Handle <think> tags in the base OpenAI-compatible provider (#8989)

This commit is contained in:
Matt Rubens 2025-11-02 22:22:49 -05:00 committed by GitHub
parent 8d98b303b3
commit 613255c09a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 303 additions and 83 deletions

View file

@ -0,0 +1,286 @@
// npx vitest run api/providers/__tests__/base-openai-compatible-provider.spec.ts
import { Anthropic } from "@anthropic-ai/sdk"
import OpenAI from "openai"
import type { ModelInfo } from "@roo-code/types"
import { BaseOpenAiCompatibleProvider } from "../base-openai-compatible-provider"
// Create mock functions
const mockCreate = vi.fn()
// Mock OpenAI module
vi.mock("openai", () => ({
default: vi.fn(() => ({
chat: {
completions: {
create: mockCreate,
},
},
})),
}))
// Create a concrete test implementation of the abstract base class
class TestOpenAiCompatibleProvider extends BaseOpenAiCompatibleProvider<"test-model"> {
constructor(apiKey: string) {
const testModels: Record<"test-model", ModelInfo> = {
"test-model": {
maxTokens: 4096,
contextWindow: 128000,
supportsImages: false,
supportsPromptCache: false,
inputPrice: 0.5,
outputPrice: 1.5,
},
}
super({
providerName: "TestProvider",
baseURL: "https://test.example.com/v1",
defaultProviderModelId: "test-model",
providerModels: testModels,
apiKey,
})
}
}
describe("BaseOpenAiCompatibleProvider", () => {
let handler: TestOpenAiCompatibleProvider
beforeEach(() => {
vi.clearAllMocks()
handler = new TestOpenAiCompatibleProvider("test-api-key")
})
afterEach(() => {
vi.restoreAllMocks()
})
describe("XmlMatcher reasoning tags", () => {
it("should handle reasoning tags (<think>) from stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "<think>Let me think" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " about this</think>" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "The answer is 42" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
// XmlMatcher yields chunks as they're processed
expect(chunks).toEqual([
{ type: "reasoning", text: "Let me think" },
{ type: "reasoning", text: " about this" },
{ type: "text", text: "The answer is 42" },
])
})
it("should handle complete <think> tag in a single chunk", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "Regular text before " } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "<think>Complete thought</think>" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " regular text after" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
// When a complete tag arrives in one chunk, XmlMatcher may not parse it
// This test documents the actual behavior
expect(chunks.length).toBeGreaterThan(0)
expect(chunks[0]).toEqual({ type: "text", text: "Regular text before " })
})
it("should handle incomplete <think> tag at end of stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "<think>Incomplete thought" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
// XmlMatcher should handle incomplete tags and flush remaining content
expect(chunks.length).toBeGreaterThan(0)
expect(
chunks.some(
(c) => (c.type === "text" || c.type === "reasoning") && c.text.includes("Incomplete thought"),
),
).toBe(true)
})
it("should handle text without any <think> tags", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "Just regular text" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " without reasoning" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
expect(chunks).toEqual([
{ type: "text", text: "Just regular text" },
{ type: "text", text: " without reasoning" },
])
})
it("should handle <think> tags that start at beginning of stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "<think>reasoning" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " content</think>" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " normal text" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
expect(chunks).toEqual([
{ type: "reasoning", text: "reasoning" },
{ type: "reasoning", text: " content" },
{ type: "text", text: " normal text" },
])
})
})
describe("Basic functionality", () => {
it("should create stream with correct parameters", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
async next() {
return { done: true }
},
}),
}
})
const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]
const messageGenerator = handler.createMessage(systemPrompt, messages)
await messageGenerator.next()
expect(mockCreate).toHaveBeenCalledWith(
expect.objectContaining({
model: "test-model",
temperature: 0,
messages: expect.arrayContaining([{ role: "system", content: systemPrompt }]),
stream: true,
stream_options: { include_usage: true },
}),
undefined,
)
})
it("should yield usage data from stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vi
.fn()
.mockResolvedValueOnce({
done: false,
value: {
choices: [{ delta: {} }],
usage: { prompt_tokens: 100, completion_tokens: 50 },
},
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const firstChunk = await stream.next()
expect(firstChunk.done).toBe(false)
expect(firstChunk.value).toEqual({ type: "usage", inputTokens: 100, outputTokens: 50 })
})
})
})

