Skip to content

B
bit-pm
Unverified 提交 fa80014e authored 作者: Will Chen's avatar Will Chen 提交者： GitHub
浏览文件
操作

Remove budget saver mode (#378)

This code was quite complex and hairy and resulted in very opaque errors (for both free and pro users). There's not much benefit to budget saver because Google removed 2.5 Pro free quota a while ago (after it graduated the model from experimental to preview). Dyad Pro users can still use 2.5 Flash free quota by disabling Dyad Pro by clicking on the Dyad Pro button at the top.
上级 534cbad9
隐藏空白字符变更
内嵌 并排
正在显示 包含 14 行增加 和 217 行删除
src/components/ProModeSelector.tsx
浏览文件 @ fa80014e
......@@ -18,10 +18,6 @@ import { IpcClient } from "@/ipc/ipc_client";
export function ProModeSelector() {
const { settings, updateSettings } = useSettings();
const toggleSaverMode = () => {
updateSettings({ enableProSaverMode: !settings?.enableProSaverMode });
};
const toggleLazyEdits = () => {
updateSettings({
enableProLazyEditsMode: !settings?.enableProLazyEditsMode,
......@@ -76,16 +72,6 @@ export function ProModeSelector() {
</a>
</div>
)}
<SelectorRow
id="saver-mode"
label="Saver Mode"
description="Uses your free Gemini API quota before consuming Dyad Pro AI credits"
tooltip="Note: using the free Gemini API lets Google use your data to
improve their models."
proEnabled={proEnabled}
settingEnabled={Boolean(settings?.enableProSaverMode)}
toggle={toggleSaverMode}
/>
<SelectorRow
id="lazy-edits"
label="Turbo Edits"
......
src/ipc/handlers/chat_stream_handlers.ts
浏览文件 @ fa80014e
import { ipcMain } from "electron";
import { CoreMessage, TextPart, ImagePart } from "ai";
import { CoreMessage, TextPart, ImagePart, streamText } from "ai";
import { db } from "../../db";
import { chats, messages } from "../../db/schema";
import { and, eq, isNull } from "drizzle-orm";
......@@ -32,7 +32,6 @@ import * as crypto from "crypto";
import { readFile, writeFile, unlink } from "fs/promises";
import { getMaxTokens } from "../utils/token_utils";
import { MAX_CHAT_TURNS_IN_CONTEXT } from "@/constants/settings_constants";
import { streamTextWithBackup } from "../utils/stream_utils";
import { validateChatContext } from "../utils/context_paths_utils";
const logger = log.scope("chat_stream_handlers");
......@@ -244,8 +243,11 @@ export function registerChatStreamHandlers() {
"estimated tokens",
codebaseInfo.length / 4,
);
const { modelClient, backupModelClients, isEngineEnabled } =
await getModelClient(settings.selectedModel, settings, files);
const { modelClient, isEngineEnabled } = await getModelClient(
settings.selectedModel,
settings,
files,
);
// Prepare message history for the AI
const messageHistory = updatedChat.messages.map((message) => ({
......@@ -394,11 +396,11 @@ This conversation includes one or more image attachments. When the user uploads
}
// When calling streamText, the messages need to be properly formatted for mixed content
const { textStream } = streamTextWithBackup({
const { textStream } = streamText({
maxTokens: await getMaxTokens(settings.selectedModel),
temperature: 0,
model: modelClient,
backupModelClients: backupModelClients,
maxRetries: 2,
model: modelClient.model,
system: systemPrompt,
messages: chatMessages.filter((m) => m.content),
onError: (error: any) => {
......
src/ipc/utils/get_model_client.ts
浏览文件 @ fa80014e
......@@ -10,7 +10,6 @@ import { getEnvVar } from "./read_env";
import log from "electron-log";
import { getLanguageModelProviders } from "../shared/language_model_helpers";
import { LanguageModelProvider } from "../ipc_types";
import { llmErrorStore } from "@/main/llm_error_store";
import { createDyadEngine } from "./llm_engine_provider";
import { findLanguageModel } from "./findLanguageModel";
import { LM_STUDIO_BASE_URL } from "./lm_studio_utils";
......@@ -50,7 +49,6 @@ export async function getModelClient(
files?: File[],
): Promise<{
modelClient: ModelClient;
backupModelClients: ModelClient[];
isEngineEnabled?: boolean;
}> {
const allProviders = await getLanguageModelProviders();
......@@ -143,42 +141,11 @@ export async function getModelClient(
),
builtinProviderId: "auto",
};
const googleSettings = settings.providerSettings?.google;
// Budget saver mode logic (all must be true):
// 1. Pro Saver Mode is enabled
// 2. Provider is Google
// 3. API Key is set
// 4. Has no recent errors
if (
settings.enableProSaverMode &&
providerConfig.id === "google" &&
googleSettings &&
googleSettings.apiKey?.value &&
llmErrorStore.modelHasNoRecentError({
model: model.name,
provider: providerConfig.id,
})
) {
return {
modelClient: getRegularModelClient(
{
provider: providerConfig.id,
name: model.name,
},
settings,
providerConfig,
).modelClient,
backupModelClients: [autoModelClient],
