Revert "[Refactor]: Core/API Split" (#1550)

Revert "[Refactor]: Core/API Split (#1506)"

This reverts commit ab7b4d5ee9.

api/openai/chat.go

@@ -0,0 +1,399 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/pkg/grammar"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+	"github.com/valyala/fasthttp"
+)
+
+func ChatEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	emptyMessage := ""
+	id := uuid.New().String()
+	created := int(time.Now().Unix())
+
+	process := func(s string, req *schema.OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan schema.OpenAIResponse) {
+		initialMessage := schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: []schema.Choice{{Delta: &schema.Message{Role: "assistant", Content: &emptyMessage}}},
+			Object:  "chat.completion.chunk",
+		}
+		responses <- initialMessage
+
+		ComputeChoices(req, s, config, o, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
+			resp := schema.OpenAIResponse{
+				ID:      id,
+				Created: created,
+				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []schema.Choice{{Delta: &schema.Message{Content: &s}, Index: 0}},
+				Object:  "chat.completion.chunk",
+				Usage: schema.OpenAIUsage{
+					PromptTokens:     usage.Prompt,
+					CompletionTokens: usage.Completion,
+					TotalTokens:      usage.Prompt + usage.Completion,
+				},
+			}
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+	return func(c *fiber.Ctx) error {
+		processFunctions := false
+		funcs := grammar.Functions{}
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+		log.Debug().Msgf("Configuration read: %+v", config)
+
+		// Allow the user to set custom actions via config file
+		// to be "embedded" in each model
+		noActionName := "answer"
+		noActionDescription := "use this action to answer without performing any action"
+
+		if config.FunctionsConfig.NoActionFunctionName != "" {
+			noActionName = config.FunctionsConfig.NoActionFunctionName
+		}
+		if config.FunctionsConfig.NoActionDescriptionName != "" {
+			noActionDescription = config.FunctionsConfig.NoActionDescriptionName
+		}
+
+		if input.ResponseFormat.Type == "json_object" {
+			input.Grammar = grammar.JSONBNF
+		}
+
+		// process functions if we have any defined or if we have a function call string
+		if len(input.Functions) > 0 && config.ShouldUseFunctions() {
+			log.Debug().Msgf("Response needs to process functions")
+
+			processFunctions = true
+
+			noActionGrammar := grammar.Function{
+				Name:        noActionName,
+				Description: noActionDescription,
+				Parameters: map[string]interface{}{
+					"properties": map[string]interface{}{
+						"message": map[string]interface{}{
+							"type":        "string",
+							"description": "The message to reply the user with",
+						}},
+				},
+			}
+
+			// Append the no action function
+			funcs = append(funcs, input.Functions...)
+			if !config.FunctionsConfig.DisableNoAction {
+				funcs = append(funcs, noActionGrammar)
+			}
+
+			// Force picking one of the functions by the request
+			if config.FunctionToCall() != "" {
+				funcs = funcs.Select(config.FunctionToCall())
+			}
+
+			// Update input grammar
+			jsStruct := funcs.ToJSONStructure()
+			config.Grammar = jsStruct.Grammar("")
+		} else if input.JSONFunctionGrammarObject != nil {
+			config.Grammar = input.JSONFunctionGrammarObject.Grammar("")
+		}
+
+		// functions are not supported in stream mode (yet?)
+		toStream := input.Stream && !processFunctions
+
+		log.Debug().Msgf("Parameters: %+v", config)
+
+		var predInput string
+
+		suppressConfigSystemPrompt := false
+		mess := []string{}
+		for messageIndex, i := range input.Messages {
+			var content string
+			role := i.Role
+
+			// if function call, we might want to customize the role so we can display better that the "assistant called a json action"
+			// if an "assistant_function_call" role is defined, we use it, otherwise we use the role that is passed by in the request
+			if i.FunctionCall != nil && i.Role == "assistant" {
+				roleFn := "assistant_function_call"
+				r := config.Roles[roleFn]
+				if r != "" {
+					role = roleFn
+				}
+			}
+			r := config.Roles[role]
+			contentExists := i.Content != nil && i.StringContent != ""
+			// First attempt to populate content via a chat message specific template
+			if config.TemplateConfig.ChatMessage != "" {
+				chatMessageData := model.ChatMessageTemplateData{
+					SystemPrompt: config.SystemPrompt,
+					Role:         r,
+					RoleName:     role,
+					Content:      i.StringContent,
+					MessageIndex: messageIndex,
+				}
+				templatedChatMessage, err := o.Loader.EvaluateTemplateForChatMessage(config.TemplateConfig.ChatMessage, chatMessageData)
+				if err != nil {
+					log.Error().Msgf("error processing message %+v using template \"%s\": %v. Skipping!", chatMessageData, config.TemplateConfig.ChatMessage, err)
+				} else {
+					if templatedChatMessage == "" {
+						log.Warn().Msgf("template \"%s\" produced blank output for %+v. Skipping!", config.TemplateConfig.ChatMessage, chatMessageData)
+						continue // TODO: This continue is here intentionally to skip over the line `mess = append(mess, content)` below, and to prevent the sprintf
+					}
+					log.Debug().Msgf("templated message for chat: %s", templatedChatMessage)
+					content = templatedChatMessage
+				}
+			}
+			// If this model doesn't have such a template, or if that template fails to return a value, template at the message level.
+			if content == "" {
+				if r != "" {
+					if contentExists {
+						content = fmt.Sprint(r, i.StringContent)
+					}
+					if i.FunctionCall != nil {
+						j, err := json.Marshal(i.FunctionCall)
+						if err == nil {
+							if contentExists {
+								content += "\n" + fmt.Sprint(r, " ", string(j))
+							} else {
+								content = fmt.Sprint(r, " ", string(j))
+							}
+						}
+					}
+				} else {
+					if contentExists {
+						content = fmt.Sprint(i.StringContent)
+					}
+					if i.FunctionCall != nil {
+						j, err := json.Marshal(i.FunctionCall)
+						if err == nil {
+							if contentExists {
+								content += "\n" + string(j)
+							} else {
+								content = string(j)
+							}
+						}
+					}
+				}
+				// Special Handling: System. We care if it was printed at all, not the r branch, so check seperately
