Add support for reasoning_content field in chat completion messages for DeepSeek R1

chat.go

@@ -104,6 +104,12 @@ type ChatCompletionMessage struct {
 	// - https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
 	Name string `json:"name,omitempty"`
 
+	// This property is used for the "reasoning" feature supported by deepseek-reasoner
+	// which is not in the official documentation.
+	// the doc from deepseek:
+	// - https://api-docs.deepseek.com/api/create-chat-completion#responses
+	ReasoningContent string `json:"reasoning_content,omitempty"`
+
 	FunctionCall *FunctionCall `json:"function_call,omitempty"`
 
 	// For Role=assistant prompts this may be set to the tool calls generated by the model, such as function calls.
@@ -119,56 +125,60 @@ func (m ChatCompletionMessage) MarshalJSON() ([]byte, error) {
 	}
 	if len(m.MultiContent) > 0 {
 		msg := struct {
-			Role         string            `json:"role"`
-			Content      string            `json:"-"`
-			Refusal      string            `json:"refusal,omitempty"`
-			MultiContent []ChatMessagePart `json:"content,omitempty"`
-			Name         string            `json:"name,omitempty"`
-			FunctionCall *FunctionCall     `json:"function_call,omitempty"`
-			ToolCalls    []ToolCall        `json:"tool_calls,omitempty"`
-			ToolCallID   string            `json:"tool_call_id,omitempty"`
+			Role             string            `json:"role"`
+			Content          string            `json:"-"`
+			Refusal          string            `json:"refusal,omitempty"`
+			MultiContent     []ChatMessagePart `json:"content,omitempty"`
+			Name             string            `json:"name,omitempty"`
+			ReasoningContent string            `json:"reasoning_content,omitempty"`
+			FunctionCall     *FunctionCall     `json:"function_call,omitempty"`
+			ToolCalls        []ToolCall        `json:"tool_calls,omitempty"`
+			ToolCallID       string            `json:"tool_call_id,omitempty"`
 		}(m)
 		return json.Marshal(msg)
 	}
 
 	msg := struct {
-		Role         string            `json:"role"`
-		Content      string            `json:"content,omitempty"`
-		Refusal      string            `json:"refusal,omitempty"`
-		MultiContent []ChatMessagePart `json:"-"`
-		Name         string            `json:"name,omitempty"`
-		FunctionCall *FunctionCall     `json:"function_call,omitempty"`
-		ToolCalls    []ToolCall        `json:"tool_calls,omitempty"`
-		ToolCallID   string            `json:"tool_call_id,omitempty"`
+		Role             string            `json:"role"`
+		Content          string            `json:"content,omitempty"`
+		Refusal          string            `json:"refusal,omitempty"`
+		MultiContent     []ChatMessagePart `json:"-"`
+		Name             string            `json:"name,omitempty"`
+		ReasoningContent string            `json:"reasoning_content,omitempty"`
+		FunctionCall     *FunctionCall     `json:"function_call,omitempty"`
+		ToolCalls        []ToolCall        `json:"tool_calls,omitempty"`
+		ToolCallID       string            `json:"tool_call_id,omitempty"`
 	}(m)
 	return json.Marshal(msg)
 }
 
 func (m *ChatCompletionMessage) UnmarshalJSON(bs []byte) error {
 	msg := struct {
-		Role         string `json:"role"`
-		Content      string `json:"content,omitempty"`
-		Refusal      string `json:"refusal,omitempty"`
-		MultiContent []ChatMessagePart
-		Name         string        `json:"name,omitempty"`
-		FunctionCall *FunctionCall `json:"function_call,omitempty"`
-		ToolCalls    []ToolCall    `json:"tool_calls,omitempty"`
-		ToolCallID   string        `json:"tool_call_id,omitempty"`
+		Role             string `json:"role"`
+		Content          string `json:"content"`
+		Refusal          string `json:"refusal,omitempty"`
+		MultiContent     []ChatMessagePart
+		Name             string        `json:"name,omitempty"`
+		ReasoningContent string        `json:"reasoning_content,omitempty"`
+		FunctionCall     *FunctionCall `json:"function_call,omitempty"`
+		ToolCalls        []ToolCall    `json:"tool_calls,omitempty"`
+		ToolCallID       string        `json:"tool_call_id,omitempty"`
 	}{}
 
 	if err := json.Unmarshal(bs, &msg); err == nil {
 		*m = ChatCompletionMessage(msg)
 		return nil
 	}
 	multiMsg := struct {
-		Role         string `json:"role"`
-		Content      string
-		Refusal      string            `json:"refusal,omitempty"`
-		MultiContent []ChatMessagePart `json:"content"`
-		Name         string            `json:"name,omitempty"`
-		FunctionCall *FunctionCall     `json:"function_call,omitempty"`
-		ToolCalls    []ToolCall        `json:"tool_calls,omitempty"`
-		ToolCallID   string            `json:"tool_call_id,omitempty"`
+		Role             string `json:"role"`
+		Content          string
+		Refusal          string            `json:"refusal,omitempty"`
+		MultiContent     []ChatMessagePart `json:"content"`
+		Name             string            `json:"name,omitempty"`
+		ReasoningContent string            `json:"reasoning_content,omitempty"`
+		FunctionCall     *FunctionCall     `json:"function_call,omitempty"`
+		ToolCalls        []ToolCall        `json:"tool_calls,omitempty"`
+		ToolCallID       string            `json:"tool_call_id,omitempty"`
 	}{}
 	if err := json.Unmarshal(bs, &multiMsg); err != nil {
 		return err

