feat(ai-proxy): Adjust the streaming response structure to keep it consistent with the openai (#2391)

Signed-off-by: Xijun Dai <daixijun1990@gmail.com>
2026-03-08 02:30:56 +08:00 · 2025-06-12 16:25:35 +08:00
parent 86934b3203
commit fb8dd819e9
12 changed files with 180 additions and 178 deletions
--- a/plugins/wasm-go/extensions/ai-proxy/provider/bedrock.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/bedrock.go
@@ -42,8 +42,7 @@ const (
 	requestIdHeader        = "X-Amzn-Requestid"
 )

-type bedrockProviderInitializer struct {
-}
+type bedrockProviderInitializer struct{}

 func (b *bedrockProviderInitializer) ValidateConfig(config *ProviderConfig) error {
 	if len(config.awsAccessKey) == 0 || len(config.awsSecretKey) == 0 {
@@ -104,7 +103,7 @@ func (b *bedrockProvider) convertEventFromBedrockToOpenAI(ctx wrapper.HttpContex
 		chatChoice.Delta = &chatMessage{Content: bedrockEvent.Delta.Text}
 	}
 	if bedrockEvent.StopReason != nil {
-		chatChoice.FinishReason = stopReasonBedrock2OpenAI(*bedrockEvent.StopReason)
+		chatChoice.FinishReason = util.Ptr(stopReasonBedrock2OpenAI(*bedrockEvent.StopReason))
 	}
 	choices = append(choices, chatChoice)
 	requestId := ctx.GetStringContext(requestIdHeader, "")
@@ -118,7 +117,7 @@ func (b *bedrockProvider) convertEventFromBedrockToOpenAI(ctx wrapper.HttpContex
 	}
 	if bedrockEvent.Usage != nil {
 		openAIFormattedChunk.Choices = choices[:0]
-		openAIFormattedChunk.Usage = usage{
+		openAIFormattedChunk.Usage = &usage{
 			CompletionTokens: bedrockEvent.Usage.OutputTokens,
 			PromptTokens:     bedrockEvent.Usage.InputTokens,
 			TotalTokens:      bedrockEvent.Usage.TotalTokens,
@@ -756,7 +755,7 @@ func (b *bedrockProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, b
 				Role:    bedrockResponse.Output.Message.Role,
 				Content: outputContent,
 			},
-			FinishReason: stopReasonBedrock2OpenAI(bedrockResponse.StopReason),
+			FinishReason: util.Ptr(stopReasonBedrock2OpenAI(bedrockResponse.StopReason)),
 		},
 	}
 	requestId := ctx.GetStringContext(requestIdHeader, "")
@@ -768,7 +767,7 @@ func (b *bedrockProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, b
 		SystemFingerprint: "",
 		Object:            objectChatCompletion,
 		Choices:           choices,
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     bedrockResponse.Usage.InputTokens,
 			CompletionTokens: bedrockResponse.Usage.OutputTokens,
 			TotalTokens:      bedrockResponse.Usage.TotalTokens,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/claude.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/claude.go
@@ -341,7 +341,7 @@ func (c *claudeProvider) responseClaude2OpenAI(ctx wrapper.HttpContext, origResp
 	choice := chatCompletionChoice{
 		Index:        0,
 		Message:      &chatMessage{Role: roleAssistant, Content: origResponse.Content[0].Text},
-		FinishReason: stopReasonClaude2OpenAI(origResponse.StopReason),
+		FinishReason: util.Ptr(stopReasonClaude2OpenAI(origResponse.StopReason)),
 	}