+				if contentExists && role == "system" {
+					suppressConfigSystemPrompt = true
+				}
+			}
+
+			mess = append(mess, content)
+		}
+
+		predInput = strings.Join(mess, "\n")
+		log.Debug().Msgf("Prompt (before templating): %s", predInput)
+
+		if toStream {
+			log.Debug().Msgf("Stream request received")
+			c.Context().SetContentType("text/event-stream")
+			//c.Response().Header.SetContentType(fiber.MIMETextHTMLCharsetUTF8)
+			//	c.Set("Content-Type", "text/event-stream")
+			c.Set("Cache-Control", "no-cache")
+			c.Set("Connection", "keep-alive")
+			c.Set("Transfer-Encoding", "chunked")
+		}
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Chat != "" && !processFunctions {
+			templateFile = config.TemplateConfig.Chat
+		}
+
+		if config.TemplateConfig.Functions != "" && processFunctions {
+			templateFile = config.TemplateConfig.Functions
+		}
+
+		if templateFile != "" {
+			templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.ChatPromptTemplate, templateFile, model.PromptTemplateData{
+				SystemPrompt:         config.SystemPrompt,
+				SuppressSystemPrompt: suppressConfigSystemPrompt,
+				Input:                predInput,
+				Functions:            funcs,
+			})
+			if err == nil {
+				predInput = templatedInput
+				log.Debug().Msgf("Template found, input modified to: %s", predInput)
+			} else {
+				log.Debug().Msgf("Template failed loading: %s", err.Error())
+			}
+		}
+
+		log.Debug().Msgf("Prompt (after templating): %s", predInput)
+		if processFunctions {
+			log.Debug().Msgf("Grammar: %+v", config.Grammar)
+		}
+
+		if toStream {
+			responses := make(chan schema.OpenAIResponse)
+
+			go process(predInput, input, config, o.Loader, responses)
+
+			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
+
+				usage := &schema.OpenAIUsage{}
+
+				for ev := range responses {
+					usage = &ev.Usage // Copy a pointer to the latest usage chunk so that the stop message can reference it
+					var buf bytes.Buffer
+					enc := json.NewEncoder(&buf)
+					enc.Encode(ev)
+					log.Debug().Msgf("Sending chunk: %s", buf.String())
+					_, err := fmt.Fprintf(w, "data: %v\n", buf.String())
+					if err != nil {
+						log.Debug().Msgf("Sending chunk failed: %v", err)
+						input.Cancel()
+						break
+					}
+					w.Flush()
+				}
+
+				resp := &schema.OpenAIResponse{
+					ID:      id,
+					Created: created,
+					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []schema.Choice{
+						{
+							FinishReason: "stop",
+							Index:        0,
+							Delta:        &schema.Message{Content: &emptyMessage},
+						}},
+					Object: "chat.completion.chunk",
+					Usage:  *usage,
+				}
+				respData, _ := json.Marshal(resp)
+
+				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
+				w.WriteString("data: [DONE]\n\n")
+				w.Flush()
+			}))
+			return nil
+		}
+
+		result, tokenUsage, err := ComputeChoices(input, predInput, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+			if processFunctions {
+				// As we have to change the result before processing, we can't stream the answer (yet?)
+				ss := map[string]interface{}{}
+				// This prevent newlines to break JSON parsing for clients
+				s = utils.EscapeNewLines(s)
+				json.Unmarshal([]byte(s), &ss)
+				log.Debug().Msgf("Function return: %s %+v", s, ss)
+
+				// The grammar defines the function name as "function", while OpenAI returns "name"
+				func_name := ss["function"]
+				// Similarly, while here arguments is a map[string]interface{}, OpenAI actually want a stringified object
+				args := ss["arguments"] // arguments needs to be a string, but we return an object from the grammar result (TODO: fix)
+				d, _ := json.Marshal(args)
+
+				ss["arguments"] = string(d)
+				ss["name"] = func_name
+
+				// if do nothing, reply with a message
+				if func_name == noActionName {
+					log.Debug().Msgf("nothing to do, computing a reply")
+
+					// If there is a message that the LLM already sends as part of the JSON reply, use it
+					arguments := map[string]interface{}{}
+					json.Unmarshal([]byte(d), &arguments)
+					m, exists := arguments["message"]
+					if exists {
+						switch message := m.(type) {
+						case string:
+							if message != "" {
+								log.Debug().Msgf("Reply received from LLM: %s", message)
+								message = backend.Finetune(*config, predInput, message)
+								log.Debug().Msgf("Reply received from LLM(finetuned): %s", message)
+
+								*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &message}})
+								return
+							}
+						}
+					}
+
+					log.Debug().Msgf("No action received from LLM, without a message, computing a reply")
+					// Otherwise ask the LLM to understand the JSON output and the context, and return a message
+					// Note: This costs (in term of CPU) another computation
+					config.Grammar = ""
+					images := []string{}
+					for _, m := range input.Messages {
+						images = append(images, m.StringImages...)
+					}
+					predFunc, err := backend.ModelInference(input.Context, predInput, images, o.Loader, *config, o, nil)
+					if err != nil {
+						log.Error().Msgf("inference error: %s", err.Error())
+						return
+					}
+
+					prediction, err := predFunc()
+					if err != nil {
+						log.Error().Msgf("inference error: %s", err.Error())
+						return
+					}
+
+					fineTunedResponse := backend.Finetune(*config, predInput, prediction.Response)
+					*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &fineTunedResponse}})
+				} else {
+					// otherwise reply with the function call
+					*c = append(*c, schema.Choice{
+						FinishReason: "function_call",
+						Message:      &schema.Message{Role: "assistant", FunctionCall: ss},
+					})
+				}
+
+				return
+			}
+			*c = append(*c, schema.Choice{FinishReason: "stop", Index: 0, Message: &schema.Message{Role: "assistant", Content: &s}})
+		}, nil)
+		if err != nil {
+			return err
+		}
+
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "chat.completion",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     tokenUsage.Prompt,
+				CompletionTokens: tokenUsage.Completion,
+				TotalTokens:      tokenUsage.Prompt + tokenUsage.Completion,
+			},
+		}
+		respData, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", respData)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/completion.go