chat_stream.go

@@ -11,6 +11,12 @@ type ChatCompletionStreamChoiceDelta struct {
 	FunctionCall *FunctionCall `json:"function_call,omitempty"`
 	ToolCalls    []ToolCall    `json:"tool_calls,omitempty"`
 	Refusal      string        `json:"refusal,omitempty"`
+
+	// This property is used for the "reasoning" feature supported by deepseek-reasoner
+	// which is not in the official documentation.
+	// the doc from deepseek:
+	// - https://api-docs.deepseek.com/api/create-chat-completion#responses
+	ReasoningContent string `json:"reasoning_content,omitempty"`
 }
 
 type ChatCompletionStreamChoiceLogprobs struct {

chat_test.go

@@ -411,6 +411,23 @@ func TestO3ModelChatCompletions(t *testing.T) {
 	checks.NoError(t, err, "CreateChatCompletion error")
 }
 
+func TestDeepseekR1ModelChatCompletions(t *testing.T) {
+	client, server, teardown := setupOpenAITestServer()
+	defer teardown()
+	server.RegisterHandler("/v1/chat/completions", handleDeepseekR1ChatCompletionEndpoint)
+	_, err := client.CreateChatCompletion(context.Background(), openai.ChatCompletionRequest{
+		Model:               "deepseek-reasoner",
+		MaxCompletionTokens: 100,
+		Messages: []openai.ChatCompletionMessage{
+			{
+				Role:    openai.ChatMessageRoleUser,
+				Content: "Hello!",
+			},
+		},
+	})
+	checks.NoError(t, err, "CreateChatCompletion error")
+}
+
 // TestCompletions Tests the completions endpoint of the API using the mocked server.
 func TestChatCompletionsWithHeaders(t *testing.T) {
 	client, server, teardown := setupOpenAITestServer()
@@ -822,6 +839,68 @@ func handleChatCompletionEndpoint(w http.ResponseWriter, r *http.Request) {
 	fmt.Fprintln(w, string(resBytes))
 }
 
+func handleDeepseekR1ChatCompletionEndpoint(w http.ResponseWriter, r *http.Request) {
+	var err error
+	var resBytes []byte
+
+	// completions only accepts POST requests
+	if r.Method != "POST" {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+	}
+	var completionReq openai.ChatCompletionRequest
+	if completionReq, err = getChatCompletionBody(r); err != nil {
+		http.Error(w, "could not read request", http.StatusInternalServerError)
+		return
+	}
+	res := openai.ChatCompletionResponse{
+		ID:      strconv.Itoa(int(time.Now().Unix())),
+		Object:  "test-object",
+		Created: time.Now().Unix(),
+		// would be nice to validate Model during testing, but
+		// this may not be possible with how much upkeep
+		// would be required / wouldn't make much sense
+		Model: completionReq.Model,
+	}
+	// create completions
+	n := completionReq.N
+	if n == 0 {
+		n = 1
+	}
+	if completionReq.MaxCompletionTokens == 0 {
+		completionReq.MaxCompletionTokens = 1000
+	}
+	for i := 0; i < n; i++ {
+		reasoningContent := "User says hello! And I need to reply"
+		completionStr := strings.Repeat("a", completionReq.MaxCompletionTokens-numTokens(reasoningContent))
+		res.Choices = append(res.Choices, openai.ChatCompletionChoice{
+			Message: openai.ChatCompletionMessage{
+				Role:             openai.ChatMessageRoleAssistant,
+				ReasoningContent: reasoningContent,
+				Content:          completionStr,
+			},
+			Index: i,
+		})
+	}
+	inputTokens := numTokens(completionReq.Messages[0].Content) * n
+	completionTokens := completionReq.MaxTokens * n
+	res.Usage = openai.Usage{
+		PromptTokens:     inputTokens,
+		CompletionTokens: completionTokens,
+		TotalTokens:      inputTokens + completionTokens,
+	}
+	resBytes, _ = json.Marshal(res)
+	w.Header().Set(xCustomHeader, xCustomHeaderValue)
+	for k, v := range rateLimitHeaders {
+		switch val := v.(type) {
+		case int:
+			w.Header().Set(k, strconv.Itoa(val))
+		default:
+			w.Header().Set(k, fmt.Sprintf("%s", v))
+		}
+	}
+	fmt.Fprintln(w, string(resBytes))
+}
+
 // getChatCompletionBody Returns the body of the request to create a completion.
 func getChatCompletionBody(r *http.Request) (openai.ChatCompletionRequest, error) {
 	completion := openai.ChatCompletionRequest{}

openai_test.go

@@ -29,7 +29,7 @@ func setupAzureTestServer() (client *openai.Client, server *test.ServerTest, tea
 
 // numTokens Returns the number of GPT-3 encoded tokens in the given text.
 // This function approximates based on the rule of thumb stated by OpenAI:
-// https://beta.openai.com/tokenizer/
+// https://beta.openai.com/tokenizer.
 //
 // TODO: implement an actual tokenizer for GPT-3 and Codex (once available).
 func numTokens(s string) int {