 	return &chatCompletionResponse{
@@ -351,7 +351,7 @@ func (c *claudeProvider) responseClaude2OpenAI(ctx wrapper.HttpContext, origResp
 		SystemFingerprint: "",
 		Object:            objectChatCompletion,
 		Choices:           []chatCompletionChoice{choice},
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     origResponse.Usage.InputTokens,
 			CompletionTokens: origResponse.Usage.OutputTokens,
 			TotalTokens:      origResponse.Usage.InputTokens + origResponse.Usage.OutputTokens,
@@ -404,7 +404,7 @@ func (c *claudeProvider) streamResponseClaude2OpenAI(ctx wrapper.HttpContext, or
 		choice := chatCompletionChoice{
 			Index:        origResponse.Index,
 			Delta:        &chatMessage{},
-			FinishReason: stopReasonClaude2OpenAI(origResponse.Delta.StopReason),
+			FinishReason: util.Ptr(stopReasonClaude2OpenAI(origResponse.Delta.StopReason)),
 		}
 		return c.createChatCompletionResponse(ctx, origResponse, choice)
 	case "message_stop":
@@ -415,7 +415,7 @@ func (c *claudeProvider) streamResponseClaude2OpenAI(ctx wrapper.HttpContext, or
 			Object:      objectChatCompletionChunk,
 			Choices:     []chatCompletionChoice{},
 			ServiceTier: c.serviceTier,
-			Usage: usage{
+			Usage: &usage{
 				PromptTokens:     c.usage.PromptTokens,
 				CompletionTokens: c.usage.CompletionTokens,
 				TotalTokens:      c.usage.TotalTokens,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/dify.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/dify.go
@@ -116,23 +116,23 @@ func (d *difyProvider) responseDify2OpenAI(ctx wrapper.HttpContext, response *Di
 		choice = chatCompletionChoice{
 			Index:        0,
 			Message:      &chatMessage{Role: roleAssistant, Content: response.Answer},
-			FinishReason: finishReasonStop,
+			FinishReason: util.Ptr(finishReasonStop),
 		}
-		//response header中增加conversationId字段
+		// response header中增加conversationId字段
 		_ = proxywasm.ReplaceHttpResponseHeader("ConversationId", response.ConversationId)
 		id = response.ConversationId
 	case BotTypeCompletion:
 		choice = chatCompletionChoice{
 			Index:        0,
 			Message:      &chatMessage{Role: roleAssistant, Content: response.Answer},
-			FinishReason: finishReasonStop,
+			FinishReason: util.Ptr(finishReasonStop),
 		}
 		id = response.MessageId
 	case BotTypeWorkflow:
 		choice = chatCompletionChoice{
 			Index:        0,
 			Message:      &chatMessage{Role: roleAssistant, Content: response.Data.Outputs[d.config.outputVariable]},
-			FinishReason: finishReasonStop,
+			FinishReason: util.Ptr(finishReasonStop),
 		}
 		id = response.Data.WorkflowId
 	}
@@ -143,7 +143,7 @@ func (d *difyProvider) responseDify2OpenAI(ctx wrapper.HttpContext, response *Di
 		SystemFingerprint: "",
 		Object:            objectChatCompletion,
 		Choices:           []chatCompletionChoice{choice},
-		Usage:             response.MetaData.Usage,
+		Usage:             &response.MetaData.Usage,
 	}
 }

@@ -188,7 +188,7 @@ func (d *difyProvider) OnStreamingResponseBody(ctx wrapper.HttpContext, name Api
 func (d *difyProvider) streamResponseDify2OpenAI(ctx wrapper.HttpContext, response *DifyChunkChatResponse) *chatCompletionResponse {
 	var choice chatCompletionChoice
 	var id string
-	var responseUsage usage
+	var responseUsage *usage
 	switch d.config.botType {
 	case BotTypeChat, BotTypeAgent:
 		choice = chatCompletionChoice{
@@ -211,9 +211,9 @@ func (d *difyProvider) streamResponseDify2OpenAI(ctx wrapper.HttpContext, respon
 		id = response.Data.WorkflowId
 	}
 	if response.Event == "message_end" || response.Event == "workflow_finished" {
-		choice.FinishReason = finishReasonStop
+		choice.FinishReason = util.Ptr(finishReasonStop)
 		if response.Event == "message_end" {
-			responseUsage = usage{
+			responseUsage = &usage{
 				PromptTokens:     response.MetaData.Usage.PromptTokens,
 				CompletionTokens: response.MetaData.Usage.CompletionTokens,
 				TotalTokens:      response.MetaData.Usage.TotalTokens,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/gemini.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/gemini.go
@@ -500,7 +500,7 @@ func (g *geminiProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, re
 		Object:  objectChatCompletion,
 		Created: time.Now().UnixMilli() / 1000,
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     response.UsageMetadata.PromptTokenCount,
 			CompletionTokens: response.UsageMetadata.CandidatesTokenCount,
 			TotalTokens:      response.UsageMetadata.TotalTokenCount,
@@ -514,7 +514,7 @@ func (g *geminiProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, re
 				Message: &chatMessage{
 					Role: roleAssistant,
 				},
-				FinishReason: finishReasonStop,
+				FinishReason: util.Ptr(finishReasonStop),
 			}
 			if part.FunctionCall != nil {
 				choice.Message.ToolCalls = g.buildToolCalls(&candidate)
@@ -524,7 +524,7 @@ func (g *geminiProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, re
 				choice.Message.Content = part.Text
 			}