@@ -0,0 +1,199 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/pkg/grammar"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+	"github.com/valyala/fasthttp"
+)
+
+// https://platform.openai.com/docs/api-reference/completions
+func CompletionEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	id := uuid.New().String()
+	created := int(time.Now().Unix())
+
+	process := func(s string, req *schema.OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan schema.OpenAIResponse) {
+		ComputeChoices(req, s, config, o, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
+			resp := schema.OpenAIResponse{
+				ID:      id,
+				Created: created,
+				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []schema.Choice{
+					{
+						Index: 0,
+						Text:  s,
+					},
+				},
+				Object: "text_completion",
+				Usage: schema.OpenAIUsage{
+					PromptTokens:     usage.Prompt,
+					CompletionTokens: usage.Completion,
+					TotalTokens:      usage.Prompt + usage.Completion,
+				},
+			}
+			log.Debug().Msgf("Sending goroutine: %s", s)
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+
+	return func(c *fiber.Ctx) error {
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("`input`: %+v", input)
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		if input.ResponseFormat.Type == "json_object" {
+			input.Grammar = grammar.JSONBNF
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		if input.Stream {
+			log.Debug().Msgf("Stream request received")
+			c.Context().SetContentType("text/event-stream")
+			//c.Response().Header.SetContentType(fiber.MIMETextHTMLCharsetUTF8)
+			//c.Set("Content-Type", "text/event-stream")
+			c.Set("Cache-Control", "no-cache")
+			c.Set("Connection", "keep-alive")
+			c.Set("Transfer-Encoding", "chunked")
+		}
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Completion != "" {
+			templateFile = config.TemplateConfig.Completion
+		}
+
+		if input.Stream {
+			if len(config.PromptStrings) > 1 {
+				return errors.New("cannot handle more than 1 `PromptStrings` when Streaming")
+			}
+
+			predInput := config.PromptStrings[0]
+
+			if templateFile != "" {
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.CompletionPromptTemplate, templateFile, model.PromptTemplateData{
+					Input: predInput,
+				})
+				if err == nil {
+					predInput = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", predInput)
+				}
+			}
+
+			responses := make(chan schema.OpenAIResponse)
+
+			go process(predInput, input, config, o.Loader, responses)
+
+			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
+
+				for ev := range responses {
+					var buf bytes.Buffer
+					enc := json.NewEncoder(&buf)
+					enc.Encode(ev)
+
+					log.Debug().Msgf("Sending chunk: %s", buf.String())
+					fmt.Fprintf(w, "data: %v\n", buf.String())
+					w.Flush()
+				}
+
+				resp := &schema.OpenAIResponse{
+					ID:      id,
+					Created: created,
+					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []schema.Choice{
+						{
+							Index:        0,
+							FinishReason: "stop",
+						},
+					},
+					Object: "text_completion",
+				}
+				respData, _ := json.Marshal(resp)
+
+				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
+				w.WriteString("data: [DONE]\n\n")
+				w.Flush()
+			}))
+			return nil
+		}
+
+		var result []schema.Choice
+
+		totalTokenUsage := backend.TokenUsage{}
+
+		for k, i := range config.PromptStrings {
+			if templateFile != "" {
+				// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.CompletionPromptTemplate, templateFile, model.PromptTemplateData{
+					SystemPrompt: config.SystemPrompt,
+					Input:        i,
+				})
+				if err == nil {
+					i = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", i)
+				}
+			}
+
+			r, tokenUsage, err := ComputeChoices(
+				input, i, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+					*c = append(*c, schema.Choice{Text: s, FinishReason: "stop", Index: k})
+				}, nil)
+			if err != nil {
+				return err
+			}
+
+			totalTokenUsage.Prompt += tokenUsage.Prompt
+			totalTokenUsage.Completion += tokenUsage.Completion
+
+			result = append(result, r...)
+		}
+
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "text_completion",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     totalTokenUsage.Prompt,
+				CompletionTokens: totalTokenUsage.Completion,
+				TotalTokens:      totalTokenUsage.Prompt + totalTokenUsage.Completion,
+			},
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/edit.go

