Merge branch 'sashabaranov:master' into master

Author: fcce

.github/workflows/pr.yml

@@ -13,14 +13,14 @@ jobs:
     - name: Setup Go
       uses: actions/setup-go@v5
       with:
-        go-version: '1.21'
+        go-version: '1.24'
     - name: Run vet
       run: |
         go vet .
     - name: Run golangci-lint
-      uses: golangci/golangci-lint-action@v4
+      uses: golangci/golangci-lint-action@v6
       with:
-        version: latest
+        version: v1.64.5
     - name: Run tests
       run: go test -race -covermode=atomic -coverprofile=coverage.out -v .
     - name: Upload coverage reports to Codecov

.golangci.yml

@@ -139,11 +139,6 @@ linters-settings:
     # Default: false
     all: true
 
-  varcheck:
-    # Check usage of exported fields and variables.
-    # Default: false
-    exported-fields: false # default false # TODO: enable after fixing false positives
-
 
 linters:
   disable-all: true
@@ -167,9 +162,7 @@ linters:
     - durationcheck # check for two durations multiplied together
     - errname # Checks that sentinel errors are prefixed with the Err and error types are suffixed with the Error.
     - errorlint # errorlint is a linter for that can be used to find code that will cause problems with the error wrapping scheme introduced in Go 1.13.
-    # Removed execinquery (deprecated). execinquery is a linter about query string checker in Query function which reads your Go src files and warning it finds
     - exhaustive # check exhaustiveness of enum switch statements
-    - exportloopref # checks for pointers to enclosing loop variables
     - forbidigo # Forbids identifiers
     - funlen # Tool for detection of long functions
     # - gochecknoglobals # check that no global variables exist
@@ -201,11 +194,11 @@ linters:
     - rowserrcheck # checks whether Err of rows is checked successfully
     - sqlclosecheck # Checks that sql.Rows and sql.Stmt are closed.
     - stylecheck # Stylecheck is a replacement for golint
-    - tenv # tenv is analyzer that detects using os.Setenv instead of t.Setenv since Go1.17
     - testpackage # linter that makes you use a separate _test package
     - tparallel # tparallel detects inappropriate usage of t.Parallel() method in your Go test codes
     - unconvert # Remove unnecessary type conversions
     - unparam # Reports unused function parameters
+    - usetesting # Reports uses of functions with replacement inside the testing package
     - wastedassign # wastedassign finds wasted assignment statements.
     - whitespace # Tool for detection of leading and trailing whitespace
     ## you may want to enable
@@ -238,12 +231,6 @@ linters:
     #- tagliatelle # Checks the struct tags.
     #- thelper # thelper detects golang test helpers without t.Helper() call and checks the consistency of test helpers
     #- wsl # [too strict and mostly code is not more readable] Whitespace Linter - Forces you to use empty lines!
-    ## deprecated
-    #- exhaustivestruct # [deprecated, replaced by exhaustruct] Checks if all struct's fields are initialized
-    #- golint # [deprecated, replaced by revive] Golint differs from gofmt. Gofmt reformats Go source code, whereas golint prints out style mistakes
-    #- interfacer # [deprecated] Linter that suggests narrower interface types
-    #- maligned # [deprecated, replaced by govet fieldalignment] Tool to detect Go structs that would take less memory if their fields were sorted
-    #- scopelint # [deprecated, replaced by exportloopref] Scopelint checks for unpinned variables in go programs
 
 
 issues:

audio_api_test.go

@@ -40,12 +40,9 @@ func TestAudio(t *testing.T) {
 
 	ctx := context.Background()
 
-	dir, cleanup := test.CreateTestDirectory(t)
-	defer cleanup()
-
 	for _, tc := range testcases {
 		t.Run(tc.name, func(t *testing.T) {
-			path := filepath.Join(dir, "fake.mp3")
+			path := filepath.Join(t.TempDir(), "fake.mp3")
 			test.CreateTestFile(t, path)
 
 			req := openai.AudioRequest{
@@ -90,12 +87,9 @@ func TestAudioWithOptionalArgs(t *testing.T) {
 
 	ctx := context.Background()
 
-	dir, cleanup := test.CreateTestDirectory(t)
-	defer cleanup()
-
 	for _, tc := range testcases {
 		t.Run(tc.name, func(t *testing.T) {
-			path := filepath.Join(dir, "fake.mp3")
+			path := filepath.Join(t.TempDir(), "fake.mp3")
 			test.CreateTestFile(t, path)
 
 			req := openai.AudioRequest{

audio_test.go

@@ -13,9 +13,7 @@ import (
 )
 
 func TestAudioWithFailingFormBuilder(t *testing.T) {
-	dir, cleanup := test.CreateTestDirectory(t)
-	defer cleanup()
-	path := filepath.Join(dir, "fake.mp3")
+	path := filepath.Join(t.TempDir(), "fake.mp3")
 	test.CreateTestFile(t, path)
 
 	req := AudioRequest{
@@ -63,9 +61,7 @@ func TestAudioWithFailingFormBuilder(t *testing.T) {
 
