aient 1.2.38__tar.gz → 1.2.40__tar.gz

{aient-1.2.38 → aient-1.2.40}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aient
-Version: 1.2.38
+Version: 1.2.40
 Summary: Aient: The Awakening of Agent.
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{aient-1.2.38 → aient-1.2.40}/aient/models/chatgpt.py

@@ -12,7 +12,7 @@ from typing import Union, Optional, Callable
 from .base import BaseLLM
 from ..plugins.registry import registry
 from ..plugins import PLUGINS, get_tools_result_async, function_call_list, update_tools_config
-from ..utils.scripts import safe_get, async_generator_to_sync, parse_function_xml, parse_continuous_json, convert_functions_to_xml, remove_xml_tags_and_content
+from ..utils.scripts import safe_get, async_generator_to_sync, parse_function_xml, parse_continuous_json, convert_functions_to_xml, remove_xml_tags_and_content, find_most_frequent_phrase
 from ..core.request import prepare_request_payload
 from ..core.response import fetch_response_stream, fetch_response
 from ..architext.architext import Messages, SystemMessage, UserMessage, AssistantMessage, ToolCalls, ToolResults, Texts, RoleMessage, Images, Files
@@ -81,6 +81,14 @@ class TaskComplete(Exception):
         super().__init__(f"Task completed with message: {message}")
 
 
+class RepetitiveResponseError(Exception):
+    """Custom exception for detecting repetitive and meaningless generated strings."""
+    def __init__(self, message, phrase, count):
+        super().__init__(message)
+        self.phrase = phrase
+        self.count = count
+
+
 class chatgpt(BaseLLM):
     """
     Official ChatGPT API
@@ -438,6 +446,13 @@ class chatgpt(BaseLLM):
 
         if not full_response.strip() and not need_function_call:
             raise EmptyResponseError("Response is empty")
+        most_frequent_phrase, most_frequent_phrase_count = find_most_frequent_phrase(full_response)
+        if most_frequent_phrase_count > 100:
+            raise RepetitiveResponseError(
+                f"Detected repetitive and meaningless content. The phrase '{most_frequent_phrase}' appeared {most_frequent_phrase_count} times.",
+                most_frequent_phrase,
+                most_frequent_phrase_count
+            )
 
         if self.print_log:
             self.logger.info(f"total_tokens: {total_tokens}")
@@ -660,17 +675,6 @@ class chatgpt(BaseLLM):
             self.add_to_conversation(full_response, response_role, convo_id=convo_id, total_tokens=total_tokens, pass_history=pass_history)
             self.function_calls_counter = {}
 
-            # 清理翻译引擎相关的历史记录
-            if pass_history <= 2 and len(self.conversation[convo_id]) >= 2 \
-            and (
-                "You are a translation engine" in self.conversation[convo_id][-2]["content"] \
-                or "You are a translation engine" in safe_get(self.conversation, convo_id, -2, "content", 0, "text", default="") \
-                or "你是一位精通简体中文的专业翻译" in self.conversation[convo_id][-2]["content"] \
-                or "你是一位精通简体中文的专业翻译" in safe_get(self.conversation, convo_id, -2, "content", 0, "text", default="")
-            ):
-                self.conversation[convo_id].pop(-1)
-                self.conversation[convo_id].pop(-1)
-
     async def _ask_stream_handler(
         self,
         prompt: list,
@@ -811,6 +815,9 @@ class chatgpt(BaseLLM):
             except EmptyResponseError as e:
                 self.logger.warning(f"{e}, retrying...")
                 continue
+            except RepetitiveResponseError as e:
+                self.logger.warning(f"{e}, retrying...")
+                continue
             except TaskComplete as e:
                 raise
             except ModelNotFoundError as e:

{aient-1.2.38 → aient-1.2.40}/aient/utils/scripts.py

@@ -3,10 +3,65 @@ import re
 import json
 import fnmatch
 import requests
+import collections
 import urllib.parse
 
 from ..core.utils import get_image_message
 
+def find_most_frequent_phrase(s, min_len=4, max_phrase_len=20):
+    """
+    查找字符串中出现次数最多的短语（单词序列）。
+    此版本经过性能优化，并增加了最大短语长度限制。
+
+    Args:
+        s: 输入字符串。
+        min_len: 短语的最小字符长度。
+        max_phrase_len: 要搜索的最大短语长度（以单词为单位）。
+
+    Returns:
+        一个元组 (most_frequent_phrase, count)，其中
+        most_frequent_phrase 是出现次数最多的短语，
+        count 是它的出现次数。
+        如果没有找到符合条件的重复短语，则返回 ("", 0)。
+    """
+    # start_time = time.time()
+    if not s or len(s) < min_len:
+        return "", 0
+
+    words = [word for word in re.split(r'[\s\n]+', s) if word]
+    if not words:
+        return "", 0
+    n = len(words)
+
+    phrase_counts = collections.defaultdict(int)
+
+    # 确定要检查的实际最大长度
+    effective_max_len = min(n // 2, max_phrase_len)
+
+    # 优化的核心：直接在单词列表上生成并统计所有可能的短语（n-grams）
+    for length in range(1, effective_max_len + 1):
+        for i in range(n - length + 1):
+            phrase_tuple = tuple(words[i:i + length])
+            phrase_counts[phrase_tuple] += 1
+    # 筛选出重复次数大于1且满足最小长度要求的短语
+    best_phrase = ""
+    max_count = 0
+
+    for phrase_tuple, count in phrase_counts.items():
+        if count > 1:
+            phrase = " ".join(phrase_tuple)
+            if len(phrase) >= min_len:
+                if count > max_count:
+                    max_count = count
+                    best_phrase = phrase
+                elif count == max_count and len(phrase) > len(best_phrase):
+                    best_phrase = phrase
+
+    if max_count > 0:
+        return best_phrase, max_count
+    else:
+        return "", 0
+
 def get_doc_from_url(url):
     filename = urllib.parse.unquote(url.split("/")[-1])
     response = requests.get(url, stream=True)

{aient-1.2.38 → aient-1.2.40}/aient.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aient
-Version: 1.2.38
+Version: 1.2.40
 Summary: Aient: The Awakening of Agent.
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{aient-1.2.38 → aient-1.2.40}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "aient"
-version = "1.2.38"
+version = "1.2.40"
 description = "Aient: The Awakening of Agent."
 readme = "README.md"
 requires-python = ">=3.11"