@@ -0,0 +1,94 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+
+	"github.com/rs/zerolog/log"
+)
+
+func EditEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Edit != "" {
+			templateFile = config.TemplateConfig.Edit
+		}
+
+		var result []schema.Choice
+		totalTokenUsage := backend.TokenUsage{}
+
+		for _, i := range config.InputStrings {
+			if templateFile != "" {
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.EditPromptTemplate, templateFile, model.PromptTemplateData{
+					Input:        i,
+					Instruction:  input.Instruction,
+					SystemPrompt: config.SystemPrompt,
+				})
+				if err == nil {
+					i = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", i)
+				}
+			}
+
+			r, tokenUsage, err := ComputeChoices(input, i, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+				*c = append(*c, schema.Choice{Text: s})
+			}, nil)
+			if err != nil {
+				return err
+			}
+
+			totalTokenUsage.Prompt += tokenUsage.Prompt
+			totalTokenUsage.Completion += tokenUsage.Completion
+
+			result = append(result, r...)
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "edit",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     totalTokenUsage.Prompt,
+				CompletionTokens: totalTokenUsage.Completion,
+				TotalTokens:      totalTokenUsage.Prompt + totalTokenUsage.Completion,
+			},
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/embeddings.go

@@ -0,0 +1,78 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/google/uuid"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+// https://platform.openai.com/docs/api-reference/embeddings
+func EmbeddingsEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		model, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(model, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+		items := []schema.Item{}
+
+		for i, s := range config.InputToken {
+			// get the model function to call for the result
+			embedFn, err := backend.ModelEmbedding("", s, o.Loader, *config, o)
+			if err != nil {
+				return err
+			}
+
+			embeddings, err := embedFn()
+			if err != nil {
+				return err
+			}
+			items = append(items, schema.Item{Embedding: embeddings, Index: i, Object: "embedding"})
+		}
+
+		for i, s := range config.InputStrings {
+			// get the model function to call for the result
+			embedFn, err := backend.ModelEmbedding(s, []int{}, o.Loader, *config, o)
+			if err != nil {
+				return err
+			}
+
+			embeddings, err := embedFn()
+			if err != nil {
+				return err
+			}
+			items = append(items, schema.Item{Embedding: embeddings, Index: i, Object: "embedding"})
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Data:    items,
+			Object:  "list",
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/image.go

