Add support for O3-mini (#930)
* Add support for O3-mini - Add support for the o3 mini set of models, including tests that match the constraints in OpenAI's API docs (https://platform.openai.com/docs/models#o3-mini). * Deprecate and refactor - Deprecate `ErrO1BetaLimitationsLogprobs` and `ErrO1BetaLimitationsOther` - Implement `validationRequestForReasoningModels`, which works on both o1 & o3, and has per-model-type restrictions on functionality (eg, o3 class are allowed function calls and system messages, o1 isn't) * Move reasoning validation to `reasoning_validator.go` - Add a `NewReasoningValidator` which exposes a `Validate()` method for a given request - Also adds a test for chat streams * Final nits
This commit is contained in:
@@ -2,24 +2,9 @@ package openai
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"net/http"
|
||||
)
|
||||
|
||||
var (
|
||||
ErrO1MaxTokensDeprecated = errors.New("this model is not supported MaxTokens, please use MaxCompletionTokens") //nolint:lll
|
||||
ErrCompletionUnsupportedModel = errors.New("this model is not supported with this method, please use CreateChatCompletion client method instead") //nolint:lll
|
||||
ErrCompletionStreamNotSupported = errors.New("streaming is not supported with this method, please use CreateCompletionStream") //nolint:lll
|
||||
ErrCompletionRequestPromptTypeNotSupported = errors.New("the type of CompletionRequest.Prompt only supports string and []string") //nolint:lll
|
||||
)
|
||||
|
||||
var (
|
||||
ErrO1BetaLimitationsMessageTypes = errors.New("this model has beta-limitations, user and assistant messages only, system messages are not supported") //nolint:lll
|
||||
ErrO1BetaLimitationsTools = errors.New("this model has beta-limitations, tools, function calling, and response format parameters are not supported") //nolint:lll
|
||||
ErrO1BetaLimitationsLogprobs = errors.New("this model has beta-limitations, logprobs not supported") //nolint:lll
|
||||
ErrO1BetaLimitationsOther = errors.New("this model has beta-limitations, temperature, top_p and n are fixed at 1, while presence_penalty and frequency_penalty are fixed at 0") //nolint:lll
|
||||
)
|
||||
|
||||
// GPT3 Defines the models provided by OpenAI to use when generating
|
||||
// completions from OpenAI.
|
||||
// GPT3 Models are designed for text-based tasks. For code-specific
|
||||
@@ -31,6 +16,8 @@ const (
|
||||
O1Preview20240912 = "o1-preview-2024-09-12"
|
||||
O1 = "o1"
|
||||
O120241217 = "o1-2024-12-17"
|
||||
O3Mini = "o3-mini"
|
||||
O3Mini20250131 = "o3-mini-2025-01-31"
|
||||
GPT432K0613 = "gpt-4-32k-0613"
|
||||
GPT432K0314 = "gpt-4-32k-0314"
|
||||
GPT432K = "gpt-4-32k"
|
||||
@@ -96,21 +83,14 @@ const (
|
||||
CodexCodeDavinci001 = "code-davinci-001"
|
||||
)
|
||||
|
||||
// O1SeriesModels List of new Series of OpenAI models.
|
||||
// Some old api attributes not supported.
|
||||
var O1SeriesModels = map[string]struct{}{
|
||||
O1Mini: {},
|
||||
O1Mini20240912: {},
|
||||
O1Preview: {},
|
||||
O1Preview20240912: {},
|
||||
}
|
||||
|
||||
var disabledModelsForEndpoints = map[string]map[string]bool{
|
||||
"/completions": {
|
||||
O1Mini: true,
|
||||
O1Mini20240912: true,
|
||||
O1Preview: true,
|
||||
O1Preview20240912: true,
|
||||
O3Mini: true,
|
||||
O3Mini20250131: true,
|
||||
GPT3Dot5Turbo: true,
|
||||
GPT3Dot5Turbo0301: true,
|
||||
GPT3Dot5Turbo0613: true,
|
||||
@@ -183,64 +163,6 @@ func checkPromptType(prompt any) bool {
|
||||
return true // all items in the slice are string, so it is []string
|
||||
}
|
||||
|
||||
var unsupportedToolsForO1Models = map[ToolType]struct{}{
|
||||
ToolTypeFunction: {},
|
||||
}
|
||||
|
||||
var availableMessageRoleForO1Models = map[string]struct{}{
|
||||
ChatMessageRoleUser: {},
|
||||
ChatMessageRoleAssistant: {},
|
||||
}
|
||||
|
||||
// validateRequestForO1Models checks for deprecated fields of OpenAI models.
|
||||
func validateRequestForO1Models(request ChatCompletionRequest) error {
|
||||
if _, found := O1SeriesModels[request.Model]; !found {
|
||||
return nil
|
||||
}
|
||||
|
||||
if request.MaxTokens > 0 {
|
||||
return ErrO1MaxTokensDeprecated
|
||||
}
|
||||
|
||||
// Logprobs: not supported.
|
||||
if request.LogProbs {
|
||||
return ErrO1BetaLimitationsLogprobs
|
||||
}
|
||||
|
||||
// Message types: user and assistant messages only, system messages are not supported.
|
||||
for _, m := range request.Messages {
|
||||
if _, found := availableMessageRoleForO1Models[m.Role]; !found {
|
||||
return ErrO1BetaLimitationsMessageTypes
|
||||
}
|
||||
}
|
||||
|
||||
// Tools: tools, function calling, and response format parameters are not supported
|
||||
for _, t := range request.Tools {
|
||||
if _, found := unsupportedToolsForO1Models[t.Type]; found {
|
||||
return ErrO1BetaLimitationsTools
|
||||
}
|
||||
}
|
||||
|
||||
// Other: temperature, top_p and n are fixed at 1, while presence_penalty and frequency_penalty are fixed at 0.
|
||||
if request.Temperature > 0 && request.Temperature != 1 {
|
||||
return ErrO1BetaLimitationsOther
|
||||
}
|
||||
if request.TopP > 0 && request.TopP != 1 {
|
||||
return ErrO1BetaLimitationsOther
|
||||
}
|
||||
if request.N > 0 && request.N != 1 {
|
||||
return ErrO1BetaLimitationsOther
|
||||
}
|
||||
if request.PresencePenalty > 0 {
|
||||
return ErrO1BetaLimitationsOther
|
||||
}
|
||||
if request.FrequencyPenalty > 0 {
|
||||
return ErrO1BetaLimitationsOther
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// CompletionRequest represents a request structure for completion API.
|
||||
type CompletionRequest struct {
|
||||
Model string `json:"model"`
|
||||
|
||||
Reference in New Issue
Block a user