-			choice.FinishReason = candidate.FinishReason
+			choice.FinishReason = util.Ptr(candidate.FinishReason)
 			fullTextResponse.Choices = append(fullTextResponse.Choices, choice)
 			choiceIndex += 1
 		}
@@ -567,7 +567,7 @@ func (g *geminiProvider) buildChatCompletionStreamResponse(ctx wrapper.HttpConte
 		Created: time.Now().UnixMilli() / 1000,
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		Choices: []chatCompletionChoice{choice},
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     geminiResp.UsageMetadata.PromptTokenCount,
 			CompletionTokens: geminiResp.UsageMetadata.CandidatesTokenCount,
 			TotalTokens:      geminiResp.UsageMetadata.TotalTokenCount,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/hunyuan.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/hunyuan.go
@@ -387,7 +387,7 @@ func (m *hunyuanProvider) convertChunkFromHunyuanToOpenAI(ctx wrapper.HttpContex
 		Model:             ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		SystemFingerprint: "",
 		Object:            objectChatCompletionChunk,
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     hunyuanFormattedChunk.Usage.PromptTokens,
 			CompletionTokens: hunyuanFormattedChunk.Usage.CompletionTokens,
 			TotalTokens:      hunyuanFormattedChunk.Usage.TotalTokens,
@@ -400,7 +400,7 @@ func (m *hunyuanProvider) convertChunkFromHunyuanToOpenAI(ctx wrapper.HttpContex
 	if hunyuanFormattedChunk.Choices[0].FinishReason == hunyuanStreamEndMark {
 		// log.Debugf("@@@ --- 最后chunk: ")
 		openAIFormattedChunk.Choices = append(openAIFormattedChunk.Choices, chatCompletionChoice{
-			FinishReason: hunyuanFormattedChunk.Choices[0].FinishReason,
+			FinishReason: util.Ptr(hunyuanFormattedChunk.Choices[0].FinishReason),
 		})
 	} else {
 		deltaMsg := chatMessage{
@@ -495,7 +495,7 @@ func (m *hunyuanProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, h
 				Content:   choice.Message.Content,
 				ToolCalls: nil,
 			},
-			FinishReason: choice.FinishReason,
+			FinishReason: util.Ptr(choice.FinishReason),
 		})
 	}
 	return &chatCompletionResponse{
@@ -505,7 +505,7 @@ func (m *hunyuanProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, h
 		SystemFingerprint: "",
 		Object:            objectChatCompletion,
 		Choices:           choices,
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     hunyuanResponse.Response.Usage.PromptTokens,
 			CompletionTokens: hunyuanResponse.Response.Usage.CompletionTokens,
 			TotalTokens:      hunyuanResponse.Response.Usage.TotalTokens,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/minimax.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/minimax.go
@@ -36,8 +36,7 @@ const (
 	defaultSenderName        string = "小明"
 )

-type minimaxProviderInitializer struct {
-}
+type minimaxProviderInitializer struct{}

 func (m *minimaxProviderInitializer) ValidateConfig(config *ProviderConfig) error {
 	// If using the chat completion Pro API, a group ID must be set.
@@ -368,7 +367,7 @@ func (m *minimaxProvider) responseProToOpenAI(response *minimaxChatCompletionPro
 				Content: message.Text,
 			}
 			choices = append(choices, chatCompletionChoice{
-				FinishReason: choice.FinishReason,
+				FinishReason: util.Ptr(choice.FinishReason),
 				Index:        messageIndex,
 				Message:      message,
 			})
@@ -381,7 +380,7 @@ func (m *minimaxProvider) responseProToOpenAI(response *minimaxChatCompletionPro
 		Created: response.Created,
 		Model:   response.Model,
 		Choices: choices,
-		Usage: usage{
+		Usage: &usage{
 			TotalTokens:      int(response.Usage.TotalTokens),
 			PromptTokens:     int(response.Usage.PromptTokens),
 			CompletionTokens: int(response.Usage.CompletionTokens),
--- a/plugins/wasm-go/extensions/ai-proxy/provider/model.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/model.go
@@ -138,15 +138,15 @@ type chatCompletionResponse struct {
 	ServiceTier       string                 `json:"service_tier,omitempty"`
 	SystemFingerprint string                 `json:"system_fingerprint,omitempty"`
 	Object            string                 `json:"object,omitempty"`
-	Usage             usage                  `json:"usage,omitempty"`
+	Usage             *usage                 `json:"usage"`
 }

 type chatCompletionChoice struct {
 	Index        int                    `json:"index"`
 	Message      *chatMessage           `json:"message,omitempty"`
 	Delta        *chatMessage           `json:"delta,omitempty"`
-	FinishReason string                 `json:"finish_reason,omitempty"`
-	Logprobs     map[string]interface{} `json:"logprobs,omitempty"`
+	FinishReason *string                `json:"finish_reason"`
+	Logprobs     map[string]interface{} `json:"logprobs"`
 }