@@ -0,0 +1,239 @@
+package openai
+
+import (
+	"bufio"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/google/uuid"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+func downloadFile(url string) (string, error) {
+	// Get the data
+	resp, err := http.Get(url)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+
+	// Create the file
+	out, err := os.CreateTemp("", "image")
+	if err != nil {
+		return "", err
+	}
+	defer out.Close()
+
+	// Write the body to file
+	_, err = io.Copy(out, resp.Body)
+	return out.Name(), err
+}
+
+// https://platform.openai.com/docs/api-reference/images/create
+
+/*
+*
+
+	curl http://localhost:8080/v1/images/generations \
+	  -H "Content-Type: application/json" \
+	  -d '{
+	    "prompt": "A cute baby sea otter",
+	    "n": 1,
+	    "size": "512x512"
+	  }'
+
+*
+*/
+func ImageEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		m, input, err := readInput(c, o, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		if m == "" {
+			m = model.StableDiffusionBackend
+		}
+		log.Debug().Msgf("Loading model: %+v", m)
+
+		config, input, err := readConfig(m, input, cm, o.Loader, o.Debug, 0, 0, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		src := ""
+		if input.File != "" {
+
+			fileData := []byte{}
+			// check if input.File is an URL, if so download it and save it
+			// to a temporary file
+			if strings.HasPrefix(input.File, "http://") || strings.HasPrefix(input.File, "https://") {
+				out, err := downloadFile(input.File)
+				if err != nil {
+					return fmt.Errorf("failed downloading file:%w", err)
+				}
+				defer os.RemoveAll(out)
+
+				fileData, err = os.ReadFile(out)
+				if err != nil {
+					return fmt.Errorf("failed reading file:%w", err)
+				}
+
+			} else {
+				// base 64 decode the file and write it somewhere
+				// that we will cleanup
+				fileData, err = base64.StdEncoding.DecodeString(input.File)
+				if err != nil {
+					return err
+				}
+			}
+
+			// Create a temporary file
+			outputFile, err := os.CreateTemp(o.ImageDir, "b64")
+			if err != nil {
+				return err
+			}
+			// write the base64 result
+			writer := bufio.NewWriter(outputFile)
+			_, err = writer.Write(fileData)
+			if err != nil {
+				outputFile.Close()
+				return err
+			}
+			outputFile.Close()
+			src = outputFile.Name()
+			defer os.RemoveAll(src)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		switch config.Backend {
+		case "stablediffusion":
+			config.Backend = model.StableDiffusionBackend
+		case "tinydream":
+			config.Backend = model.TinyDreamBackend
+		case "":
+			config.Backend = model.StableDiffusionBackend
+		}
+
+		sizeParts := strings.Split(input.Size, "x")
+		if len(sizeParts) != 2 {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+		width, err := strconv.Atoi(sizeParts[0])
+		if err != nil {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+		height, err := strconv.Atoi(sizeParts[1])
+		if err != nil {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+
+		b64JSON := false
+		if input.ResponseFormat.Type == "b64_json" {
+			b64JSON = true
+		}
+		// src and clip_skip
+		var result []schema.Item
+		for _, i := range config.PromptStrings {
+			n := input.N
+			if input.N == 0 {
+				n = 1
+			}
+			for j := 0; j < n; j++ {
+				prompts := strings.Split(i, "|")
+				positive_prompt := prompts[0]
+				negative_prompt := ""
+				if len(prompts) > 1 {
+					negative_prompt = prompts[1]
+				}
+
+				mode := 0
+				step := config.Step
+				if step == 0 {
+					step = 15
+				}
+
+				if input.Mode != 0 {
+					mode = input.Mode
+				}
+
+				if input.Step != 0 {
+					step = input.Step
+				}
+
+				tempDir := ""
+				if !b64JSON {
+					tempDir = o.ImageDir
+				}
+				// Create a temporary file
+				outputFile, err := os.CreateTemp(tempDir, "b64")
+				if err != nil {
+					return err
+				}
+				outputFile.Close()
+				output := outputFile.Name() + ".png"
+				// Rename the temporary file
+				err = os.Rename(outputFile.Name(), output)
+				if err != nil {
+					return err
+				}
+
+				baseURL := c.BaseURL()
+
+				fn, err := backend.ImageGeneration(height, width, mode, step, input.Seed, positive_prompt, negative_prompt, src, output, o.Loader, *config, o)
+				if err != nil {
+					return err
+				}
+				if err := fn(); err != nil {
+					return err
+				}
+
+				item := &schema.Item{}
+
+				if b64JSON {
+					defer os.RemoveAll(output)
+					data, err := os.ReadFile(output)
+					if err != nil {
+						return err
+					}
+					item.B64JSON = base64.StdEncoding.EncodeToString(data)
+				} else {
+					base := filepath.Base(output)
+					item.URL = baseURL + "/generated-images/" + base
+				}
+
+				result = append(result, *item)
+			}
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Data:    result,
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/inference.go

