Bug Description

https://docs.rs/openai/1.0.0-alpha.9/openai/completions/struct.CompletionBuilder.html#method.presence_penalty

presence penalty and frequency penalty are i8 but the API requests a float between -2.0 and 2.0.

I have a slight suspicion that it might be applying some scaling (i.e. dividing by 10.0 before sending) but I can't find any proof on the code of this.

Anyway, whatever it is, it needs to be either corrected or clarified. If it's really sending an integer, the problem is that the only options here are -2,-1,0,1,2; if it's actually sending a fractional number, the documentation doesn't explain.

There might be other methods affected by this issue in this repo. I'd recommend checking if it applies. For example frequency_penalty in this struct seems to have the same problem. It's possible this also is on other APIs too.

Terminal Output

No response

Reproduction Steps

https://docs.rs/openai/1.0.0-alpha.9/openai/completions/struct.CompletionBuilder.html#method.presence_penalty

Try to send an API request with presence penalty of 0.32. Won't compile..

Expected Behavior

https://docs.rs/openai/1.0.0-alpha.9/openai/completions/struct.CompletionBuilder.html#method.presence_penalty

API accepting a float, fixed number or docs updated to explain how decimal numbers can be sent to the API.

Workaround

No response

Library Version

1.0.0-alpha.9

rustc Version

rustc 1.68.2 (9eb3afe9e 2023-03-27)

cargo Version

cargo 1.68.2 (6feb7c9cf 2023-03-26)

Platform

Linux

Problem

I prefer not using OpenAI's official API endpoint https://api.openai.com/v1/, but some mirror site like chatanywhere/GPT_API_free.

Solution

Seems that this feature is easy to implement, just make BASE_URL in lib.rs customizable -- export a function like set_key; for example set_base_url.

Alternatives

No response

Problem

ive been using this api for a while and one of the most infuriating things is that i cant copy or partialeq some structs/enums.
as an example, the ChatCompletionBuilder doesnt derive Clone, and the enum ChatCompletionMessageRole doesnt derive partialeq.

Solution

add core derives to all structs/enums

Alternatives

No response

Problem

We're currently using Mutexes but it makes sense to use a RwLock since it allows multiple readers at the same time.

Solution

Use RwLock

Let me know your thoughts

Problem

OpenAI allows a single ChatCompletionMessage to contain multiple function calls, as is presented here:
https://platform.openai.com/docs/guides/function-calling/parallel-function-calling.
However, the current implementation only allows for a single Function Call to be present per response.

from source:

pub function_call: Option<ChatCompletionFunctionCall>

(https://github.com/rellfy/openai/blob/master/src/chat.rs#L60)

Solution

Given my lack of knowledge about the API in use, changing

pub function_call: Option<ChatCompletionFunctionCall>

to

pub function_call: Vec<ChatCompletionFunctionCall>

should resolve the problem, though I am not familiar with the function calling API enough to be sure that that would work

Alternatives

No response

Problem

ChatGPT just released function calling API that would be very valuable to have in this crate

https://platform.openai.com/docs/guides/gpt/function-calling

Solution

Add function call API

Alternatives

No response

In the OpenAI docs, they describe the function call flow here
https://platform.openai.com/docs/guides/gpt/function-calling

TL;DR is that after the model tells you a function to call, and you call it, then you add a message to the chain with the role of function with the name and result of the function.

So basically need to be able to create this

ChatCompletionMessage {
    role: ChatCompletionMessageRole::Function,
    name: Some("update_ability_score".into()),
    content: Some("Your constitution has dropped by 2 points".into()),
    function_call: None, //docs say use None, but this seems odd
});

I'm able to use the rest of the function calling, works great! Seem like the role just never got added here

Is there a better way to do this that I'm missing or should I put up a PR to add this?

Problem

Implement new assistants API released by OpenAI

Solution

Implement new assistants API released by OpenAI

Alternatives

No response