 type usage struct {
--- a/plugins/wasm-go/extensions/ai-proxy/provider/qwen.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/qwen.go
@@ -302,7 +302,7 @@ func (m *qwenProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, qwen
 		message := qwenMessageToChatMessage(qwenChoice.Message, m.config.reasoningContentMode)
 		choices = append(choices, chatCompletionChoice{
 			Message:      &message,
-			FinishReason: qwenChoice.FinishReason,
+			FinishReason: util.Ptr(qwenChoice.FinishReason),
 		})
 	}
 	return &chatCompletionResponse{
@@ -312,7 +312,7 @@ func (m *qwenProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, qwen
 		SystemFingerprint: "",
 		Object:            objectChatCompletion,
 		Choices:           choices,
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     qwenResponse.Usage.InputTokens,
 			CompletionTokens: qwenResponse.Usage.OutputTokens,
 			TotalTokens:      qwenResponse.Usage.TotalTokens,
@@ -413,11 +413,11 @@ func (m *qwenProvider) buildChatCompletionStreamingResponse(ctx wrapper.HttpCont

 	if finished {
 		finishResponse := *&baseMessage
-		finishResponse.Choices = append(finishResponse.Choices, chatCompletionChoice{Delta: &chatMessage{}, FinishReason: qwenChoice.FinishReason})
+		finishResponse.Choices = append(finishResponse.Choices, chatCompletionChoice{Delta: &chatMessage{}, FinishReason: util.Ptr(qwenChoice.FinishReason)})

 		usageResponse := *&baseMessage
 		usageResponse.Choices = []chatCompletionChoice{{Delta: &chatMessage{}}}
-		usageResponse.Usage = usage{
+		usageResponse.Usage = &usage{
 			PromptTokens:     qwenResponse.Usage.InputTokens,
 			CompletionTokens: qwenResponse.Usage.OutputTokens,
 			TotalTokens:      qwenResponse.Usage.TotalTokens,
--- a/plugins/wasm-go/extensions/ai-proxy/provider/spark.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/spark.go
@@ -150,7 +150,7 @@ func (p *sparkProvider) responseSpark2OpenAI(ctx wrapper.HttpContext, response *
 		Object:  objectChatCompletion,
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		Choices: choices,
-		Usage:   response.Usage,
+		Usage:   &response.Usage,
 	}
 }

@@ -168,7 +168,7 @@ func (p *sparkProvider) streamResponseSpark2OpenAI(ctx wrapper.HttpContext, resp
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		Object:  objectChatCompletion,
 		Choices: choices,
-		Usage:   response.Usage,
+		Usage:   &response.Usage,
 	}
 }

--- a/plugins/wasm-go/extensions/ai-proxy/provider/vertex.go
+++ b/plugins/wasm-go/extensions/ai-proxy/provider/vertex.go
@@ -32,8 +32,7 @@ const (
 	vertexEmbeddingAction            = "predict"
 )

-type vertexProviderInitializer struct {
-}
+type vertexProviderInitializer struct{}

 func (v *vertexProviderInitializer) ValidateConfig(config *ProviderConfig) error {
 	if config.vertexAuthKey == "" {
@@ -245,7 +244,7 @@ func (v *vertexProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, re
 		Created: time.Now().UnixMilli() / 1000,
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		Choices: make([]chatCompletionChoice, 0, len(response.Candidates)),
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     response.UsageMetadata.PromptTokenCount,
 			CompletionTokens: response.UsageMetadata.CandidatesTokenCount,
 			TotalTokens:      response.UsageMetadata.TotalTokenCount,
@@ -257,7 +256,7 @@ func (v *vertexProvider) buildChatCompletionResponse(ctx wrapper.HttpContext, re
 			Message: &chatMessage{
 				Role: roleAssistant,
 			},
-			FinishReason: candidate.FinishReason,
+			FinishReason: util.Ptr(candidate.FinishReason),
 		}
 		if len(candidate.Content.Parts) > 0 {
 			choice.Message.Content = candidate.Content.Parts[0].Text
@@ -310,7 +309,7 @@ func (v *vertexProvider) buildChatCompletionStreamResponse(ctx wrapper.HttpConte
 		Created: time.Now().UnixMilli() / 1000,
 		Model:   ctx.GetStringContext(ctxKeyFinalRequestModel, ""),
 		Choices: []chatCompletionChoice{choice},
-		Usage: usage{
+		Usage: &usage{
 			PromptTokens:     vertexResp.UsageMetadata.PromptTokenCount,
 			CompletionTokens: vertexResp.UsageMetadata.CandidatesTokenCount,
 			TotalTokens:      vertexResp.UsageMetadata.TotalTokenCount,
--- a/plugins/wasm-go/extensions/ai-proxy/util/ptr.go
+++ b/plugins/wasm-go/extensions/ai-proxy/util/ptr.go
@@ -0,0 +1,5 @@
+package util
+
+func Ptr[T any](v T) *T {
+	return &v
+}