Configure per-configuration temperature

Roo Code · Roo Code · commit bc9773ff9aab · 2025-02-08T00:21:13.000-05:00
diff --git a/src/api/providers/anthropic.ts b/src/api/providers/anthropic.ts
@@ -44,7 +44,7 @@ export class AnthropicHandler implements ApiHandler, SingleCompletionHandler {
 					{
 						model: modelId,
 						max_tokens: this.getModel().info.maxTokens || 8192,
-						temperature: 0,
+						temperature: this.options.modelTemperature ?? 0,
 						system: [{ text: systemPrompt, type: "text", cache_control: { type: "ephemeral" } }], // setting cache breakpoint for system prompt so new tasks can reuse it
 						messages: messages.map((message, index) => {
 							if (index === lastUserMsgIndex || index === secondLastMsgUserIndex) {
@@ -96,7 +96,7 @@ export class AnthropicHandler implements ApiHandler, SingleCompletionHandler {
 				stream = (await this.client.messages.create({
 					model: modelId,
 					max_tokens: this.getModel().info.maxTokens || 8192,
-					temperature: 0,
+					temperature: this.options.modelTemperature ?? 0,
 					system: [{ text: systemPrompt, type: "text" }],
 					messages,
 					// tools,
@@ -179,7 +179,7 @@ export class AnthropicHandler implements ApiHandler, SingleCompletionHandler {
 			const response = await this.client.messages.create({
 				model: this.getModel().id,
 				max_tokens: this.getModel().info.maxTokens || 8192,
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 				messages: [{ role: "user", content: prompt }],
 				stream: false,
 			})
diff --git a/src/api/providers/bedrock.ts b/src/api/providers/bedrock.ts
@@ -104,7 +104,7 @@ export class AwsBedrockHandler implements ApiHandler, SingleCompletionHandler {
 			system: [{ text: systemPrompt }],
 			inferenceConfig: {
 				maxTokens: modelConfig.info.maxTokens || 5000,
-				temperature: 0.3,
+				temperature: this.options.modelTemperature ?? 0.3,
 				topP: 0.1,
 				...(this.options.awsUsePromptCache
 					? {
@@ -262,7 +262,7 @@ export class AwsBedrockHandler implements ApiHandler, SingleCompletionHandler {
 				]),
 				inferenceConfig: {
 					maxTokens: modelConfig.info.maxTokens || 5000,
-					temperature: 0.3,
+					temperature: this.options.modelTemperature ?? 0.3,
 					topP: 0.1,
 				},
 			}
diff --git a/src/api/providers/gemini.ts b/src/api/providers/gemini.ts
@@ -23,7 +23,7 @@ export class GeminiHandler implements ApiHandler, SingleCompletionHandler {
 			contents: messages.map(convertAnthropicMessageToGemini),
 			generationConfig: {
 				// maxOutputTokens: this.getModel().info.maxTokens,
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 			},
 		})
 
@@ -60,7 +60,7 @@ export class GeminiHandler implements ApiHandler, SingleCompletionHandler {
 			const result = await model.generateContent({
 				contents: [{ role: "user", parts: [{ text: prompt }] }],
 				generationConfig: {
-					temperature: 0,
+					temperature: this.options.modelTemperature ?? 0,
 				},
 			})
 
diff --git a/src/api/providers/glama.ts b/src/api/providers/glama.ts
@@ -79,7 +79,7 @@ export class GlamaHandler implements ApiHandler, SingleCompletionHandler {
 		}
 
 		if (this.supportsTemperature()) {
-			requestOptions.temperature = 0
+			requestOptions.temperature = this.options.modelTemperature ?? 0
 		}
 
 		const { data: completion, response } = await this.client.chat.completions
@@ -172,7 +172,7 @@ export class GlamaHandler implements ApiHandler, SingleCompletionHandler {
 			}
 
 			if (this.supportsTemperature()) {
-				requestOptions.temperature = 0
+				requestOptions.temperature = this.options.modelTemperature ?? 0
 			}
 
 			if (this.getModel().id.startsWith("anthropic/")) {
diff --git a/src/api/providers/lmstudio.ts b/src/api/providers/lmstudio.ts
@@ -27,7 +27,7 @@ export class LmStudioHandler implements ApiHandler, SingleCompletionHandler {
 			const stream = await this.client.chat.completions.create({
 				model: this.getModel().id,
 				messages: openAiMessages,
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 				stream: true,
 			})
 			for await (const chunk of stream) {
@@ -59,7 +59,7 @@ export class LmStudioHandler implements ApiHandler, SingleCompletionHandler {
 			const response = await this.client.chat.completions.create({
 				model: this.getModel().id,
 				messages: [{ role: "user", content: prompt }],
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 				stream: false,
 			})
 			return response.choices[0]?.message.content || ""
diff --git a/src/api/providers/mistral.ts b/src/api/providers/mistral.ts
@@ -30,7 +30,7 @@ export class MistralHandler implements ApiHandler {
 		const stream = await this.client.chat.stream({
 			model: this.getModel().id,
 			// max_completion_tokens: this.getModel().info.maxTokens,
-			temperature: 0,
+			temperature: this.options.modelTemperature ?? 0,
 			messages: [{ role: "system", content: systemPrompt }, ...convertToMistralMessages(messages)],
 			stream: true,
 		})
diff --git a/src/api/providers/ollama.ts b/src/api/providers/ollama.ts
@@ -20,7 +20,7 @@ export class OllamaHandler implements ApiHandler, SingleCompletionHandler {
 
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
 		const modelId = this.getModel().id
-		const useR1Format = modelId.toLowerCase().includes('deepseek-r1')
+		const useR1Format = modelId.toLowerCase().includes("deepseek-r1")
 		const openAiMessages: OpenAI.Chat.ChatCompletionMessageParam[] = [
 			{ role: "system", content: systemPrompt },
 			...(useR1Format ? convertToR1Format(messages) : convertToOpenAiMessages(messages)),
@@ -29,7 +29,7 @@ export class OllamaHandler implements ApiHandler, SingleCompletionHandler {
 		const stream = await this.client.chat.completions.create({
 			model: this.getModel().id,
 			messages: openAiMessages,
-			temperature: 0,
+			temperature: this.options.modelTemperature ?? 0,
 			stream: true,
 		})
 		for await (const chunk of stream) {
@@ -53,11 +53,13 @@ export class OllamaHandler implements ApiHandler, SingleCompletionHandler {
 	async completePrompt(prompt: string): Promise<string> {
 		try {
 			const modelId = this.getModel().id
-			const useR1Format = modelId.toLowerCase().includes('deepseek-r1')
+			const useR1Format = modelId.toLowerCase().includes("deepseek-r1")
 			const response = await this.client.chat.completions.create({
 				model: this.getModel().id,
-				messages: useR1Format ? convertToR1Format([{ role: "user", content: prompt }]) : [{ role: "user", content: prompt }],
-				temperature: 0,
+				messages: useR1Format
+					? convertToR1Format([{ role: "user", content: prompt }])
+					: [{ role: "user", content: prompt }],
+				temperature: this.options.modelTemperature ?? 0,
 				stream: false,
 			})
 			return response.choices[0]?.message.content || ""
diff --git a/src/api/providers/openai-native.ts b/src/api/providers/openai-native.ts
@@ -88,7 +88,7 @@ export class OpenAiNativeHandler implements ApiHandler, SingleCompletionHandler
 	): ApiStream {
 		const stream = await this.client.chat.completions.create({
 			model: modelId,
-			temperature: 0,
+			temperature: this.options.modelTemperature ?? 0,
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
@@ -189,7 +189,7 @@ export class OpenAiNativeHandler implements ApiHandler, SingleCompletionHandler
 		return {
 			model: modelId,
 			messages: [{ role: "user", content: prompt }],
-			temperature: 0,
+			temperature: this.options.modelTemperature ?? 0,
 		}
 	}
 }
diff --git a/src/api/providers/openai.ts b/src/api/providers/openai.ts
@@ -57,7 +57,7 @@ export class OpenAiHandler implements ApiHandler, SingleCompletionHandler {
 			}
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 				model: modelId,
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? (deepseekReasoner ? 0.6 : 0),
 				messages: deepseekReasoner
 					? convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
 					: [systemMessage, ...convertToOpenAiMessages(messages)],
diff --git a/src/api/providers/openrouter.ts b/src/api/providers/openrouter.ts
@@ -115,7 +115,7 @@ export class OpenRouterHandler implements ApiHandler, SingleCompletionHandler {
 				break
 		}
 
-		let temperature = 0
+		let defaultTemperature = 0
 		let topP: number | undefined = undefined
 
 		// Handle models based on deepseek-r1
@@ -124,9 +124,8 @@ export class OpenRouterHandler implements ApiHandler, SingleCompletionHandler {
 			this.getModel().id === "perplexity/sonar-reasoning"
 		) {
 			// Recommended temperature for DeepSeek reasoning models
-			temperature = 0.6
-			// DeepSeek highly recommends using user instead of system
-			// role
+			defaultTemperature = 0.6
+			// DeepSeek highly recommends using user instead of system role
 			openAiMessages = convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
 			// Some provider support topP and 0.95 is value that Deepseek used in their benchmarks
 			topP = 0.95
@@ -137,7 +136,7 @@ export class OpenRouterHandler implements ApiHandler, SingleCompletionHandler {
 		const stream = await this.client.chat.completions.create({
 			model: this.getModel().id,
 			max_tokens: maxTokens,
-			temperature: temperature,
+			temperature: this.options.modelTemperature ?? defaultTemperature,
 			top_p: topP,
 			messages: openAiMessages,
 			stream: true,
@@ -224,7 +223,7 @@ export class OpenRouterHandler implements ApiHandler, SingleCompletionHandler {
 			const response = await this.client.chat.completions.create({
 				model: this.getModel().id,
 				messages: [{ role: "user", content: prompt }],
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 				stream: false,
 			})
 
diff --git a/src/api/providers/unbound.ts b/src/api/providers/unbound.ts
@@ -79,7 +79,7 @@ export class UnboundHandler implements ApiHandler, SingleCompletionHandler {
 				{
 					model: this.getModel().id.split("/")[1],
 					max_tokens: maxTokens,
-					temperature: 0,
+					temperature: this.options.modelTemperature ?? 0,
 					messages: openAiMessages,
 					stream: true,
 				},
@@ -146,7 +146,7 @@ export class UnboundHandler implements ApiHandler, SingleCompletionHandler {
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 				model: this.getModel().id.split("/")[1],
 				messages: [{ role: "user", content: prompt }],
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 			}
 
 			if (this.getModel().id.startsWith("anthropic/")) {
diff --git a/src/api/providers/vertex.ts b/src/api/providers/vertex.ts
@@ -22,7 +22,7 @@ export class VertexHandler implements ApiHandler, SingleCompletionHandler {
 		const stream = await this.client.messages.create({
 			model: this.getModel().id,
 			max_tokens: this.getModel().info.maxTokens || 8192,
-			temperature: 0,
+			temperature: this.options.modelTemperature ?? 0,
 			system: systemPrompt,
 			messages,
 			stream: true,
@@ -89,7 +89,7 @@ export class VertexHandler implements ApiHandler, SingleCompletionHandler {
 			const response = await this.client.messages.create({
 				model: this.getModel().id,
 				max_tokens: this.getModel().info.maxTokens || 8192,
-				temperature: 0,
+				temperature: this.options.modelTemperature ?? 0,
 				messages: [{ role: "user", content: prompt }],
 				stream: false,
 			})
diff --git a/src/core/webview/ClineProvider.ts b/src/core/webview/ClineProvider.ts
@@ -119,6 +119,7 @@ type GlobalStateKey =
 	| "autoApprovalEnabled"
 	| "customModes" // Array of custom modes
 	| "unboundModelId"
+	| "modelTemperature"
 
 export const GlobalFileNames = {
 	apiConversationHistory: "api_conversation_history.json",
@@ -1538,6 +1539,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			mistralApiKey,
 			unboundApiKey,
 			unboundModelId,
+			modelTemperature,
 		} = apiConfiguration
 		await this.updateGlobalState("apiProvider", apiProvider)
 		await this.updateGlobalState("apiModelId", apiModelId)
@@ -1578,6 +1580,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 		await this.storeSecret("mistralApiKey", mistralApiKey)
 		await this.storeSecret("unboundApiKey", unboundApiKey)
 		await this.updateGlobalState("unboundModelId", unboundModelId)
+		await this.updateGlobalState("modelTemperature", modelTemperature)
 		if (this.cline) {
 			this.cline.api = buildApiHandler(apiConfiguration)
 		}
@@ -2254,6 +2257,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			experiments,
 			unboundApiKey,
 			unboundModelId,
+			modelTemperature,
 		] = await Promise.all([
 			this.getGlobalState("apiProvider") as Promise<ApiProvider | undefined>,
 			this.getGlobalState("apiModelId") as Promise<string | undefined>,
@@ -2328,6 +2332,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			this.getGlobalState("experiments") as Promise<Record<ExperimentId, boolean> | undefined>,
 			this.getSecret("unboundApiKey") as Promise<string | undefined>,
 			this.getGlobalState("unboundModelId") as Promise<string | undefined>,
+			this.getGlobalState("modelTemperature") as Promise<number | undefined>,
 		])
 
 		let apiProvider: ApiProvider
@@ -2385,6 +2390,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 				vsCodeLmModelSelector,
 				unboundApiKey,
 				unboundModelId,
+				modelTemperature,
 			},
 			lastShownAnnouncementId,
 			customInstructions,
diff --git a/src/shared/api.ts b/src/shared/api.ts
@@ -60,6 +60,7 @@ export interface ApiHandlerOptions {
 	includeMaxTokens?: boolean
 	unboundApiKey?: string
 	unboundModelId?: string
+	modelTemperature?: number
 }
 
 export type ApiConfiguration = ApiHandlerOptions & {
diff --git a/webview-ui/src/components/settings/ApiOptions.tsx b/webview-ui/src/components/settings/ApiOptions.tsx
@@ -2,6 +2,7 @@ import { memo, useCallback, useEffect, useMemo, useState } from "react"
 import { useEvent, useInterval } from "react-use"
 import { Checkbox, Dropdown, Pane, type DropdownOption } from "vscrui"
 import { VSCodeLink, VSCodeRadio, VSCodeRadioGroup, VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
+import { TemperatureControl } from "./TemperatureControl"
 import * as vscodemodels from "vscode"
 
 import {
@@ -1361,6 +1362,18 @@ const ApiOptions = ({ apiErrorMessage, modelIdErrorMessage }: ApiOptionsProps) =
 					</>
 				)}
 
+			<div style={{ marginTop: "10px" }}>
+				<TemperatureControl
+					value={apiConfiguration?.modelTemperature}
+					onChange={(value) => {
+						handleInputChange("modelTemperature")({
+							target: { value },
+						})
+					}}
+					maxValue={2}
+				/>
+			</div>
+
 			{modelIdErrorMessage && (
 				<p
 					style={{
diff --git a/webview-ui/src/components/settings/TemperatureControl.tsx b/webview-ui/src/components/settings/TemperatureControl.tsx

Original file line number	Diff line number	Diff line change
`@@ -79,7 +79,7 @@ export class GlamaHandler implements ApiHandler, SingleCompletionHandler {`
`79`	`79`	`}`
`80`	`80`
`81`	`81`	`if (this.supportsTemperature()) {`
`82`		`- requestOptions.temperature = 0`
	`82`	`+ requestOptions.temperature = this.options.modelTemperature ?? 0`
`83`	`83`	`}`
`84`	`84`
`85`	`85`	`const { data: completion, response } = await this.client.chat.completions`
`@@ -172,7 +172,7 @@ export class GlamaHandler implements ApiHandler, SingleCompletionHandler {`
`172`	`172`	`}`
`173`	`173`
`174`	`174`	`if (this.supportsTemperature()) {`
`175`		`- requestOptions.temperature = 0`
	`175`	`+ requestOptions.temperature = this.options.modelTemperature ?? 0`
`176`	`176`	`}`
`177`	`177`
`178`	`178`	`if (this.getModel().id.startsWith("anthropic/")) {`
Original file line number	Diff line number	Diff line change
`@@ -88,7 +88,7 @@ export class OpenAiNativeHandler implements ApiHandler, SingleCompletionHandler`
`88`	`88`	`): ApiStream {`
`89`	`89`	`const stream = await this.client.chat.completions.create({`
`90`	`90`	`model: modelId,`
`91`		`- temperature: 0,`
	`91`	`+ temperature: this.options.modelTemperature ?? 0,`
`92`	`92`	`messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],`
`93`	`93`	`stream: true,`
`94`	`94`	`stream_options: { include_usage: true },`
`@@ -189,7 +189,7 @@ export class OpenAiNativeHandler implements ApiHandler, SingleCompletionHandler`
`189`	`189`	`return {`
`190`	`190`	`model: modelId,`
`191`	`191`	`messages: [{ role: "user", content: prompt }],`
`192`		`- temperature: 0,`
	`192`	`+ temperature: this.options.modelTemperature ?? 0,`
`193`	`193`	`}`
`194`	`194`	`}`
`195`	`195`	`}`
Original file line number	Diff line number	Diff line change
`@@ -57,7 +57,7 @@ export class OpenAiHandler implements ApiHandler, SingleCompletionHandler {`
`57`	`57`	`}`
`58`	`58`	`const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {`
`59`	`59`	`model: modelId,`
`60`		`- temperature: 0,`
	`60`	`+ temperature: this.options.modelTemperature ?? (deepseekReasoner ? 0.6 : 0),`
`61`	`61`	`messages: deepseekReasoner`
`62`	`62`	`? convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])`
`63`	`63`	`: [systemMessage, ...convertToOpenAiMessages(messages)],`