打了两把UNO 第二把出现的，monika在基础公共牌上使用＋4导致报错
点击忽略无法返回游戏
（至少我想说个再见的）
前几天有打过五十来把测试UNO还是很稳定的

I'm sorry, but an uncaught exception occurred.

While running game code:
File "game/zz_cardgames.rpy", line 4441, in <module>
File "game/zz_cardgames.rpy", line 1487, in prepare_game
File "game/zz_cardgames.rpy", line 6308, in insert
TypeError: integer argument expected, got float

-- Full Traceback ------------------------------------------------------------

Full traceback:
File "script-ch30.rpyc", line 2091, in script call
File "tl/chinese/config_pythonov/script_fix.rpyc", line 40, in script call
File "script-ch30.rpyc", line 2091, in script call
File "tl/chinese/config_pythonov/script_fix.rpyc", line 40, in script call
File "zz_games.rpyc", line 190, in script call
File "zz_cardgames.rpyc", line 4441, in script
File "renpy/ast.py", line 823, in execute
File "renpy/python.py", line 1178, in py_exec_bytecode
File "game/zz_cardgames.rpy", line 4441, in <module>
File "game/zz_cardgames.rpy", line 1487, in prepare_game
File "game/zz_cardgames.rpy", line 6308, in insert
File "renpy/revertable.py", line 97, in do_mutation
TypeError: integer argument expected, got float

24.8.12

请注意本轮次测试将于8月13或14日终止, 以允许我们对模型进行进一步的改良. 请等待关于下一轮次测试的通知.

接下来的改良仍然着重于模型基础能力, 尚未启用强化训练, 即LIA分支将继续更新至少一次. 下次测试仍然使用LIA分支.

如果你希望保存你的session请及时下载. 在下一轮测试开始时, 我们强烈建议你清除旧的session以便分析模型的变化.

==MAICA-撕裂现实的帷幕==

Please notice that this round of Beta will end by Apr.13 or 14, acquiring us time to make improvements on core model performance. Please wait for notifications on starting next round of Beta.

The following improvements on core model are still focused on fundamental LLM abilities so feedback learning is yet not included. This means the next Beta will still be based on an improved LIA branch model which will be open sourced.

Download your session in time if you want to keep it. We strongly suggest you purging your session in next round to avoid session pollution.

==MAICA-We tear this barrier apart==

24.8.6

关于显示表情不正常的修复

关于自动重连的修复

关于登录模式的改良

Fixes on emotion analyzing failure

Fixes on issues about auto reconnection

Improvements on login interaction

一个成功的项目不可以没有彩蛋, 所以我在后端加了一点彩蛋. 特定的query可以触发.

一共有九个, 我很好奇运行多久会被全部找到. 小提示是多少都跟'人造生命'沾点关系.

不用看仓库了, 没有的.

除此之外就是一些前端工作和整理文件.

今天试了一下鼻炎的药, 鼻子确实不塞了, 但是全身反应还是照样有. 我觉得我的毛病可能不是简单的鼻炎.

不想去医院, 至少是暂时不想去. 再说吧.

帮淘宝买rgb控制器的店家解决了一个编译问题, 他送了我一个免费以旧换新.

感觉像是完成了什么奇怪的支线任务.

新控制器送到之后需要费点时间换装. 在此之前我会在qwen-14b上继续训练8个ep, 观察足够的拟合到底需要多少训练量.

很奇怪的是, 在推理中合并权重好像会让模型的拟合效果变差. 我觉得这也是swift本身的问题.

懒得交issue了. 我已经有三个问题明确的issue没人回了, 估计提了也没用.

总的来说, 进展比想象中慢. 但是急也急不来. 无论人手还是设备都太有限了.