You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After several months of experimentation, I've made an interesting discovery: standard semantic models (like Claude, OpenAI ChatGpt, Qwen, CodeLlama-Instruct or StarCoder2-Instruct) can actually perform better at code completion than FIM models when given the right system prompt and user message structure.
Based on these findings, I've recently added additional settings to customize both:
System prompts for providing context and instruction to the model
User message templates for structuring the code completion requests
The default settings are optimized based on my testing, but I'd love to hear your experiences and suggestions:
Have you experimented with different prompts and message structures?
What combinations work best for your specific use cases?
Any ideas for improving the defaults?
Thoughts on the balance between FIM and instruction models for code completion?
You can find these settings in:
Settings -> Qode Assist -> Code Completion
"Use System Prompt" and system prompt text
"Use User Template" and message template
Let's work together to optimize QodeAssist's code completion capabilities!
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
-
After several months of experimentation, I've made an interesting discovery: standard semantic models (like Claude, OpenAI ChatGpt, Qwen, CodeLlama-Instruct or StarCoder2-Instruct) can actually perform better at code completion than FIM models when given the right system prompt and user message structure.
Based on these findings, I've recently added additional settings to customize both:
The default settings are optimized based on my testing, but I'd love to hear your experiences and suggestions:
You can find these settings in:
Settings -> Qode Assist -> Code Completion
Let's work together to optimize QodeAssist's code completion capabilities!
Beta Was this translation helpful? Give feedback.
All reactions