Replies: 3 comments 1 reply
-
Hi - please share your tabby setup: e.g what kind of model are you using for code completion. Besides, can you provide actual cases that you feel could use improvements? e.g in form of screenshort / recording. |
Beta Was this translation helpful? Give feedback.
-
Hi,thanks for your answer, I use code-llama-7b, and the context uses 40 lines. I have currently modified the code on the plugin side and reduced some calls that I think are unnecessary (not necessarily correct, but I think it is more reasonable). The rules are as follows:
I tried to modify the code in this way, but unfortunately, the content displayed every time I entered would shake, the actual number of calls dropped significantly, and indicators such as the adoption rate also increased significantly (I think it is closer to real user feedback and can better reflect usability😂). |
Beta Was this translation helpful? Give feedback.
-
Compared to mainstream products, Tabby's code completion is indeed more frequent. |
Beta Was this translation helpful? Give feedback.
-
In the code generation scenario, there are often some invalid generation, such as the deletion of some indentation characters. I am not sure if this is what you expected. If this is an optimization point, does the tabby official team have plans to optimize this scenario?
Please reply with a 👍 if you want this feature.
Beta Was this translation helpful? Give feedback.
All reactions