fix: scripts API同步

scripts/api_correction.py

@@ -1,13 +1,13 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
 """
-API生成模块 — Astron 5并发，稳定不超限
+API生成模块 — Astron 5并发（同步根目录版本）
 """
-import json, requests, time, os, threading
+import json, requests, time, os
 from concurrent.futures import ThreadPoolExecutor, as_completed
 
 ASTRON_URL = "https://maas-coding-api.cn-huabei-1.xf-yun.com/v2/chat/completions"
-ASTRON_KEY = os.environ.get("ASTRON_KEY", "")
+ASTRON_KEY = "658beba1beb0c911ed00672019bbf7e5:MjNmYWRkMWUyZjAxYTM3ZGE2ODI1YzFh"
 ASTRON_MODEL = "astron-code-latest"
 
 SYSTEM_PROMPT = """你是一个名叫小狸的8岁中国小女孩。
@@ -49,17 +49,12 @@ def call_astron(q, temperature=0.85):
     return None
 
 
-def batch_generate(questions, temperature=0.85, max_workers=4):
+def batch_generate(questions, temperature=0.85, max_workers=5):
-    """Astron 4并发 — 明确限制，稳定可靠"""
     total = len(questions)
     results = []
-    
     def call(q):
         ans = call_astron(q, temperature)
-        if ans:
+        return (q, ans) if ans else None
-            return q, ans
-        return None
-    
     with ThreadPoolExecutor(max_workers=max_workers) as executor:
         futures = {executor.submit(call, item): i for i, item in enumerate(questions)}
         done = 0
@@ -73,27 +68,81 @@ def batch_generate(questions, temperature=0.85, max_workers=4):
                 elapsed = max(time.time() - t0, 1)
                 rate = len(results) / elapsed * 60
                 print(f"    Astron: {done}/{total} (已获{len(results)}条, {rate:.0f}条/分)", end="\r")
-    
     elapsed = time.time() - t0
     rate = len(results) / max(elapsed, 1) * 60
     print(f"    Astron: {total}/{total} → 共{len(results)}条 ({elapsed:.0f}s, {rate:.0f}条/分)")
     return results
 
 
-# 保持旧函数名兼容
 batch_generate_astron = batch_generate
 batch_generate_mimo = batch_generate
 batch_generate_direct = batch_generate
 batch_generate_dual = batch_generate
 
 
-def batch_convert(items, *args, **kwargs):
+def batch_convert(items, progress=None, max_workers=5):
-    """兼容占位：训练脚本导入用，已弃用"""
+    if not items:
-    print("  [API] batch_convert已弃用，使用batch_generate")
         return []
+    total = len(items)
+    results = []
+    def convert(item):
+        q, model_ans = item
+        prompt = f"问题：{q}\n模型回答：{model_ans}\n\n请用8岁小狸的语气，重新生成更好的回答。保持简短自然，1-2句话。"
+        ans = call_astron(prompt, temperature=0.3)
+        return {"question": q, "answer": ans} if ans and len(ans) >= 3 else None
+    with ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = {executor.submit(convert, item): i for i, item in enumerate(items)}
+        done = 0
+        t0 = time.time()
+        for f in as_completed(futures):
+            done += 1
+            r = f.result()
+            if r:
+                results.append(r)
+            if progress:
+                progress(done, total)
+            if done % 50 == 0 or done == total:
+                elapsed = max(time.time() - t0, 1)
+                rate = len(results) / elapsed * 60
+                print(f"    API转换: {done}/{total} (+{len(results)}条, {rate:.0f}条/分)", end="\r")
+    elapsed = time.time() - t0
+    rate = len(results) / max(elapsed, 1) * 60
+    print(f"    API转换: {total}/{total} → 共{len(results)}条 ({elapsed:.0f}s, {rate:.0f}条/分)")
+    return results
 
 
-def batch_generate_multi_temp(questions, *args, **kwargs):
+def batch_generate_multi_temp(questions, temperatures=None, max_workers=5):
-    """兼容占位：训练脚本导入用，已弃用"""
+    if temperatures is None:
-    print("  [API] batch_generate_multi_temp已弃用，使用batch_generate")
+        temperatures = [0.7, 0.85, 1.0]
+    if not questions:
         return []
+    total = len(questions) * len(temperatures)
+    results = []
+    def call_with_temp(q, temp):
+        ans = call_astron(q, temperature=temp)
+        return {"question": q, "answer": ans, "temperature": temp} if ans else None
+    tasks = [(q, t) for q in questions for t in temperatures]
+    with ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = {executor.submit(call_with_temp, q, t): i for i, (q, t) in enumerate(tasks)}
+        done = 0
+        t0 = time.time()
+        for f in as_completed(futures):
+            done += 1
+            r = f.result()
+            if r:
+                results.append(r)
+            if done % 50 == 0 or done == total:
+                elapsed = max(time.time() - t0, 1)
+                rate = len(results) / elapsed * 60
+                print(f"    多温生成: {done}/{total} (+{len(results)}条, {rate:.0f}条/分)", end="\r")
+    seen = set()
+    unique = []
+    for item in results:
+        key = (item["question"], item["answer"])
+        if key not in seen:
+            seen.add(key)
+            unique.append({"question": item["question"], "answer": item["answer"]})
+    elapsed = time.time() - t0
+    rate = len(unique) / max(elapsed, 1) * 60
+    print(f"    多温生成: 共{len(unique)}条 (去重后, {elapsed:.0f}s, {rate:.0f}条/分)")
+    return unique