 func TestCreateFileField(t *testing.T) {
 	t.Run("createFileField failing file", func(t *testing.T) {
-		dir, cleanup := test.CreateTestDirectory(t)
-		defer cleanup()
-		path := filepath.Join(dir, "fake.mp3")
+		path := filepath.Join(t.TempDir(), "fake.mp3")
 		test.CreateTestFile(t, path)
 
 		req := AudioRequest{

chat.go

@@ -14,6 +14,7 @@ const (
 	ChatMessageRoleAssistant = "assistant"
 	ChatMessageRoleFunction  = "function"
 	ChatMessageRoleTool      = "tool"
+	ChatMessageRoleDeveloper = "developer"
 )
 
 const chatCompletionsSuffix = "/chat/completions"
@@ -93,7 +94,7 @@ type ChatMessagePart struct {
 
 type ChatCompletionMessage struct {
 	Role         string `json:"role"`
-	Content      string `json:"content"`
+	Content      string `json:"content,omitempty"`
 	Refusal      string `json:"refusal,omitempty"`
 	MultiContent []ChatMessagePart
 
@@ -132,7 +133,7 @@ func (m ChatCompletionMessage) MarshalJSON() ([]byte, error) {
 
 	msg := struct {
 		Role         string            `json:"role"`
-		Content      string            `json:"content"`
+		Content      string            `json:"content,omitempty"`
 		Refusal      string            `json:"refusal,omitempty"`
 		MultiContent []ChatMessagePart `json:"-"`
 		Name         string            `json:"name,omitempty"`
@@ -146,7 +147,7 @@ func (m ChatCompletionMessage) MarshalJSON() ([]byte, error) {
 func (m *ChatCompletionMessage) UnmarshalJSON(bs []byte) error {
 	msg := struct {
 		Role         string `json:"role"`
-		Content      string `json:"content"`
+		Content      string `json:"content,omitempty"`
 		Refusal      string `json:"refusal,omitempty"`
 		MultiContent []ChatMessagePart
 		Name         string        `json:"name,omitempty"`
@@ -258,6 +259,8 @@ type ChatCompletionRequest struct {
 	// Store can be set to true to store the output of this completion request for use in distillations and evals.
 	// https://platform.openai.com/docs/api-reference/chat/create#chat-create-store
 	Store bool `json:"store,omitempty"`
+	// Controls effort on reasoning for reasoning models. It can be set to "low", "medium", or "high".
+	ReasoningEffort string `json:"reasoning_effort,omitempty"`
 	// Metadata to store with the completion.
 	Metadata map[string]string `json:"metadata,omitempty"`
 }
@@ -390,7 +393,8 @@ func (c *Client) CreateChatCompletion(
 		return
 	}
 
-	if err = validateRequestForO1Models(request); err != nil {
+	reasoningValidator := NewReasoningValidator()
+	if err = reasoningValidator.Validate(request); err != nil {
 		return
 	}
 

chat_stream.go

@@ -80,7 +80,8 @@ func (c *Client) CreateChatCompletionStream(
 	}
 
 	request.Stream = true
-	if err = validateRequestForO1Models(request); err != nil {
+	reasoningValidator := NewReasoningValidator()
+	if err = reasoningValidator.Validate(request); err != nil {
 		return
 	}
 

chat_stream_test.go

@@ -792,6 +792,173 @@ func compareChatResponses(r1, r2 openai.ChatCompletionStreamResponse) bool {
 	return true
 }
 
