Fix: Provider like SiliconFlow show wrong tokens number on task header

Author: misskanagi

src/api/providers/openai.ts

@@ -60,6 +60,9 @@ export class OpenAiHandler implements ApiHandler, SingleCompletionHandler {
 		const deepseekReasoner = modelId.includes("deepseek-reasoner")
 		const ark = modelUrl.includes(".volces.com")
 
+		let last_prompt_tokens = 0
+		let last_completion_tokens = 0
+
 		if (this.options.openAiStreamingEnabled ?? true) {
 			const systemMessage: OpenAI.Chat.ChatCompletionSystemMessageParam = {
 				role: "system",
@@ -107,7 +110,9 @@ export class OpenAiHandler implements ApiHandler, SingleCompletionHandler {
 					}
 				}
 				if (chunk.usage) {
-					yield this.processUsageMetrics(chunk.usage)
+					yield this.processUsageMetrics(chunk.usage, last_prompt_tokens, last_completion_tokens)
+					last_prompt_tokens = chunk.usage?.prompt_tokens || 0
+					last_completion_tokens = chunk.usage?.completion_tokens || 0
 				}
 			}
 		} else {
@@ -130,15 +135,27 @@ export class OpenAiHandler implements ApiHandler, SingleCompletionHandler {
 				type: "text",
 				text: response.choices[0]?.message.content || "",
 			}
-			yield this.processUsageMetrics(response.usage)
+			yield this.processUsageMetrics(response.usage, last_prompt_tokens, last_completion_tokens)
 		}
 	}
 
-	protected processUsageMetrics(usage: any): ApiStreamUsageChunk {
+	protected processUsageMetrics(
+		usage: any,
+		last_prompt_tokens: number,
+		last_completion_tokens: number,
+	): ApiStreamUsageChunk {
+		if (this.options.openAiUsageCumulation) {
+			return {
+				type: "usage",
+				inputTokens: usage?.prompt_tokens || 0,
+				outputTokens: usage?.completion_tokens || 0,
+			}
+		}
+
 		return {
 			type: "usage",
-			inputTokens: usage?.prompt_tokens || 0,
-			outputTokens: usage?.completion_tokens || 0,
+			inputTokens: usage?.prompt_tokens - last_prompt_tokens || 0,
+			outputTokens: usage?.completion_tokens - last_completion_tokens || 0,
 		}
 	}
 

src/core/webview/ClineProvider.ts

@@ -84,6 +84,7 @@ type GlobalStateKey =
 	| "openAiModelId"
 	| "openAiCustomModelInfo"
 	| "openAiUseAzure"
+	| "openAiUsageCumulation"
 	| "ollamaModelId"
 	| "ollamaBaseUrl"
 	| "lmStudioModelId"
@@ -1615,6 +1616,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			openAiModelId,
 			openAiCustomModelInfo,
 			openAiUseAzure,
+			openAiUsageCumulation,
 			ollamaModelId,
 			ollamaBaseUrl,
 			lmStudioModelId,
@@ -1660,6 +1662,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 		await this.updateGlobalState("openAiModelId", openAiModelId)
 		await this.updateGlobalState("openAiCustomModelInfo", openAiCustomModelInfo)
 		await this.updateGlobalState("openAiUseAzure", openAiUseAzure)
+		await this.updateGlobalState("openAiUsageCumulation", openAiUsageCumulation)
 		await this.updateGlobalState("ollamaModelId", ollamaModelId)
 		await this.updateGlobalState("ollamaBaseUrl", ollamaBaseUrl)
 		await this.updateGlobalState("lmStudioModelId", lmStudioModelId)
@@ -2481,6 +2484,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			openAiModelId,
 			openAiCustomModelInfo,
 			openAiUseAzure,
+			openAiUsageCumulation,
 			ollamaModelId,
 			ollamaBaseUrl,
 			lmStudioModelId,
@@ -2561,6 +2565,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			this.getGlobalState("openAiModelId") as Promise<string | undefined>,
 			this.getGlobalState("openAiCustomModelInfo") as Promise<ModelInfo | undefined>,
 			this.getGlobalState("openAiUseAzure") as Promise<boolean | undefined>,
+			this.getGlobalState("openAiUsageCumulation") as Promise<boolean | undefined>,
 			this.getGlobalState("ollamaModelId") as Promise<string | undefined>,
 			this.getGlobalState("ollamaBaseUrl") as Promise<string | undefined>,
 			this.getGlobalState("lmStudioModelId") as Promise<string | undefined>,
@@ -2658,6 +2663,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 				openAiModelId,
 				openAiCustomModelInfo,
 				openAiUseAzure,
+				openAiUsageCumulation,
 				ollamaModelId,
 				ollamaBaseUrl,
 				lmStudioModelId,

src/shared/api.ts

@@ -45,6 +45,7 @@ export interface ApiHandlerOptions {
 	openAiModelId?: string
 	openAiCustomModelInfo?: ModelInfo
 	openAiUseAzure?: boolean
+	openAiUsageCumulation?: boolean
 	ollamaModelId?: string
 	ollamaBaseUrl?: string
 	lmStudioModelId?: string

webview-ui/src/components/settings/ApiOptions.tsx

@@ -628,6 +628,15 @@ const ApiOptions = ({ apiErrorMessage, modelIdErrorMessage, fromWelcomeView }: A
 							Enable streaming
 						</Checkbox>
 					</div>
+					<Checkbox
+						checked={apiConfiguration?.openAiUsageCumulation ?? true}
+						onChange={(checked: boolean) => {
+							handleInputChange("openAiUsageCumulation")({
+								target: { value: checked },
+							})
+						}}>
+						Enable token usage cumulation
+					</Checkbox>
 					<Checkbox
 						checked={apiConfiguration?.openAiUseAzure ?? false}
 						onChange={(checked: boolean) => {