config/config.yaml

@@ -17,14 +17,8 @@ ollama:
   format: "json"
 
 vllm:
-  vl_model:
+  base_url: "http://172.17.0.1:8001/v1"
-    base_url: "http://192.168.6.115:8001/v1"
+  vl_model: "qwen2.5-vl-3b-awq"
-    model: "qwen2.5-vl-3b-awq"
-    timeout: "120s"
-    level: "info"
-  text_model:
-    base_url: "http://192.168.6.115:8002/v1"
-    model: "qwen3-8b-fp8"
   timeout: "120s"
   level: "info"
 

config/config_env.yaml

@@ -14,14 +14,8 @@ ollama:
   format: "json"
 
 vllm:
-  vl_model:
+  base_url: "http://117.175.169.61:16001/v1"
-    base_url: "http://192.168.6.115:8001/v1"
+  vl_model: "qwen2.5-vl-3b-awq"
-    model: "qwen2.5-vl-3b-awq"
-    timeout: "120s"
-    level: "info"
-  text_model:
-    base_url: "http://192.168.6.115:8002/v1"
-    model: "qwen3-8b-fp8"
   timeout: "120s"
   level: "info"
 

config/config_test.yaml

@@ -14,14 +14,8 @@ ollama:
   format: "json"
 
 vllm:
-  vl_model:
+  base_url: "http://host.docker.internal:8001/v1"
-    base_url: "http://192.168.6.115:8001/v1"
+  vl_model: "qwen2.5-vl-3b-awq"
-    model: "qwen2.5-vl-3b-awq"
-    timeout: "120s"
-    level: "info"
-  text_model:
-    base_url: "http://192.168.6.115:8002/v1"
-    model: "qwen3-8b-fp8"
   timeout: "120s"
   level: "info"
 

internal/biz/do/handle.go

@@ -36,7 +36,6 @@ import (
 
 type Handle struct {
 	Ollama                *llm_service.OllamaService
-	Vllm                  *llm_service.VllmService
 	toolManager           *tools.Manager
 	conf                  *config.Config
 	sessionImpl           *impl.SessionImpl
@@ -48,7 +47,6 @@ type Handle struct {
 
 func NewHandle(
 	Ollama *llm_service.OllamaService,
-	Vllm *llm_service.VllmService,
 	toolManager *tools.Manager,
 	conf *config.Config,
 	sessionImpl *impl.SessionImpl,
@@ -59,7 +57,6 @@ func NewHandle(
 ) *Handle {
 	return &Handle{
 		Ollama:                Ollama,
-		Vllm:                  Vllm,
 		toolManager:           toolManager,
 		conf:                  conf,
 		sessionImpl:           sessionImpl,
@@ -75,8 +72,7 @@ func (r *Handle) Recognize(ctx context.Context, rec *entitys.Recognize, promptPr
 
 	prompt, err := promptProcessor.CreatePrompt(ctx, rec)
 	//意图识别
-	// recognizeMsg, err := r.Ollama.IntentRecognize(ctx, &entitys.ToolSelect{
+	recognizeMsg, err := r.Ollama.IntentRecognize(ctx, &entitys.ToolSelect{
-	recognizeMsg, err := r.Vllm.IntentRecognize(ctx, &entitys.ToolSelect{
 		Prompt: prompt,
 		Tools:  rec.Tasks,
 	})

