autoglm-9b-multilingual model output is unreasonable

[ahuizxc]

tested on android world dataset,
the autoglm-9b works fine but due to it's a chinese language model, some task perform bad
Multilingual version is totally unusable！

[K1XE]

same here, very interested in autoglm-9b. do you have any preliminary results on the Android World benchmark? thanks in advance.

[ahuizxc]

same here, very interested in autoglm-9b. do you have any preliminary results on the Android World benchmark? thanks in advance.

autoglm-9b跑benchmark要改一下解析输出格式的代码，MobileRL的代码解析不了模型的输出，并且由于是中文版，成功率略低，经常launch中文app，需要改一下英文的app字典做适配，有些简单任务才能成功，困难一点的成不了，不如mai-ui-8b

[xuyifan-0731]

Using our inference framework's android_world-reasoning-test-nolaunch-coordinate, we can correctly infer autoglm-9b-multilingual. There are two different system prompts in autoglm-9b-multilingual, one for Chinese and one for English; you need to choose the corresponding one for it to output correctly.