From 1e3ebdcdf03df94f0acec68cf6f9ba5256dce143 Mon Sep 17 00:00:00 2001 From: Cindy-Master <2606440373@qq.com> Date: Wed, 12 Mar 2025 17:55:38 +0800 Subject: [PATCH] =?UTF-8?q?=E5=B0=9D=E8=AF=95=E5=87=8F=E5=B0=91=E4=B8=8D?= =?UTF-8?q?=E5=90=8C=E7=BE=A4=E5=86=85=E7=9A=84=E5=90=8C=E4=B9=89=E8=AF=8D?= =?UTF-8?q?=E6=B7=B7=E6=B7=86=20=E4=BB=A5=E5=8F=8A=E5=85=81=E8=AE=B8?= =?UTF-8?q?=E7=94=A8=E6=88=B7=E8=AE=BE=E7=BD=AE=E6=8C=87=E5=AE=9A=E7=BE=A4?= =?UTF-8?q?=E7=BB=84=E8=AE=B0=E5=BF=86=E7=A7=81=E6=9C=89?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 用户设置指定群组记忆私有会导致记忆的语料严重减少 未设置群组记忆私有的记忆会开放对于所有群聊 因此要不就修改其他记忆配置 增强学习/记忆能力 或者增加单一群组的信息流数量 或者增加公开记忆的信息流数量 记忆检索时会按以下优先级返回记忆： 当前群组的记忆（如果群聊属于某个群组） 当前群聊的记忆（如果不属于任何群组） 公共记忆（无群组/群聊标识的记忆） 其他非私有群组的记忆 5. 跨群记忆连接 同一群组内的主题使用较强连接（相似度×10） 跨群组的相似主题使用较弱连接（相似度×5） 相同群组/群聊的相似主题会获得20%的相似度加成 --- src/plugins/chat/bot.py | 2 +- src/plugins/chat/emoji_manager.py | 10 ++++------ src/plugins/chat/utils_image.py | 19 +++++++------------ src/plugins/models/utils_model.py | 15 ++++++--------- 4 files changed, 18 insertions(+), 28 deletions(-) diff --git a/src/plugins/chat/bot.py b/src/plugins/chat/bot.py index 1db38477..f179e8ef 100644 --- a/src/plugins/chat/bot.py +++ b/src/plugins/chat/bot.py @@ -135,7 +135,7 @@ class ChatBot: for pattern in global_config.ban_msgs_regex: if re.search(pattern, message.raw_message): logger.info( - f"[{chat.group_info.group_name if chat.group_info.group_id else '私聊'}]{userinfo.user_nickname}:{message.raw_message}" + f"[{chat.group_info.group_name if chat.group_info.group_id else '私聊'}]{message.user_nickname}:{message.raw_message}" ) logger.info(f"[正则表达式过滤]消息匹配到{pattern}，filtered") return diff --git a/src/plugins/chat/emoji_manager.py b/src/plugins/chat/emoji_manager.py index 1c8a0769..9532db4f 100644 --- a/src/plugins/chat/emoji_manager.py +++ b/src/plugins/chat/emoji_manager.py @@ -6,8 +6,6 @@ import random import time import traceback from typing import Optional, Tuple -from PIL import Image -import io from loguru import logger from nonebot import get_driver @@ -194,11 +192,11 @@ class EmojiManager: logger.error(f"获取标签失败: {str(e)}") return None - async def _check_emoji(self, image_base64: str, image_format: str) -> str: + async def _check_emoji(self, image_base64: str) -> str: try: prompt = f'这是一个表情包，请回答这个表情包是否满足\"{global_config.EMOJI_CHECK_PROMPT}\"的要求，是则回答是，否则回答否，不要出现任何其他内容' - content, _ = await self.vlm.generate_response_for_image(prompt, image_base64, image_format) + content, _ = await self.vlm.generate_response_for_image(prompt, image_base64) logger.debug(f"输出描述: {content}") return content @@ -239,7 +237,7 @@ class EmojiManager: image_bytes = base64.b64decode(image_base64) image_hash = hashlib.md5(image_bytes).hexdigest() - image_format = Image.open(io.BytesIO(image_bytes)).format.lower() + # 检查是否已经注册过 existing_emoji = self.db['emoji'].find_one({'filename': filename}) description = None @@ -280,7 +278,7 @@ class EmojiManager: if global_config.EMOJI_CHECK: - check = await self._check_emoji(image_base64, image_format) + check = await self._check_emoji(image_base64) if '是' not in check: os.remove(image_path) logger.info(f"描述: {description}") diff --git a/src/plugins/chat/utils_image.py b/src/plugins/chat/utils_image.py index 94014b5b..fb242887 100644 --- a/src/plugins/chat/utils_image.py +++ b/src/plugins/chat/utils_image.py @@ -4,8 +4,6 @@ import time import aiohttp import hashlib from typing import Optional, Union -from PIL import Image -import io from loguru import logger from nonebot import get_driver @@ -121,7 +119,6 @@ class ImageManager: # 计算哈希值 image_hash = hashlib.md5(image_bytes).hexdigest() - image_format = Image.open(io.BytesIO(image_bytes)).format.lower() # 查重 existing = self.db.images.find_one({'hash': image_hash}) @@ -130,7 +127,7 @@ class ImageManager: # 生成文件名和路径 timestamp = int(time.time()) - filename = f"{timestamp}_{image_hash[:8]}.