internal/biz/llm_service/vllm.go

@@ -1,153 +0,0 @@
-package llm_service
-
-import (
-	"ai_scheduler/internal/config"
-	"ai_scheduler/internal/entitys"
-	"ai_scheduler/internal/pkg"
-	"ai_scheduler/internal/pkg/utils_vllm"
-	"context"
-	"encoding/base64"
-	"errors"
-	"fmt"
-	"strings"
-
-	"github.com/cloudwego/eino/schema"
-	"github.com/ollama/ollama/api"
-)
-
-type VllmService struct {
-	client *utils_vllm.Client
-	config *config.Config
-}
-
-func NewVllmService(
-	client *utils_vllm.Client,
-	config *config.Config,
-) *VllmService {
-	return &VllmService{
-		client: client,
-		config: config,
-	}
-}
-
-func (s *VllmService) IntentRecognize(ctx context.Context, req *entitys.ToolSelect) (msg string, err error) {
-	msgs := s.convertMessages(req.Prompt)
-	tools := s.convertTools(req.Tools)
-
-	resp, err := s.client.ToolSelect(ctx, msgs, tools)
-	if err != nil {
-		return
-	}
-
-	if resp.Content == "" {
-		if len(resp.ToolCalls) > 0 {
-			call := resp.ToolCalls[0]
-			var matchFromTools = &entitys.Match{
-				Confidence: 1,
-				Index:      call.Function.Name,
-				Parameters: call.Function.Arguments,
-				IsMatch:    true,
-			}
-			msg = pkg.JsonStringIgonErr(matchFromTools)
-		} else {
-			err = errors.New("不太明白你想表达的意思呢，可以在仔细描述一下您所需要的内容吗，感谢感谢")
-			return
-		}
-	} else {
-		msg = resp.Content
-	}
-	return
-}
-
-func (s *VllmService) convertMessages(prompts []api.Message) []*schema.Message {
-	msgs := make([]*schema.Message, 0, len(prompts))
-	for _, p := range prompts {
-		msg := &schema.Message{
-			Role:    schema.RoleType(p.Role),
-			Content: p.Content,
-		}
-
-		// 这里实际应该不会走进来
-		if len(p.Images) > 0 {
-			parts := []schema.MessageInputPart{
-				{Type: schema.ChatMessagePartTypeText, Text: p.Content},
-			}
-			for _, imgData := range p.Images {
-				b64 := base64.StdEncoding.EncodeToString(imgData)
-				mimeType := "image/jpeg"
-				parts = append(parts, schema.MessageInputPart{
-					Type: schema.ChatMessagePartTypeImageURL,
-					Image: &schema.MessageInputImage{
-						MessagePartCommon: schema.MessagePartCommon{
-							MIMEType:   mimeType,
-							Base64Data: &b64,
-						},
-					},
-				})
-			}
-			msg.UserInputMultiContent = parts
-		}
-		msgs = append(msgs, msg)
-	}
-	return msgs
-}
-
-func (s *VllmService) convertTools(tasks []entitys.RegistrationTask) []*schema.ToolInfo {
-	tools := make([]*schema.ToolInfo, 0, len(tasks))
-	for _, task := range tasks {
-		params := make(map[string]*schema.ParameterInfo)
-		for k, v := range task.TaskConfigDetail.Param.Properties {
-			dt := schema.String
-
-			// Handle v.Type dynamically to support both string and []string (compiler suggests []string)
-			// Using fmt.Sprint handles both cases safely without knowing exact type structure
-			typeStr := fmt.Sprintf("%v", v.Type)
-			typeStr = strings.Trim(typeStr, "[]") // normalize "[string]" -> "string"
-
-			switch typeStr {
-			case "string":
-				dt = schema.String
-			case "integer", "int":
-				dt = schema.Integer
-			case "number", "float":
-				dt = schema.Number
-			case "boolean", "bool":
-				dt = schema.Boolean
-			case "object":
-				dt = schema.Object
-			case "array":
-				dt = schema.Array
-			}
-
-			required := false
-			for _, r := range task.TaskConfigDetail.Param.Required {
-				if r == k {
-					required = true
-					break
-				}
-			}
-
-			desc := v.Description
-			if len(v.Enum) > 0 {
-				var enumStrs []string
-				for _, e := range v.Enum {
-					enumStrs = append(enumStrs, fmt.Sprintf("%v", e))
-				}
-				desc += " Enum: " + strings.Join(enumStrs, ", ")
-			}
-
-			params[k] = &schema.ParameterInfo{
-				Type:     dt,
-				Desc:     desc,
-				Required: required,
-			}
-		}
-
-		tools = append(tools, &schema.ToolInfo{
-			Name:        task.Name,
-			Desc:        task.Desc,
-			ParamsOneOf: schema.NewParamsOneOfByParams(params),
-		})
-	}
-	return tools
-}

internal/biz/provider_set.go

@@ -13,7 +13,6 @@ var ProviderSetBiz = wire.NewSet(
 	NewChatHistoryBiz,
 	//llm_service.NewLangChainGenerate,
 	llm_service.NewOllamaGenerate,
-	llm_service.NewVllmService,
 	//handle.NewHandle,
 	do.NewDo,
 	do.NewHandle,

internal/config/config.go

@@ -122,13 +122,8 @@ type OllamaConfig struct {
 }
 
 type VllmConfig struct {
-	VLModel   VllmModel `mapstructure:"vl_model"`
-	TextModel VllmModel `mapstructure:"text_model"`
-}
-
-type VllmModel struct {
 	BaseURL string        `mapstructure:"base_url"`
-	Model   string        `mapstructure:"model"`
+	VlModel string        `mapstructure:"vl_model"`
 	Timeout time.Duration `mapstructure:"timeout"`
 	Level   string        `mapstructure:"level"`
 }

internal/pkg/utils_vllm/client.go

@@ -7,63 +7,33 @@ import (
 	"encoding/base64"
 
