Merge 91e27664a2 into eafed9705c

src/api/providers/__tests__/qwen-code-native-tools.spec.ts

@@ -89,19 +89,19 @@ describe("QwenCodeHandler Native Tools", () => {
 			})
 			await stream.next()
 
-			expect(mockCreate).toHaveBeenCalledWith(
+			const callArgs = mockCreate.mock.calls[0][0]
-				expect.objectContaining({
+			expect(callArgs.tools).toEqual(
-					tools: expect.arrayContaining([
+				expect.arrayContaining([
-						expect.objectContaining({
+					expect.objectContaining({
-							type: "function",
+						type: "function",
-							function: expect.objectContaining({
+						function: expect.objectContaining({
-								name: "test_tool",
+							name: "test_tool",
-							}),
 						}),
-					]),
+					}),
-					parallel_tool_calls: true,
+				]),
-				}),
 			)
+			// DashScope does not support parallel_tool_calls
+			expect(callArgs).not.toHaveProperty("parallel_tool_calls")
 		})
 
 		it("should include tool_choice when provided", async () => {
@@ -145,7 +145,8 @@ describe("QwenCodeHandler Native Tools", () => {
 			const callArgs = mockCreate.mock.calls[mockCreate.mock.calls.length - 1][0]
 			expect(callArgs).toHaveProperty("tools")
 			expect(callArgs).toHaveProperty("tool_choice")
-			expect(callArgs).toHaveProperty("parallel_tool_calls", true)
+			// DashScope does not support parallel_tool_calls
+			expect(callArgs).not.toHaveProperty("parallel_tool_calls")
 		})
 
 		it("should yield tool_call_partial chunks during streaming", async () => {
@@ -215,7 +216,7 @@ describe("QwenCodeHandler Native Tools", () => {
 			})
 		})
 
-		it("should set parallel_tool_calls based on metadata", async () => {
+		it("should not include parallel_tool_calls even when metadata provides it (DashScope unsupported)", async () => {
 			mockCreate.mockImplementationOnce(() => ({
 				[Symbol.asyncIterator]: async function* () {
 					yield {
@@ -231,11 +232,9 @@ describe("QwenCodeHandler Native Tools", () => {
 			})
 			await stream.next()
 
-			expect(mockCreate).toHaveBeenCalledWith(
+			const callArgs = mockCreate.mock.calls[0][0]
-				expect.objectContaining({
+			// DashScope does not support parallel_tool_calls - should never be sent
-					parallel_tool_calls: true,
+			expect(callArgs).not.toHaveProperty("parallel_tool_calls")
-				}),
-			)
 		})
 
 		it("should yield tool_call_end events when finish_reason is tool_calls", async () => {
@@ -370,4 +369,98 @@ describe("QwenCodeHandler Native Tools", () => {
 			expect(endChunks).toHaveLength(1)
 		})
 	})
+
+	describe("DashScope API Compatibility", () => {
+		it("should use max_tokens instead of max_completion_tokens", async () => {
+			mockCreate.mockImplementationOnce(() => ({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						choices: [{ delta: { content: "Test response" } }],
+					}
+				},
+			}))
+
+			const stream = handler.createMessage("test prompt", [], {
+				taskId: "test-task-id",
+			})
+			await stream.next()
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).toHaveProperty("max_tokens")
+			expect(callArgs).not.toHaveProperty("max_completion_tokens")
+		})
+
+		it("should not include stream_options", async () => {
+			mockCreate.mockImplementationOnce(() => ({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						choices: [{ delta: { content: "Test response" } }],
+					}
+				},
+			}))
+
+			const stream = handler.createMessage("test prompt", [], {
+				taskId: "test-task-id",
+			})
+			await stream.next()
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).not.toHaveProperty("stream_options")
+		})
+
+		it("should not include parallel_tool_calls", async () => {
+			mockCreate.mockImplementationOnce(() => ({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						choices: [{ delta: { content: "Test response" } }],
+					}
+				},
+			}))
+
+			const stream = handler.createMessage("test prompt", [], {
+				taskId: "test-task-id",
+				tools: testTools,
+			})
+			await stream.next()
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).not.toHaveProperty("parallel_tool_calls")
+		})
+
+		it("should not set strict: true on tool definitions", async () => {
+			mockCreate.mockImplementationOnce(() => ({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						choices: [{ delta: { content: "Test response" } }],
+					}
+				},
+			}))
+
+			const stream = handler.createMessage("test prompt", [], {
+				taskId: "test-task-id",
+				tools: testTools,
+			})
+			await stream.next()
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			const tools = callArgs.tools
+			expect(tools).toBeDefined()
+			for (const tool of tools) {
+				// DashScope does not support strict mode
+				expect(tool.function).not.toHaveProperty("strict")
+			}
+		})
+
+		it("should use max_tokens in completePrompt", async () => {
+			mockCreate.mockImplementationOnce(() => ({
+				choices: [{ message: { content: "response" } }],
+			}))
+
+			await handler.completePrompt("test prompt")
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).toHaveProperty("max_tokens")
+			expect(callArgs).not.toHaveProperty("max_completion_tokens")
+		})
+	})
 })

src/api/providers/qwen-code.ts

@@ -219,16 +219,19 @@ export class QwenCodeHandler extends BaseProvider implements SingleCompletionHan
 
 		const convertedMessages = [systemMessage, ...convertToOpenAiMessages(messages)]
 
+		// DashScope's OpenAI-compatible API does not support several OpenAI-specific
+		// parameters. Using them causes a 400 Bad Request error. Specifically:
+		// - max_completion_tokens -> use max_tokens instead
+		// - parallel_tool_calls -> not supported
+		// - stream_options -> not supported
 		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 			model: model.id,
 			temperature: 0,
 			messages: convertedMessages,
 			stream: true,
-			stream_options: { include_usage: true },
+			max_tokens: model.info.maxTokens,
-			max_completion_tokens: model.info.maxTokens,
 			tools: this.convertToolsForOpenAI(metadata?.tools),
 			tool_choice: metadata?.tool_choice,
-			parallel_tool_calls: metadata?.parallelToolCalls ?? true,
 		}
 
 		const stream = await this.callApiWithRetry(() => client.chat.completions.create(requestOptions))
@@ -321,6 +324,28 @@ export class QwenCodeHandler extends BaseProvider implements SingleCompletionHan
 		return { id, info }
 	}
 
+	/**
+	 * Override to skip strict mode for DashScope compatibility.
+	 * DashScope's OpenAI-compatible API does not support OpenAI's strict mode
+	 * on tool definitions (strict: true, additionalProperties: false enforcement).
+	 * Sending these causes a 400 Bad Request error.
+	 */
+	protected override convertToolsForOpenAI(tools: any[] | undefined): any[] | undefined {
+		if (!tools) {
+			return undefined
+		}
+
+		return tools
+			.filter((tool) => tool.type === "function")
+			.map((tool) => ({
+				...tool,
+				function: {
+					...tool.function,
+					// Do not set strict: true - DashScope does not support it
+				},
+			}))
+	}
+
 	async completePrompt(prompt: string): Promise<string> {
 		await this.ensureAuthenticated()
 		const client = this.ensureClient()
@@ -329,7 +354,7 @@ export class QwenCodeHandler extends BaseProvider implements SingleCompletionHan
 		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 			model: model.id,
 			messages: [{ role: "user", content: prompt }],
-			max_completion_tokens: model.info.maxTokens,
+			max_tokens: model.info.maxTokens,
 		}
 
 		const response = await this.callApiWithRetry(() => client.chat.completions.create(requestOptions))