@@ -0,0 +1,55 @@
+package openai
+
+import (
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ComputeChoices(
+	req *schema.OpenAIRequest,
+	predInput string,
+	config *config.Config,
+	o *options.Option,
+	loader *model.ModelLoader,
+	cb func(string, *[]schema.Choice),
+	tokenCallback func(string, backend.TokenUsage) bool) ([]schema.Choice, backend.TokenUsage, error) {
+	n := req.N // number of completions to return
+	result := []schema.Choice{}
+
+	if n == 0 {
+		n = 1
+	}
+
+	images := []string{}
+	for _, m := range req.Messages {
+		images = append(images, m.StringImages...)
+	}
+
+	// get the model function to call for the result
+	predFunc, err := backend.ModelInference(req.Context, predInput, images, loader, *config, o, tokenCallback)
+	if err != nil {
+		return result, backend.TokenUsage{}, err
+	}
+
+	tokenUsage := backend.TokenUsage{}
+
+	for i := 0; i < n; i++ {
+		prediction, err := predFunc()
+		if err != nil {
+			return result, backend.TokenUsage{}, err
+		}
+
+		tokenUsage.Prompt += prediction.Usage.Prompt
+		tokenUsage.Completion += prediction.Usage.Completion
+
+		finetunedResponse := backend.Finetune(*config, predInput, prediction.Response)
+		cb(finetunedResponse, &result)
+
+		//result = append(result, Choice{Text: prediction})
+
+	}
+	return result, tokenUsage, err
+}

api/openai/list.go

@@ -0,0 +1,69 @@
+package openai
+
+import (
+	"regexp"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+)
+
+func ListModelsEndpoint(loader *model.ModelLoader, cm *config.ConfigLoader) func(ctx *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		models, err := loader.ListModels()
+		if err != nil {
+			return err
+		}
+		var mm map[string]interface{} = map[string]interface{}{}
+
+		dataModels := []schema.OpenAIModel{}
+
+		var filterFn func(name string) bool
+		filter := c.Query("filter")
+
+		// If filter is not specified, do not filter the list by model name
+		if filter == "" {
+			filterFn = func(_ string) bool { return true }
+		} else {
+			// If filter _IS_ specified, we compile it to a regex which is used to create the filterFn
+			rxp, err := regexp.Compile(filter)
+			if err != nil {
+				return err
+			}
+			filterFn = func(name string) bool {
+				return rxp.MatchString(name)
+			}
+		}
+
+		// By default, exclude any loose files that are already referenced by a configuration file.
+		excludeConfigured := c.QueryBool("excludeConfigured", true)
+
+		// Start with the known configurations
+		for _, c := range cm.GetAllConfigs() {
+			if excludeConfigured {
+				mm[c.Model] = nil
+			}
+
+			if filterFn(c.Name) {
+				dataModels = append(dataModels, schema.OpenAIModel{ID: c.Name, Object: "model"})
+			}
+		}
+
+		// Then iterate through the loose files:
+		for _, m := range models {
+			// And only adds them if they shouldn't be skipped.
+			if _, exists := mm[m]; !exists && filterFn(m) {
+				dataModels = append(dataModels, schema.OpenAIModel{ID: m, Object: "model"})
+			}
+		}
+
+		return c.JSON(struct {
+			Object string               `json:"object"`
+			Data   []schema.OpenAIModel `json:"data"`
+		}{
+			Object: "list",
+			Data:   dataModels,
+		})
+	}
+}

