fix: initialize auto suggestions providers (closes #571) #576
Conversation
|
Did the fix locally and it worked :) |
There was a problem hiding this comment.
I was thinking to refactor it to make Provider as two separate instance, instead of having a singleton like this.
This actually only works for copilot, since if both auto_suggestions_provider and provider is set (let say openai and claude respectively), then generations won't work at all.
|
It seems to be working. 🤔 Autocompletion with OpenAI provider (also, the settings I'm currently using):
{
body = {
max_tokens = 4096,
model = "gpt-4o-mini",
stream = true,
temperature = 0
},
url = "https://api.openai.com/v1/chat/completions"
}Also a chat with Claude:
{
body = {
max_tokens = 4096,
model = "claude-3-haiku-20240307",
stream = true,
system = { {
cache_control = {
type = "ephemeral"
},
text = "You are an excellent programming expert.\n",
type = "text"
} },
temperature = 0
},
url = "https://api.anthropic.com/v1/messages"
} |
aarnphm
changed the title
| E.setup({ provider = provider }) | ||
|
|
||
| if auto_suggestions_provider and auto_suggestions_provider ~= provider then | ||
| E.setup({ provider = auto_suggestions_provider }) |
There was a problem hiding this comment.
The design patterns related to the provider are a bit too complex. Why is E.setup called twice here? @aarnphm
There was a problem hiding this comment.
Ye I'm intending to clean this up a bit. It has to do with we are lazily loading the env for each providers.
Fixes #571