+func TestCreateChatCompletionStreamWithReasoningModel(t *testing.T) {
+	client, server, teardown := setupOpenAITestServer()
+	defer teardown()
+	server.RegisterHandler("/v1/chat/completions", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+
+		dataBytes := []byte{}
+
+		//nolint:lll
+		dataBytes = append(dataBytes, []byte(`data: {"id":"1","object":"chat.completion.chunk","created":1729585728,"model":"o3-mini-2025-01-31","system_fingerprint":"fp_mini","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":null}]}`)...)
+		dataBytes = append(dataBytes, []byte("\n\n")...)
+
+		//nolint:lll
+		dataBytes = append(dataBytes, []byte(`data: {"id":"2","object":"chat.completion.chunk","created":1729585728,"model":"o3-mini-2025-01-31","system_fingerprint":"fp_mini","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}`)...)
+		dataBytes = append(dataBytes, []byte("\n\n")...)
+
+		//nolint:lll
+		dataBytes = append(dataBytes, []byte(`data: {"id":"3","object":"chat.completion.chunk","created":1729585728,"model":"o3-mini-2025-01-31","system_fingerprint":"fp_mini","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}`)...)
+		dataBytes = append(dataBytes, []byte("\n\n")...)
+
+		//nolint:lll
+		dataBytes = append(dataBytes, []byte(`data: {"id":"4","object":"chat.completion.chunk","created":1729585728,"model":"o3-mini-2025-01-31","system_fingerprint":"fp_mini","choices":[{"index":0,"delta":{"content":" O3Mini"},"finish_reason":null}]}`)...)
+		dataBytes = append(dataBytes, []byte("\n\n")...)
+
+		//nolint:lll
+		dataBytes = append(dataBytes, []byte(`data: {"id":"5","object":"chat.completion.chunk","created":1729585728,"model":"o3-mini-2025-01-31","system_fingerprint":"fp_mini","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}`)...)
+		dataBytes = append(dataBytes, []byte("\n\n")...)
+
+		dataBytes = append(dataBytes, []byte("data: [DONE]\n\n")...)
+
+		_, err := w.Write(dataBytes)
+		checks.NoError(t, err, "Write error")
+	})
+
+	stream, err := client.CreateChatCompletionStream(context.Background(), openai.ChatCompletionRequest{
+		MaxCompletionTokens: 2000,
+		Model:               openai.O3Mini20250131,
+		Messages: []openai.ChatCompletionMessage{
+			{
+				Role:    openai.ChatMessageRoleUser,
+				Content: "Hello!",
+			},
+		},
+		Stream: true,
+	})
+	checks.NoError(t, err, "CreateCompletionStream returned error")
+	defer stream.Close()
+
+	expectedResponses := []openai.ChatCompletionStreamResponse{
+		{
+			ID:                "1",
+			Object:            "chat.completion.chunk",
+			Created:           1729585728,
+			Model:             openai.O3Mini20250131,
+			SystemFingerprint: "fp_mini",
+			Choices: []openai.ChatCompletionStreamChoice{
+				{
+					Index: 0,
+					Delta: openai.ChatCompletionStreamChoiceDelta{
+						Role: "assistant",
+					},
+				},
+			},
+		},
+		{
+			ID:                "2",
+			Object:            "chat.completion.chunk",
+			Created:           1729585728,
+			Model:             openai.O3Mini20250131,
+			SystemFingerprint: "fp_mini",
+			Choices: []openai.ChatCompletionStreamChoice{
+				{
+					Index: 0,
+					Delta: openai.ChatCompletionStreamChoiceDelta{
+						Content: "Hello",
+					},
+				},
+			},
+		},
+		{
+			ID:                "3",
+			Object:            "chat.completion.chunk",
+			Created:           1729585728,
+			Model:             openai.O3Mini20250131,
+			SystemFingerprint: "fp_mini",
+			Choices: []openai.ChatCompletionStreamChoice{
+				{
+					Index: 0,
+					Delta: openai.ChatCompletionStreamChoiceDelta{
+						Content: " from",
+					},
+				},
+			},
+		},
+		{
+			ID:                "4",
+			Object:            "chat.completion.chunk",
+			Created:           1729585728,
+			Model:             openai.O3Mini20250131,
+			SystemFingerprint: "fp_mini",
+			Choices: []openai.ChatCompletionStreamChoice{
+				{
+					Index: 0,
+					Delta: openai.ChatCompletionStreamChoiceDelta{
+						Content: " O3Mini",
+					},
+				},
+			},
+		},
+		{
+			ID:                "5",
+			Object:            "chat.completion.chunk",
+			Created:           1729585728,
+			Model:             openai.O3Mini20250131,
+			SystemFingerprint: "fp_mini",
+			Choices: []openai.ChatCompletionStreamChoice{
+				{
+					Index:        0,
+					Delta:        openai.ChatCompletionStreamChoiceDelta{},
+					FinishReason: "stop",
+				},
+			},
+		},
+	}
+
+	for ix, expectedResponse := range expectedResponses {
+		b, _ := json.Marshal(expectedResponse)
+		t.Logf("%d: %s", ix, string(b))
+
+		receivedResponse, streamErr := stream.Recv()
+		checks.NoError(t, streamErr, "stream.Recv() failed")
+		if !compareChatResponses(expectedResponse, receivedResponse) {
+			t.Errorf("Stream response %v is %v, expected %v", ix, receivedResponse, expectedResponse)
+		}
+	}
+
+	_, streamErr := stream.Recv()
+	if !errors.Is(streamErr, io.EOF) {
+		t.Errorf("stream.Recv() did not return EOF in the end: %v", streamErr)
+	}
+}
+
+func TestCreateChatCompletionStreamReasoningValidatorFails(t *testing.T) {
+	client, _, _ := setupOpenAITestServer()
+
+	stream, err := client.CreateChatCompletionStream(context.Background(), openai.ChatCompletionRequest{
+		MaxTokens: 100, // This will trigger the validator to fail
+		Model:     openai.O3Mini,
+		Messages: []openai.ChatCompletionMessage{
+			{
+				Role:    openai.ChatMessageRoleUser,
+				Content: "Hello!",
+			},
+		},
+		Stream: true,
+	})
+
+	if stream != nil {
+		t.Error("Expected nil stream when validation fails")
+		stream.Close()
+	}
+
+	if !errors.Is(err, openai.ErrReasoningModelMaxTokensDeprecated) {
+		t.Errorf("Expected ErrReasoningModelMaxTokensDeprecated, got: %v", err)
+	}
+}
+
 func compareChatStreamResponseChoices(c1, c2 openai.ChatCompletionStreamChoice) bool {
 	if c1.Index != c2.Index {
 		return false