todo):
-                        print(f"[teacher] {finished}/{len(todo)} 完成", flush=True)
+                        elapsed = time.monotonic() - start
+                        rate = finished / elapsed * 60  # 条/分
+                        eta = (len(todo) - finished) / rate if rate > 0 else 0
+                        print(
+                            f"[teacher] {finished}/{len(todo)} 完成 | "
+                            f"{rate:.1f} 条/分 | 已用 {elapsed / 60:.1f} 分 | "
+                            f"预计剩余 {eta:.0f} 分",
+                            flush=True,
+                        )
                 record = {
                     "key": key,
                     "completion": completion,