View file

@ -178,43 +178,6 @@ describe("MiniMaxHandler", () => {
expect(firstChunk.value).toEqual({ type: "text", text: testContent })
})
it("should handle reasoning tags (<think>) from stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {
[Symbol.asyncIterator]: () => ({
next: vitest
.fn()
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "<think>Let me think" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: " about this</think>" } }] },
})
.mockResolvedValueOnce({
done: false,
value: { choices: [{ delta: { content: "The answer is 42" } }] },
})
.mockResolvedValueOnce({ done: true }),
}),
}
})
const stream = handler.createMessage("system prompt", [])
const chunks = []
for await (const chunk of stream) {
chunks.push(chunk)
}
// XmlMatcher yields chunks as they're processed
expect(chunks).toEqual([
{ type: "reasoning", text: "Let me think" },
{ type: "reasoning", text: " about this" },
{ type: "text", text: "The answer is 42" },
])
})
it("createMessage should yield usage data from stream", async () => {
mockCreate.mockImplementationOnce(() => {
return {

View file

@ -4,6 +4,7 @@ import OpenAI from "openai"
import type { ModelInfo } from "@roo-code/types"
import { type ApiHandlerOptions, getModelMaxOutputTokens } from "../../shared/api"
import { XmlMatcher } from "../../utils/xml-matcher"
import { ApiStream } from "../transform/stream"
import { convertToOpenAiMessages } from "../transform/openai-format"
@ -105,13 +106,21 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
): ApiStream {
const stream = await this.createStream(systemPrompt, messages, metadata)
const matcher = new XmlMatcher(
"think",
(chunk) =>
({
type: chunk.matched ? "reasoning" : "text",
text: chunk.data,
}) as const,
)
for await (const chunk of stream) {
const delta = chunk.choices[0]?.delta
if (delta?.content) {
yield {
type: "text",
text: delta.content,
for (const processedChunk of matcher.update(delta.content)) {
yield processedChunk
}
}
@ -127,6 +136,11 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
}
}
}
// Process any remaining content
for (const processedChunk of matcher.final()) {
yield processedChunk
}
}
async completePrompt(prompt: string): Promise<string> {

View file

@ -1,10 +1,6 @@
import { Anthropic } from "@anthropic-ai/sdk"
import { type MinimaxModelId, minimaxDefaultModelId, minimaxModels } from "@roo-code/types"
import type { ApiHandlerOptions } from "../../shared/api"
import { XmlMatcher } from "../../utils/xml-matcher"
import { ApiStream } from "../transform/stream"
import type { ApiHandlerCreateMessageMetadata } from "../index"
import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
@ -20,43 +16,4 @@ export class MiniMaxHandler extends BaseOpenAiCompatibleProvider<MinimaxModelId>
defaultTemperature: 1.0,
})
}
override async *createMessage(
systemPrompt: string,
messages: Anthropic.Messages.MessageParam[],
metadata?: ApiHandlerCreateMessageMetadata,
): ApiStream {
const stream = await this.createStream(systemPrompt, messages, metadata)
const matcher = new XmlMatcher(
"think",
(chunk) =>
({
type: chunk.matched ? "reasoning" : "text",
text: chunk.data,
}) as const,
)
for await (const chunk of stream) {
const delta = chunk.choices[0]?.delta
if (delta?.content) {
for (const matcherChunk of matcher.update(delta.content)) {
yield matcherChunk
}
}
if (chunk.usage) {
yield {
type: "usage",
inputTokens: chunk.usage.prompt_tokens || 0,
outputTokens: chunk.usage.completion_tokens || 0,
}
}
}
for (const chunk of matcher.final()) {
yield chunk
}
}
}