Completions

Request

Endpoint

Lines 66 to 71 in 6ee5bc5

    
           let request = self 
        
               .client 
        
               .post("https://api.openai.com/v1/completions") 
        
               .header("Content-Type", "application/json") 
        
               .header("Authorization", format!("Bearer {}", self.api_key)) 
        
               .json(&json_data);

Instead of https://api.openai.com/v1/completions, the endpoint URL would be in the format https://{resource_name}.openai.azure.com/openai/deployments/{deployment_id}/completions?api-version=<api_version> with 2022-12-01 being the latest API version so we must make the endpoint URL for completions configurable.

Headers

Instead of an Authorization header in the format Authorization: Bearer <api_key>, we would have the header api-key: <api_key>

Parameters

In the Azure OpenAI Service API, a completions request does not support a model parameter. Instead, this is represented by the {deployment_id} URL parameter.

Response

The OpenAI API response object has a usage property but the Azure OpenAI Service API does not.

The model property in the example response displays "ada". The models available for the Azure OpenAI Service are listed in Model Summary table and region availability. These are deployed then targeted in the {deployment_id} URL parameter for the completions endpoint. I'm not sure all the models supported by the Azure OpenAI Service are listed in

gptcommit/src/openai.rs

Lines 91 to 101 in 6ee5bc5

    
           pub(crate) fn get_prompt_token_limit_for_model(&self) -> usize { 
        
               match self.model.as_str() { 
        
                   "text-davinci-003" => 4097, 
        
                   "text-curie-001" => 2048, 
        
                   "text-babbage-001" => 2048, 
        
                   "text-ada-001" => 2048, 
        
                   "code-davinci-002" => 8000, 
        
                   "code-cushman-001" => 2048, 
        
                   _ => 4097, 
        
               } 
        
           }

At least not if "ada" is a valid model response value.

	let request = self
	.client
	.post("https://api.openai.com/v1/completions")
	.header("Content-Type", "application/json")
	.header("Authorization", format!("Bearer {}", self.api_key))
	.json(&json_data);

	pub(crate) fn get_prompt_token_limit_for_model(&self) -> usize {
	match self.model.as_str() {
	"text-davinci-003" => 4097,
	"text-curie-001" => 2048,
	"text-babbage-001" => 2048,
	"text-ada-001" => 2048,
	"code-davinci-002" => 8000,
	"code-cushman-001" => 2048,
	_ => 4097,
	}
	}

Support Azure OpenAI service #18

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions