Skip to content

feat: add reasoning effort support to Chutes provider models #8905

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed

feat: add reasoning effort support to Chutes provider models #8905

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions packages/types/src/providers/chutes.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,8 @@ export const chutesModels = {
contextWindow: 163840,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description: "DeepSeek R1 0528 model.",
Expand All @@ -60,6 +62,8 @@ export const chutesModels = {
contextWindow: 163840,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description: "DeepSeek R1 model.",
Expand Down Expand Up @@ -207,6 +211,8 @@ export const chutesModels = {
contextWindow: 163840,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description: "DeepSeek R1 Zero model.",
Expand Down Expand Up @@ -288,6 +294,8 @@ export const chutesModels = {
contextWindow: 163840,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description: "TNGTech DeepSeek R1T Chimera model.",
Expand Down Expand Up @@ -345,6 +353,8 @@ export const chutesModels = {
contextWindow: 128000,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description:
Expand Down Expand Up @@ -382,6 +392,8 @@ export const chutesModels = {
contextWindow: 262144,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0.077968332,
outputPrice: 0.31202496,
description: "Qwen3 235B A22B Thinking 2507 model with 262K context window.",
Expand All @@ -401,6 +413,8 @@ export const chutesModels = {
contextWindow: 131072,
supportsImages: false,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0,
outputPrice: 0,
description:
Expand All @@ -411,6 +425,8 @@ export const chutesModels = {
contextWindow: 262144,
supportsImages: true,
supportsPromptCache: false,
supportsReasoningEffort: true,
reasoningEffort: "medium",
inputPrice: 0.16,
outputPrice: 0.65,
description:
Expand Down
153 changes: 152 additions & 1 deletion src/api/providers/__tests__/chutes.spec.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -503,6 +503,7 @@ describe("ChutesHandler", () => {
temperature: 0.6,
stream: true,
stream_options: { include_usage: true },
reasoning_effort: "medium", // DeepSeek R1 now supports reasoning effort with default "medium"
}),
)
})
Expand Down Expand Up @@ -540,7 +541,6 @@ describe("ChutesHandler", () => {
stream: true,
stream_options: { include_usage: true },
}),
undefined,
)
})

Expand All @@ -563,4 +563,155 @@ describe("ChutesHandler", () => {
const model = handlerWithModel.getModel()
expect(model.info.temperature).toBe(0.5)
})

describe("reasoning effort support", () => {
it("should pass reasoning effort for models that support it", async () => {
const modelId: ChutesModelId = "deepseek-ai/DeepSeek-R1"

// Clear previous mocks and set up new implementation
mockCreate.mockClear()
mockCreate.mockImplementationOnce(async () => ({
[Symbol.asyncIterator]: async function* () {
yield { choices: [{ delta: { content: "test" } }], usage: null }
},
}))

const handlerWithModel = new ChutesHandler({
apiModelId: modelId,
chutesApiKey: "test-chutes-api-key",
enableReasoningEffort: true,
reasoningEffort: "high",
})

const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]

const generator = handlerWithModel.createMessage(systemPrompt, messages)
await generator.next()

expect(mockCreate).toHaveBeenCalledWith(
expect.objectContaining({
reasoning_effort: "high",
}),
)
})

it("should not pass reasoning effort for models that don't support it", async () => {
const modelId: ChutesModelId = "unsloth/Llama-3.3-70B-Instruct"

// Clear previous mocks and set up new implementation
mockCreate.mockClear()
mockCreate.mockImplementationOnce(async () => ({
[Symbol.asyncIterator]: async function* () {
yield { choices: [{ delta: { content: "test" } }], usage: null }
},
}))

const handlerWithModel = new ChutesHandler({
apiModelId: modelId,
chutesApiKey: "test-chutes-api-key",
enableReasoningEffort: true,
reasoningEffort: "high",
})

const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]

const generator = handlerWithModel.createMessage(systemPrompt, messages)
await generator.next()

const callArgs = mockCreate.mock.calls[0][0]
expect(callArgs).not.toHaveProperty("reasoning_effort")
})

it("should use model default reasoning effort when not explicitly set", async () => {
const modelId: ChutesModelId = "meituan-longcat/LongCat-Flash-Thinking-FP8"

// Clear previous mocks and set up new implementation
mockCreate.mockClear()
mockCreate.mockImplementationOnce(async () => ({
[Symbol.asyncIterator]: async function* () {
yield { choices: [{ delta: { content: "test" } }], usage: null }
},
}))

const handlerWithModel = new ChutesHandler({
apiModelId: modelId,
chutesApiKey: "test-chutes-api-key",
// Not setting enableReasoningEffort or reasoningEffort to test model defaults
})

const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]

const generator = handlerWithModel.createMessage(systemPrompt, messages)
await generator.next()

// Since we don't set enableReasoningEffort to true, and just rely on model defaults,
// the reasoning_effort will be included because the model has a default reasoningEffort
expect(mockCreate).toHaveBeenCalledWith(
expect.objectContaining({
reasoning_effort: "medium", // Should use the model's default
}),
)
})

it("should not pass reasoning effort when disabled", async () => {
const modelId: ChutesModelId = "deepseek-ai/DeepSeek-R1"

