v1.3.2

.gitignore

@@ -12,5 +12,5 @@ venv
 .venv
 nuitka/
 dist/
-.pdm.toml
+.pdm-python
 aunly_bbot/static/test.js

aunly_bbot/cli/api.py

@@ -9,7 +9,7 @@
 from playwright.async_api._generated import BrowserContext
 from graiax.playwright.installer import install_playwright
 
-from ..utils.dynamic_shot import screenshot
+from ..utils.browser_shot import screenshot
 from ..utils.fonts_provider import get_font
 from ..utils.detect_package import is_package
 

aunly_bbot/function/command/content_resolve.py

@@ -123,7 +123,7 @@ async def openai_summarization():
                                     logger.warning(f"视频 {aid} 总结失败：{ai_summary.raw}")
                                     return
 
-                            if summarise.lower().startswith("none"):
+                            if "no meaning" in summarise.lower():
                                 nonlocal chatgpt_thinks
                                 chatgpt_thinks = False
                                 raise AbortError("ChatGPT 认为这些字幕没有意义")
@@ -140,7 +140,7 @@ async def openai_summarization():
                                     quote=info_message.source,
                                 )
                         except AbortError as e:
-                            logger.info(f"视频 {aid} 总结被终止：{e}")
+                            logger.warning(f"视频 {aid} 总结被终止：{e}")
                         except Exception:
                             capture_exception()
                             logger.exception(f"视频 {aid} 总结出错")
@@ -228,7 +228,7 @@ async def openai_summarization():
                                 group, MessageChain(Image(data_bytes=image)), quote=source
                             )
                     except AbortError as e:
-                        logger.info(f"专栏 {column_id} 总结被终止：{e}")
+                        logger.warning(f"专栏 {column_id} 总结被终止：{e}")
                     except Exception:
                         capture_exception()
                         logger.exception(f"专栏 {column_id} 总结出错")

aunly_bbot/function/pusher/dynamic.py

@@ -264,7 +264,7 @@ async def push(app: Ariadne, dyn: DynamicItem):
                         MessageChain(msg),
                     )
                     insert_dyn_push_to_group(dynid, data.group)
-                    await asyncio.sleep(2)
+                    await asyncio.sleep(5)
                 except UnknownTarget:
                     logger.warning(
                         f"[BiliBili推送] {dynid} | {up_name}({up_id}) 推送失败，找不到该群 {data.group}，正在取消订阅"

aunly_bbot/function/pusher/live.py

@@ -137,7 +137,6 @@ async def main(app: Ariadne):
                                     int(data.group),
                                     MessageChain(msg),
                                 )
-                                await asyncio.sleep(1)
                             except UnknownTarget:
                                 delete = await delete_group(data.group)
                                 logger.info(
@@ -146,6 +145,7 @@ async def main(app: Ariadne):
                             except AccountMuted:
                                 group = f"{group.name}（{group.id}）" if group else data.group
                                 logger.warning(f"[BiliBili推送] 推送失败，账号在 {group} 被禁言")
+                            await asyncio.sleep(5)
 
                     insert_live_push(
                         up_id, True, len(get_sub_by_uid(up_id)), title, area_parent, area
@@ -204,7 +204,7 @@ async def main(app: Ariadne):
                             except Exception:  # noqa
                                 capture_exception()
                                 logger.exception("[BiliBili推送] 推送失败，未知错误")
-                            await asyncio.sleep(1)
+                            await asyncio.sleep(5)
                     insert_live_push(up_id, False, len(get_sub_by_uid(up_id)))
             else:
                 logger.warning(f"[BiliBili推送] 未找到订阅 UP {up_name}（{up_id}）的群，正在退订！")

aunly_bbot/utils/content_summarise.py

@@ -2,15 +2,15 @@
 
 from loguru import logger
 
-from .openai import openai_req, get_small_size_transcripts, get_user_prompt, get_simple_prompt
+from .openai import openai_req, get_small_size_transcripts, get_user_prompt, get_full_prompt
 
 
 async def subtitle_summarise(sub: list[str], title: str):
     """请求字幕总结"""
     small_size_transcripts = get_small_size_transcripts(sub)
     prompt = get_user_prompt(title, small_size_transcripts)
     logger.debug(prompt)
-    return await openai_req(get_simple_prompt(prompt))
+    return await openai_req(get_full_prompt(prompt, system=True))
 
