Method createMessage

src/api/providers/qwen-code.ts:214–322

(
		systemPrompt: string,
		messages: Anthropic.Messages.MessageParam[],
		metadata?: ApiHandlerCreateMessageMetadata,
	)

	}

	override async *createMessage(
		systemPrompt: string,
		messages: Anthropic.Messages.MessageParam[],
		metadata?: ApiHandlerCreateMessageMetadata,
	): ApiStream {
		await this.ensureAuthenticated()
		const client = this.ensureClient()
		const model = this.getModel()

		const systemMessage: OpenAI.Chat.ChatCompletionSystemMessageParam = {
			role: "system",
			content: systemPrompt,
		}

		const convertedMessages = [systemMessage, ...convertToOpenAiMessages(messages)]

		const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
			model: model.id,
			temperature: 0,
			messages: convertedMessages,
			stream: true,
			stream_options: { include_usage: true },
			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))

		let fullContent = ""

		for await (const apiChunk of stream) {
			const delta = apiChunk.choices[0]?.delta ?? {}
			const finishReason = apiChunk.choices[0]?.finish_reason

			if (delta.content) {
				let newText = delta.content
				if (newText.startsWith(fullContent)) {
					newText = newText.substring(fullContent.length)
				}
				fullContent = delta.content

				if (newText) {
					// Check for thinking blocks
					if (newText.includes("<think>") || newText.includes("</think>")) {
						// Simple parsing for thinking blocks
						const parts = newText.split(/<\/?think>/g)
						for (let i = 0; i < parts.length; i++) {
							if (parts[i]) {
								if (i % 2 === 0) {
									// Outside thinking block
									yield {
										type: "text",
										text: parts[i],
									}
								} else {
									// Inside thinking block