// Clear previous mocks and set up new implementation
mockCreate.mockClear()
mockCreate.mockImplementationOnce(async () => ({
[Symbol.asyncIterator]: async function* () {
yield { choices: [{ delta: { content: "test" } }], usage: null }
},
}))

const handlerWithModel = new ChutesHandler({
apiModelId: modelId,
chutesApiKey: "test-chutes-api-key",
enableReasoningEffort: false,
reasoningEffort: "high",
})

const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]

const generator = handlerWithModel.createMessage(systemPrompt, messages)
await generator.next()

const callArgs = mockCreate.mock.calls[0][0]
expect(callArgs).not.toHaveProperty("reasoning_effort")
})

it("should pass reasoning effort for thinking models", async () => {
const modelId: ChutesModelId = "Qwen/Qwen3-235B-A22B-Thinking-2507"

// Clear previous mocks and set up new implementation
mockCreate.mockClear()
mockCreate.mockImplementationOnce(async () => ({
[Symbol.asyncIterator]: async function* () {
yield { choices: [{ delta: { content: "test" } }], usage: null }
},
}))

const handlerWithModel = new ChutesHandler({
apiModelId: modelId,
chutesApiKey: "test-chutes-api-key",
reasoningEffort: "low", // Just set the reasoning effort, no need for enableReasoningEffort
})

const systemPrompt = "Test system prompt"
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Test message" }]

const generator = handlerWithModel.createMessage(systemPrompt, messages)
await generator.next()

expect(mockCreate).toHaveBeenCalledWith(
expect.objectContaining({
reasoning_effort: "low",
}),
)
})
})
})
48 changes: 42 additions & 6 deletions src/api/providers/chutes.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,10 +3,12 @@ import { Anthropic } from "@anthropic-ai/sdk"
import OpenAI from "openai"

import type { ApiHandlerOptions } from "../../shared/api"
import { getModelMaxOutputTokens } from "../../shared/api"
import { XmlMatcher } from "../../utils/xml-matcher"
import { convertToR1Format } from "../transform/r1-format"
import { convertToOpenAiMessages } from "../transform/openai-format"
import { ApiStream } from "../transform/stream"
import { getModelParams } from "../transform/model-params"

import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"

Expand All @@ -27,12 +29,18 @@ export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
systemPrompt: string,
messages: Anthropic.Messages.MessageParam[],
): OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming {
const {
id: model,
info: { maxTokens: max_tokens },
} = this.getModel()
const { id: model, info, reasoning } = this.getModel()

const temperature = this.options.modelTemperature ?? this.getModel().info.temperature
const temperature = this.options.modelTemperature ?? info.temperature

// Use centralized cap logic for max_tokens
const max_tokens =
getModelMaxOutputTokens({
modelId: model,
model: info,
settings: this.options,
format: "openai",
}) ?? undefined

return {
model,
Expand All @@ -41,6 +49,7 @@ export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
stream: true,
stream_options: { include_usage: true },
...(reasoning && reasoning),
}
}

Expand Down Expand Up @@ -85,19 +94,46 @@ export class ChutesHandler extends BaseOpenAiCompatibleProvider<ChutesModelId> {
yield processedChunk
}
} else {
yield* super.createMessage(systemPrompt, messages)
// For non-DeepSeek models, we need to handle reasoning effort
const stream = await this.client.chat.completions.create(this.getCompletionParams(systemPrompt, messages))

for await (const chunk of stream) {
const delta = chunk.choices[0]?.delta

if (delta?.content) {
yield {
type: "text",
text: delta.content,
}
}

if (chunk.usage) {
yield {
type: "usage",
inputTokens: chunk.usage.prompt_tokens || 0,
outputTokens: chunk.usage.completion_tokens || 0,
}
}
}
}
}

override getModel() {
const model = super.getModel()
const isDeepSeekR1 = model.id.includes("DeepSeek-R1")
const params = getModelParams({
format: "openai",
modelId: model.id,
model: model.info,
settings: this.options,
})
return {
...model,
info: {
...model.info,
temperature: isDeepSeekR1 ? DEEP_SEEK_DEFAULT_TEMPERATURE : this.defaultTemperature,
},
...params,
}
}
}
8 changes: 8 additions & 0 deletions webview-ui/src/components/settings/providers/Chutes.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,10 @@ import type { ProviderSettings } from "@roo-code/types"

import { useAppTranslation } from "@src/i18n/TranslationContext"
import { VSCodeButtonLink } from "@src/components/common/VSCodeButtonLink"
import { useSelectedModel } from "@src/components/ui/hooks/useSelectedModel"

import { inputEventTransform } from "../transforms"
import { ThinkingBudget } from "../ThinkingBudget"

type ChutesProps = {
apiConfiguration: ProviderSettings
Expand All @@ -15,6 +17,7 @@ type ChutesProps = {

export const Chutes = ({ apiConfiguration, setApiConfigurationField }: ChutesProps) => {
const { t } = useAppTranslation()
const { info: modelInfo } = useSelectedModel(apiConfiguration)

const handleInputChange = useCallback(
<K extends keyof ProviderSettings, E>(
Expand Down Expand Up @@ -45,6 +48,11 @@ export const Chutes = ({ apiConfiguration, setApiConfigurationField }: ChutesPro
{t("settings:providers.getChutesApiKey")}
</VSCodeButtonLink>
)}
<ThinkingBudget
apiConfiguration={apiConfiguration}
setApiConfigurationField={setApiConfigurationField}
modelInfo={modelInfo}
/>
</>
)
}