 	"github.com/cloudwego/eino-ext/components/model/openai"
-	"github.com/cloudwego/eino/components/model"
 	"github.com/cloudwego/eino/schema"
 )
 
 type Client struct {
-	vlModel       *openai.ChatModel
+	model  *openai.ChatModel
-	generateModel *openai.ChatModel
 	config *config.Config
 }
 
 func NewClient(config *config.Config) (*Client, func(), error) {
-	// 初始化视觉模型
+	m, err := openai.NewChatModel(context.Background(), &openai.ChatModelConfig{
-	vl, err := openai.NewChatModel(context.Background(), &openai.ChatModelConfig{
+		BaseURL: config.Vllm.BaseURL,
-		BaseURL: config.Vllm.VLModel.BaseURL,
+		Model:   config.Vllm.VlModel,
-		Model:   config.Vllm.VLModel.Model,
+		Timeout: config.Vllm.Timeout,
-		Timeout: config.Vllm.VLModel.Timeout,
 	})
 	if err != nil {
 		return nil, nil, err
 	}
-
+	c := &Client{model: m, config: config}
-	// 初始化生成模型
-	gen, err := openai.NewChatModel(context.Background(), &openai.ChatModelConfig{
-		BaseURL: config.Vllm.TextModel.BaseURL,
-		Model:   config.Vllm.TextModel.Model,
-		Timeout: config.Vllm.TextModel.Timeout,
-		ExtraFields: map[string]any{
-			"chat_template_kwargs": map[string]any{
-				"enable_thinking": false,
-			},
-		},
-	})
-	if err != nil {
-		return nil, nil, err
-	}
-
-	c := &Client{
-		vlModel:       vl,
-		generateModel: gen,
-		config:        config,
-	}
 	cleanup := func() {}
 	return c, cleanup, nil
 }
 
 func (c *Client) Chat(ctx context.Context, msgs []*schema.Message) (*schema.Message, error) {
-	// 默认聊天使用生成模型
+	return c.model.Generate(ctx, msgs)
-	return c.generateModel.Generate(ctx, msgs)
-}
-
-func (c *Client) ToolSelect(ctx context.Context, msgs []*schema.Message, tools []*schema.ToolInfo) (*schema.Message, error) {
-	// 工具选择使用生成模型
-	return c.generateModel.Generate(ctx, msgs, model.WithTools(tools))
 }
 
 func (c *Client) RecognizeWithImg(ctx context.Context, systemPrompt, userPrompt string, imgURLs []string) (*schema.Message, error) {
-	// 图片识别使用视觉模型
 	in := []*schema.Message{
 		{
 			Role:    schema.System,
@@ -88,12 +58,11 @@ func (c *Client) RecognizeWithImg(ctx context.Context, systemPrompt, userPrompt
 	}
 
 	in[1].UserInputMultiContent = parts
-	return c.vlModel.Generate(ctx, in)
+	return c.model.Generate(ctx, in)
 }
 
 // 识别图片by二进制文件
 func (c *Client) RecognizeWithImgBytes(ctx context.Context, systemPrompt, userPrompt string, imgBytes []byte, imgType string) (*schema.Message, error) {
-	// 图片识别使用视觉模型
 	in := []*schema.Message{
 		{
 			Role:    schema.System,
@@ -113,10 +82,9 @@ func (c *Client) RecognizeWithImgBytes(ctx context.Context, systemPrompt, userPr
 				MIMEType:   imgType,
 				Base64Data: util.AnyToPoint(base64.StdEncoding.EncodeToString(imgBytes)),
 			},
-			Detail: schema.ImageURLDetailHigh,
 		},
 	})
 
 	in[1].UserInputMultiContent = parts
-	return c.vlModel.Generate(ctx, in)
+	return c.model.Generate(ctx, in)
 }