api/openai/request.go

@@ -0,0 +1,336 @@
+package openai
+
+import (
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	options "github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+func readInput(c *fiber.Ctx, o *options.Option, randomModel bool) (string, *schema.OpenAIRequest, error) {
+	loader := o.Loader
+	input := new(schema.OpenAIRequest)
+	ctx, cancel := context.WithCancel(o.Context)
+	input.Context = ctx
+	input.Cancel = cancel
+	// Get input data from the request body
+	if err := c.BodyParser(input); err != nil {
+		return "", nil, fmt.Errorf("failed parsing request body: %w", err)
+	}
+
+	modelFile := input.Model
+
+	if c.Params("model") != "" {
+		modelFile = c.Params("model")
+	}
+
+	received, _ := json.Marshal(input)
+
+	log.Debug().Msgf("Request received: %s", string(received))
+
+	// Set model from bearer token, if available
+	bearer := strings.TrimLeft(c.Get("authorization"), "Bearer ")
+	bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
+
+	// If no model was specified, take the first available
+	if modelFile == "" && !bearerExists && randomModel {
+		models, _ := loader.ListModels()
+		if len(models) > 0 {
+			modelFile = models[0]
+			log.Debug().Msgf("No model specified, using: %s", modelFile)
+		} else {
+			log.Debug().Msgf("No model specified, returning error")
+			return "", nil, fmt.Errorf("no model specified")
+		}
+	}
+
+	// If a model is found in bearer token takes precedence
+	if bearerExists {
+		log.Debug().Msgf("Using model from bearer token: %s", bearer)
+		modelFile = bearer
+	}
+	return modelFile, input, nil
+}
+
+// this function check if the string is an URL, if it's an URL downloads the image in memory
+// encodes it in base64 and returns the base64 string
+func getBase64Image(s string) (string, error) {
+	if strings.HasPrefix(s, "http") {
+		// download the image
+		resp, err := http.Get(s)
+		if err != nil {
+			return "", err
+		}
+		defer resp.Body.Close()
+
+		// read the image data into memory
+		data, err := ioutil.ReadAll(resp.Body)
+		if err != nil {
+			return "", err
+		}
+
+		// encode the image data in base64
+		encoded := base64.StdEncoding.EncodeToString(data)
+
+		// return the base64 string
+		return encoded, nil
+	}
+
+	// if the string instead is prefixed with "data:image/jpeg;base64,", drop it
+	if strings.HasPrefix(s, "data:image/jpeg;base64,") {
+		return strings.ReplaceAll(s, "data:image/jpeg;base64,", ""), nil
+	}
+	return "", fmt.Errorf("not valid string")
+}
+
+func updateConfig(config *config.Config, input *schema.OpenAIRequest) {
+	if input.Echo {
+		config.Echo = input.Echo
+	}
+	if input.TopK != 0 {
+		config.TopK = input.TopK
+	}
+	if input.TopP != 0 {
+		config.TopP = input.TopP
+	}
+
+	if input.Backend != "" {
+		config.Backend = input.Backend
+	}
+
+	if input.ClipSkip != 0 {
+		config.Diffusers.ClipSkip = input.ClipSkip
+	}
+
+	if input.ModelBaseName != "" {
+		config.AutoGPTQ.ModelBaseName = input.ModelBaseName
+	}
+
+	if input.NegativePromptScale != 0 {
+		config.NegativePromptScale = input.NegativePromptScale
+	}
+
+	if input.UseFastTokenizer {
+		config.UseFastTokenizer = input.UseFastTokenizer
+	}
+
+	if input.NegativePrompt != "" {
+		config.NegativePrompt = input.NegativePrompt
+	}
+
+	if input.RopeFreqBase != 0 {
+		config.RopeFreqBase = input.RopeFreqBase
+	}
+
+	if input.RopeFreqScale != 0 {
+		config.RopeFreqScale = input.RopeFreqScale
+	}
+
+	if input.Grammar != "" {
+		config.Grammar = input.Grammar
+	}
+
+	if input.Temperature != 0 {
+		config.Temperature = input.Temperature
+	}
+
+	if input.Maxtokens != 0 {
+		config.Maxtokens = input.Maxtokens
+	}
+
+	switch stop := input.Stop.(type) {
+	case string:
+		if stop != "" {
+			config.StopWords = append(config.StopWords, stop)
+		}
+	case []interface{}:
+		for _, pp := range stop {
+			if s, ok := pp.(string); ok {
+				config.StopWords = append(config.StopWords, s)
+			}
+		}
+	}
+
+	// Decode each request's message content
+	index := 0
+	for i, m := range input.Messages {
+		switch content := m.Content.(type) {
+		case string:
+			input.Messages[i].StringContent = content
+		case []interface{}:
+			dat, _ := json.Marshal(content)
+			c := []schema.Content{}
+			json.Unmarshal(dat, &c)
+			for _, pp := range c {
+				if pp.Type == "text" {
+					input.Messages[i].StringContent = pp.Text
+				} else if pp.Type == "image_url" {
+					// Detect if pp.ImageURL is an URL, if it is download the image and encode it in base64:
+					base64, err := getBase64Image(pp.ImageURL.URL)
+					if err == nil {
+						input.Messages[i].StringImages = append(input.Messages[i].StringImages, base64) // TODO: make sure that we only return base64 stuff
+						// set a placeholder for each image
+						input.Messages[i].StringContent = fmt.Sprintf("[img-%d]", index) + input.Messages[i].StringContent
+						index++
+					} else {
+						fmt.Print("Failed encoding image", err)
+					}
+				}
+			}
+		}
+	}
+
+	if input.RepeatPenalty != 0 {
+		config.RepeatPenalty = input.RepeatPenalty
+	}
+
+	if input.Keep != 0 {
+		config.Keep = input.Keep
+	}
+
+	if input.Batch != 0 {
+		config.Batch = input.Batch
+	}
+
+	if input.F16 {
+		config.F16 = input.F16
+	}
+
+	if input.IgnoreEOS {
+		config.IgnoreEOS = input.IgnoreEOS
+	}
+
+	if input.Seed != 0 {
+		config.Seed = input.Seed
+	}
+
+	if input.Mirostat != 0 {
+		config.LLMConfig.Mirostat = input.Mirostat
+	}
+
+	if input.MirostatETA != 0 {
+		config.LLMConfig.MirostatETA = input.MirostatETA
+	}
+
+	if input.MirostatTAU != 0 {
+		config.LLMConfig.MirostatTAU = input.MirostatTAU
+	}
+
+	if input.TypicalP != 0 {
+		config.TypicalP = input.TypicalP
+	}
+
+	switch inputs := input.Input.(type) {
+	case string:
+		if inputs != "" {
+			config.InputStrings = append(config.InputStrings, inputs)
+		}
+	case []interface{}:
+		for _, pp := range inputs {
+			switch i := pp.(type) {
+			case string:
+				config.InputStrings = append(config.InputStrings, i)
+			case []interface{}:
+				tokens := []int{}
+				for _, ii := range i {
+					tokens = append(tokens, int(ii.(float64)))
+				}
+				config.InputToken = append(config.InputToken, tokens)
+			}
+		}
+	}
+
+	// Can be either a string or an object
+	switch fnc := input.FunctionCall.(type) {
+	case string:
+		if fnc != "" {
+			config.SetFunctionCallString(fnc)
+		}
+	case map[string]interface{}:
+		var name string
+		n, exists := fnc["name"]
+		if exists {
+			nn, e := n.(string)
+			if e {
+				name = nn
+			}
+		}
+		config.SetFunctionCallNameString(name)
+	}
+
+	switch p := input.Prompt.(type) {
+	case string:
+		config.PromptStrings = append(config.PromptStrings, p)
+	case []interface{}:
+		for _, pp := range p {
+			if s, ok := pp.(string); ok {
+				config.PromptStrings = append(config.PromptStrings, s)
+			}
+		}
+	}
+}
+
+func readConfig(modelFile string, input *schema.OpenAIRequest, cm *config.ConfigLoader, loader *model.ModelLoader, debug bool, threads, ctx int, f16 bool) (*config.Config, *schema.OpenAIRequest, error) {
+	// Load a config file if present after the model name
+	modelConfig := filepath.Join(loader.ModelPath, modelFile+".yaml")
+
+	var cfg *config.Config
+
+	defaults := func() {
+		cfg = config.DefaultConfig(modelFile)
+		cfg.ContextSize = ctx
+		cfg.Threads = threads
+		cfg.F16 = f16
+		cfg.Debug = debug
+	}
+
+	cfgExisting, exists := cm.GetConfig(modelFile)
+	if !exists {
+		if _, err := os.Stat(modelConfig); err == nil {
+			if err := cm.LoadConfig(modelConfig); err != nil {
+				return nil, nil, fmt.Errorf("failed loading model config (%s) %s", modelConfig, err.Error())
+			}
+			cfgExisting, exists = cm.GetConfig(modelFile)
+			if exists {
+				cfg = &cfgExisting
+			} else {
+				defaults()
+			}
+		} else {
+			defaults()
+		}
+	} else {
+		cfg = &cfgExisting
+	}
+
+	// Set the parameters for the language model prediction
+	updateConfig(cfg, input)
+
+	// Don't allow 0 as setting
+	if cfg.Threads == 0 {
+		if threads != 0 {
+			cfg.Threads = threads
+		} else {
+			cfg.Threads = 4
+		}
+	}
+
+	// Enforce debug flag if passed from CLI
+	if debug {
+		cfg.Debug = true
+	}
+
+	return cfg, input, nil
+}