 
 async def column_summarise(cv_title: str, cv_text: str):
@@ -19,4 +19,4 @@ async def column_summarise(cv_title: str, cv_text: str):
     small_size_transcripts = get_small_size_transcripts(sentences)
     prompt = get_user_prompt(cv_title, small_size_transcripts)
     logger.debug(prompt)
-    return await openai_req(get_simple_prompt(prompt))
+    return await openai_req(get_full_prompt(prompt, system=True))

aunly_bbot/utils/openai.py

@@ -25,26 +25,12 @@
 def get_user_prompt(title: str, transcript: str) -> str:
     title = title.replace("\n", " ").strip() if title else ""
     transcript = transcript.replace("\n", " ").strip() if transcript else ""
-    language = "Chinese"
-    prompt = (
-        "Your output should use the following template:\n## Summary\n## Highlights\n"
-        "- [Emoji] Bulletpoint\n\n"
-        "Your task is to summarise the video I have given you in up to 2 to 6 concise bullet points, "
-        "starting with a short highlight, each bullet point is at least 15 words. "
-        "Choose an appropriate emoji for each bullet point. "
-        f"Use the video above: {{Title}} {{Transcript}}."
-        "If you think the content in the transcript is meaningless or nonsensical, "
-        "you can choose to skip summarization and simply output 'none'."
-        f"\n\nReply in {language} Language."
-    )
-    return f'Title: "{title}"\nTranscript: "{transcript}"\n\nInstructions: {prompt}'
+    return f'Title: "{title}"\nTranscript: "{transcript}"'
 
 
-def count_tokens(contents: list[str]):
+def count_tokens(prompts: list[dict[str, str]]):
     """根据内容计算 token 数"""
 
-    content = get_simple_prompt(get_user_prompt("", " ".join(contents)))
-
     if BotConfig.Bilibili.openai_model == "gpt-3.5-turbo-0301":
         tokens_per_message = 4
         tokens_per_name = -1
@@ -55,7 +41,7 @@ def count_tokens(contents: list[str]):
         raise ValueError(f"Unknown model name {BotConfig.Bilibili.openai_model}")
 
     num_tokens = 0
-    for message in content:
+    for message in prompts:
         num_tokens += tokens_per_message
         for key, value in message.items():
             num_tokens += len(tiktoken_enc.encode(value))
@@ -67,13 +53,35 @@ def count_tokens(contents: list[str]):
 
 def get_small_size_transcripts(text_data: list[str], token_limit: int = LIMIT_COUNT):
     unique_texts = list(OrderedDict.fromkeys(text_data))
-    while count_tokens(unique_texts) > token_limit:
+    while (
+        count_tokens(
+            get_full_prompt(get_user_prompt("", " ".join(unique_texts)), system=True)
+        )
+        > token_limit
+    ):
         unique_texts.pop(random.randint(0, len(unique_texts) - 1))
     return " ".join(unique_texts)
 
 
-def get_simple_prompt(prompt: str):
-    return [{"role": "user", "content": prompt}]
+def get_full_prompt(prompt: str, system: bool = False):
+    plist: list[dict[str, str]] = []
+    if system:
+        language = "Chinese"
+        sys_prompt = (
+            "Your output should use the following template:\n## Summary\n## Highlights\n"
+            "- [Emoji] Bulletpoint\n\n"
+            "Your task is to summarise the video I have given you in up to 2 to 6 concise bullet points, "
+            "starting with a short highlight, each bullet point is at least 15 words. "
+            "Choose an appropriate emoji for each bullet point. "
+            f"Use the video above: {{Title}} {{Transcript}}."
+            "If you think that the content in the transcript is meaningless, "
+            "Or if there is very little content that cannot be well summarized, "
+            "then you can simply output the three words 'no meaning' Remember not to output anything else."
+            f"\n\nReply in {language} Language."
+        )
+        plist.append({"role": "system", "content": sys_prompt})
+    plist.append({"role": "user", "content": prompt})
+    return plist
 
 
 async def openai_req(
@@ -101,5 +109,6 @@ async def openai_req(
         )
         if req.status_code != 200:
             return AISummary(error=True, message=req.text, raw=req.json())
+        logger.info(f"[OpenAI] Response: {req.json()['choices'][0]['message']['content']}")
         logger.info(f"[OpenAI] Response token 实际: {req.json()['usage']}")
         return AISummary(summary=req.json()["choices"][0]["message"]["content"], raw=req.json())