- Introduce `AudioTranscriptionsFile` method in the OpenAI package to transcribe audio directly from a file path, improving memory efficiency. - Implement new test cases for `AudioTranscriptionsFile`, including scenarios for language options and error handling for non-existent files. - Update the process package to support the new transcription method, ensuring compatibility with existing functionality. - Enhance test coverage for audio transcription processes, validating expected outputs and error conditions.
203 lines
5 KiB
Go
203 lines
5 KiB
Go
package openai
|
|
|
|
import (
|
|
"context"
|
|
|
|
"github.com/yaoapp/gou/http"
|
|
"github.com/yaoapp/gou/process"
|
|
"github.com/yaoapp/gou/runtime/v8/bridge"
|
|
"github.com/yaoapp/kun/exception"
|
|
"github.com/yaoapp/kun/log"
|
|
)
|
|
|
|
func init() {
|
|
process.RegisterGroup("openai", map[string]process.Handler{
|
|
"tiktoken": ProcessTiktoken,
|
|
"embeddings": ProcessEmbeddings,
|
|
"chat.completions": ProcessChatCompletions,
|
|
"audio.transcriptions": ProcessAudioTranscriptions,
|
|
"audio.transcriptionsfile": ProcessAudioTranscriptionsFile,
|
|
})
|
|
}
|
|
|
|
// ProcessTiktoken openai.Tiktoken
|
|
func ProcessTiktoken(process *process.Process) interface{} {
|
|
process.ValidateArgNums(2)
|
|
model := process.ArgsString(0)
|
|
input := process.ArgsString(1)
|
|
nums, err := Tiktoken(model, input)
|
|
if err != nil {
|
|
exception.New("Tiktoken error: %s", 400, err).Throw()
|
|
}
|
|
return nums
|
|
}
|
|
|
|
// ProcessEmbeddings openai.Embeddings
|
|
func ProcessEmbeddings(process *process.Process) interface{} {
|
|
process.ValidateArgNums(2)
|
|
model := process.ArgsString(0)
|
|
input := process.Args[1]
|
|
user := ""
|
|
if process.NumOfArgs() > 2 {
|
|
user = process.ArgsString(2)
|
|
}
|
|
|
|
ai, err := New(model)
|
|
if err != nil {
|
|
exception.New("ChatCompletions error: %s", 400, err).Throw()
|
|
}
|
|
|
|
res, ex := ai.Embeddings(input, user)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
}
|
|
|
|
// ProcessAudioTranscriptions openai.audio.Transcriptions
|
|
func ProcessAudioTranscriptions(process *process.Process) interface{} {
|
|
process.ValidateArgNums(2)
|
|
model := process.ArgsString(0)
|
|
dataBase64 := process.ArgsString(1)
|
|
options := map[string]interface{}{}
|
|
if process.NumOfArgs() > 2 {
|
|
if opts, ok := process.Args[2].(map[string]interface{}); ok {
|
|
options = opts
|
|
}
|
|
}
|
|
|
|
ai, err := New(model)
|
|
if err != nil {
|
|
exception.New("ChatCompletions error: %s", 400, err).Throw()
|
|
}
|
|
|
|
res, ex := ai.AudioTranscriptions(dataBase64, options)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
}
|
|
|
|
// ProcessAudioTranscriptionsFile openai.audio.TranscriptionsFile
|
|
// Transcribe audio from an OS file path (streaming upload, no base64 overhead).
|
|
// This is the recommended way to call Whisper from TS scripts, consistent with
|
|
// office.Parse / ffmpeg.* handler style.
|
|
//
|
|
// Args:
|
|
// - connector string - AI connector name (e.g. "openai.whisper-1")
|
|
// - filePath string - OS absolute path to the audio file
|
|
// - options map - Optional: { language, model, ... }
|
|
//
|
|
// Returns: map[string]interface{} - Transcription result (e.g. {"text": "..."})
|
|
//
|
|
// Usage:
|
|
//
|
|
// var result = Process("openai.audio.transcriptionsfile", "openai.whisper-1", "/abs/path/to/audio.mp3", {"language": "en"})
|
|
func ProcessAudioTranscriptionsFile(process *process.Process) interface{} {
|
|
process.ValidateArgNums(2)
|
|
connector := process.ArgsString(0)
|
|
filePath := process.ArgsString(1)
|
|
|
|
options := map[string]interface{}{}
|
|
if process.NumOfArgs() > 2 {
|
|
if opts, ok := process.Args[2].(map[string]interface{}); ok {
|
|
options = opts
|
|
}
|
|
}
|
|
|
|
ai, err := New(connector)
|
|
if err != nil {
|
|
exception.New("AudioTranscriptionsFile error: %s", 400, err).Throw()
|
|
}
|
|
|
|
res, ex := ai.AudioTranscriptionsFile(filePath, options)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
}
|
|
|
|
// ProcessChatCompletions openai.chat.Completions
|
|
func ProcessChatCompletions(process *process.Process) interface{} {
|
|
|
|
process.ValidateArgNums(2)
|
|
ctx, cancel := context.WithCancel(context.Background())
|
|
defer cancel()
|
|
|
|
model := process.ArgsString(0)
|
|
messages := []map[string]interface{}{}
|
|
intput := process.ArgsArray(1)
|
|
for idx, v := range intput {
|
|
message, ok := v.(map[string]interface{})
|
|
if !ok {
|
|
exception.New("ChatCompletions input must be array of map, index %d", 400, idx).Throw()
|
|
}
|
|
messages = append(messages, message)
|
|
}
|
|
|
|
ai, err := New(model)
|
|
if err != nil {
|
|
exception.New("ChatCompletions error: %s", 400, err).Throw()
|
|
}
|
|
|
|
options := map[string]interface{}{}
|
|
if process.NumOfArgs() > 2 {
|
|
if opts, ok := process.Args[2].(map[string]interface{}); ok {
|
|
options = opts
|
|
}
|
|
}
|
|
|
|
if process.NumOfArgs() == 3 {
|
|
data, ex := ai.ChatCompletionsWith(ctx, messages, options, nil)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return data
|
|
}
|
|
|
|
if process.NumOfArgs() == 4 {
|
|
|
|
switch cb := process.Args[3].(type) {
|
|
case func(data []byte) int:
|
|
res, ex := ai.ChatCompletionsWith(ctx, messages, options, cb)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
|
|
case bridge.FunctionT:
|
|
res, ex := ai.ChatCompletionsWith(ctx, messages, options, func(data []byte) int {
|
|
|
|
v, err := cb.Call(string(data))
|
|
if err != nil {
|
|
log.Error("Call callback function error: %s", err.Error())
|
|
return http.HandlerReturnError
|
|
}
|
|
|
|
ret, ok := v.(int)
|
|
if !ok {
|
|
log.Error("Callback function must return int")
|
|
return http.HandlerReturnError
|
|
}
|
|
|
|
return ret
|
|
})
|
|
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
|
|
default:
|
|
exception.New("ChatCompletions error: invalid callback arguments", 400).Throw()
|
|
return nil
|
|
}
|
|
}
|
|
|
|
res, ex := ai.ChatCompletionsWith(ctx, messages, options, nil)
|
|
if ex != nil {
|
|
ex.Throw()
|
|
}
|
|
return res
|
|
|
|
}
|