api/openai/transcription.go

@@ -0,0 +1,71 @@
+package openai
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path"
+	"path/filepath"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+// https://platform.openai.com/docs/api-reference/audio/create
+func TranscriptEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		m, input, err := readInput(c, o, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(m, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+		// retrieve the file data from the request
+		file, err := c.FormFile("file")
+		if err != nil {
+			return err
+		}
+		f, err := file.Open()
+		if err != nil {
+			return err
+		}
+		defer f.Close()
+
+		dir, err := os.MkdirTemp("", "whisper")
+
+		if err != nil {
+			return err
+		}
+		defer os.RemoveAll(dir)
+
+		dst := filepath.Join(dir, path.Base(file.Filename))
+		dstFile, err := os.Create(dst)
+		if err != nil {
+			return err
+		}
+
+		if _, err := io.Copy(dstFile, f); err != nil {
+			log.Debug().Msgf("Audio file copying error %+v - %+v - err %+v", file.Filename, dst, err)
+			return err
+		}
+
+		log.Debug().Msgf("Audio file copied to: %+v", dst)
+
+		tr, err := backend.ModelTranscription(dst, input.Language, o.Loader, *config, o)
+		if err != nil {
+			return err
+		}
+
+		log.Debug().Msgf("Trascribed: %+v", tr)
+		// TODO: handle different outputs here
+		return c.Status(http.StatusOK).JSON(tr)
+	}
+}