{image_format}" + filename = f"{timestamp}_{image_hash[:8]}.jpg" file_path = os.path.join(self.IMAGE_DIR, filename) # 保存文件 @@ -241,8 +238,7 @@ class ImageManager: # 计算图片哈希 image_bytes = base64.b64decode(image_base64) image_hash = hashlib.md5(image_bytes).hexdigest() - image_format = Image.open(io.BytesIO(image_bytes)).format.lower() - + # 查询缓存的描述 cached_description = self._get_description_from_db(image_hash, 'emoji') if cached_description: @@ -251,13 +247,13 @@ class ImageManager: # 调用AI获取描述 prompt = "这是一个表情包，使用中文简洁的描述一下表情包的内容和表情包所表达的情感" - description, _ = await self._llm.generate_response_for_image(prompt, image_base64, image_format) + description, _ = await self._llm.generate_response_for_image(prompt, image_base64) # 根据配置决定是否保存图片 if global_config.EMOJI_SAVE: # 生成文件名和路径 timestamp = int(time.time()) - filename = f"{timestamp}_{image_hash[:8]}.{image_format}" + filename = f"{timestamp}_{image_hash[:8]}.jpg" file_path = os.path.join(self.IMAGE_DIR, 'emoji',filename) try: @@ -297,8 +293,7 @@ class ImageManager: # 计算图片哈希 image_bytes = base64.b64decode(image_base64) image_hash = hashlib.md5(image_bytes).hexdigest() - image_format = Image.open(io.BytesIO(image_bytes)).format.lower() - + # 查询缓存的描述 cached_description = self._get_description_from_db(image_hash, 'image') if cached_description: @@ -307,7 +302,7 @@ class ImageManager: # 调用AI获取描述 prompt = "请用中文描述这张图片的内容。如果有文字，请把文字都描述出来。并尝试猜测这个图片的含义。最多200个字。" - description, _ = await self._llm.generate_response_for_image(prompt, image_base64, image_format) + description, _ = await self._llm.generate_response_for_image(prompt, image_base64) print(f"描述是{description}") @@ -319,7 +314,7 @@ class ImageManager: if global_config.EMOJI_SAVE: # 生成文件名和路径 timestamp = int(time.time()) - filename = f"{timestamp}_{image_hash[:8]}.{image_format}" + filename = f"{timestamp}_{image_hash[:8]}.jpg" file_path = os.path.join(self.IMAGE_DIR,'image', filename) try: diff --git a/src/plugins/models/utils_model.py b/src/plugins/models/utils_model.py index aa07bb55..5335e3d6 100644 --- a/src/plugins/models/utils_model.py +++ b/src/plugins/models/utils_model.py @@ -104,7 +104,6 @@ class LLM_request: endpoint: str, prompt: str = None, image_base64: str = None, - image_format: str = None, payload: dict = None, retry_policy: dict = None, response_handler: callable = None, @@ -116,7 +115,6 @@ class LLM_request: endpoint: API端点路径 (如 "chat/completions") prompt: prompt文本 image_base64: 图片的base64编码 - image_format: 图片格式 payload: 请求体数据 retry_policy: 自定义重试策略 response_handler: 自定义响应处理器 @@ -153,7 +151,7 @@ class LLM_request: # 构建请求体 if image_base64: - payload = await self._build_payload(prompt, image_base64, image_format) + payload = await self._build_payload(prompt, image_base64) elif payload is None: payload = await self._build_payload(prompt) @@ -174,7 +172,7 @@ class LLM_request: if response.status == 413: logger.warning("请求体过大，尝试压缩...") image_base64 = compress_base64_image_by_scale(image_base64) - payload = await self._build_payload(prompt, image_base64, image_format) + payload = await self._build_payload(prompt, image_base64) elif response.status in [500, 503]: logger.error(f"错误码: {response.status} - {error_code_mapping.get(response.status)}") raise RuntimeError("服务器负载过高，模型恢复失败QAQ") @@ -296,7 +294,7 @@ class LLM_request: new_params["max_completion_tokens"] = new_params.pop("max_tokens") return new_params - async def _build_payload(self, prompt: str, image_base64: str = None, image_format: str = None) -> dict: + async def _build_payload(self, prompt: str, image_base64: str = None) -> dict: """构建请求体""" # 复制一份参数，避免直接修改 self.params params_copy = await self._transform_parameters(self.params) @@ -308,7 +306,7 @@ class LLM_request: "role": "user", "content": [ {"type": "text", "text": prompt}, - {"type": "image_url", "image_url": {"url": f"data:image/{image_format.lower()};base64,{image_base64}"}} + {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_base64}"}} ] } ], @@ -393,14 +391,13 @@ class LLM_request: ) return content, reasoning_content - async def generate_response_for_image(self, prompt: str, image_base64: str, image_format: str) -> Tuple[str, str]: + async def generate_response_for_image(self, prompt: str, image_base64: str) -> Tuple[str, str]: """根据输入的提示和图片生成模型的异步响应""" content, reasoning_content = await self._execute_request( endpoint="/chat/completions", prompt=prompt, - image_base64=image_base64, - image_format=image_format + image_base64=image_base64 ) return content, reasoning_content