isEngineEnabled,
};
} else {
return {
modelClient: autoModelClient,
backupModelClients: [],
isEngineEnabled,
};
}
return {
modelClient: autoModelClient,
isEngineEnabled,
};
} else {
logger.warn(
`Dyad Pro enabled, but provider ${model.provider} does not have a gateway prefix defined. Falling back to direct provider connection.`,
......
src/ipc/utils/stream_utils.ts deleted 100644 → 0
浏览文件 @ 534cbad9
import { streamText } from "ai";
import log from "electron-log";
import { ModelClient } from "./get_model_client";
import { llmErrorStore } from "@/main/llm_error_store";
const logger = log.scope("stream_utils");
export interface StreamTextWithBackupParams
extends Omit<Parameters<typeof streamText>[0], "model"> {
model: ModelClient; // primary client
backupModelClients?: ModelClient[]; // ordered fall-backs
}
export function streamTextWithBackup(params: StreamTextWithBackupParams): {
textStream: AsyncIterable<string>;
} {
const {
model: primaryModel,
backupModelClients = [],
onError: callerOnError,
abortSignal: callerAbort,
...rest
} = params;
const modelClients: ModelClient[] = [primaryModel, ...backupModelClients];
async function* combinedGenerator(): AsyncIterable<string> {
let lastErr: { error: unknown } | undefined = undefined;
for (let i = 0; i < modelClients.length; i++) {
const currentModelClient = modelClients[i];
/* Local abort controller for this single attempt */
const attemptAbort = new AbortController();
if (callerAbort) {
if (callerAbort.aborted) {
// Already aborted, trigger immediately
attemptAbort.abort();
} else {
callerAbort.addEventListener("abort", () => attemptAbort.abort(), {
once: true,
});
}
}
let errorFromCurrent: { error: unknown } | undefined = undefined; // set when onError fires
const providerId = currentModelClient.builtinProviderId;
if (providerId) {
llmErrorStore.clearModelError({
model: currentModelClient.model.modelId,
provider: providerId,
});
}
logger.info(
"Streaming text with model",
currentModelClient.model.modelId,
"provider",
currentModelClient.model.provider,
"builtinProviderId",
currentModelClient.builtinProviderId,
);
const { textStream } = streamText({
...rest,
maxRetries: i === modelClients.length - 1 ? 3 : 0,
model: currentModelClient.model,
abortSignal: attemptAbort.signal,
onError: (error) => {
const providerId = currentModelClient.builtinProviderId;
if (providerId) {
llmErrorStore.recordModelError({
model: currentModelClient.model.modelId,
provider: providerId,
});
}
logger.error(
`Error streaming text with ${providerId} and model ${currentModelClient.model.modelId}: ${error}`,
error,
);
errorFromCurrent = error;
attemptAbort.abort(); // kill fetch / SSE
},
});
try {
for await (const chunk of textStream) {
/* If onError fired during streaming, bail out immediately. */
if (errorFromCurrent) throw errorFromCurrent;
yield chunk;
}
/* Stream ended – check if it actually failed */
if (errorFromCurrent) throw errorFromCurrent;
/* Completed successfully – stop trying more models. */
return;
} catch (err) {
if (typeof err === "object" && err !== null && "error" in err) {
lastErr = err as { error: unknown };
} else {
lastErr = { error: err };
}
logger.warn(
`[streamTextWithBackup] model #${i} failed – ${
i < modelClients.length - 1
? "switching to backup"
: "no backups left"
}`,
err,
);
/* loop continues to next model (if any) */
}
}
/* Every model failed */
if (!lastErr) {
throw new Error("Invariant in StreamTextWithbackup failed!");
}
callerOnError?.(lastErr);
logger.error("All model invocations failed", lastErr);
// throw lastErr ?? new Error("All model invocations failed");
}
return { textStream: combinedGenerator() };
}
src/lib/schemas.ts
浏览文件 @ fa80014e
......@@ -139,7 +139,6 @@ export const UserSettingsSchema = z.object({
experiments: ExperimentsSchema.optional(),
lastShownReleaseNotesVersion: z.string().optional(),
maxChatTurnsInContext: z.number().optional(),
enableProSaverMode: z.boolean().optional(),
enableProLazyEditsMode: z.boolean().optional(),
enableProSmartFilesContextMode: z.boolean().optional(),
selectedTemplateId: z.string().optional(),
......@@ -154,6 +153,7 @@ export const UserSettingsSchema = z.object({
////////////////////////////////
// DEPRECATED.
////////////////////////////////
enableProSaverMode: z.boolean().optional(),
dyadProBudget: DyadProBudgetSchema.optional(),
runtimeMode: RuntimeModeSchema.optional(),
});
......
src/main/llm_error_store.ts deleted 100644 → 0
浏览文件 @ 534cbad9
class LlmErrorStore {
private modelErrorToTimestamp: Record<string, number> = {};
constructor() {}
recordModelError({ model, provider }: { model: string; provider: string }) {
this.modelErrorToTimestamp[this.getKey({ model, provider })] = Date.now();
}
clearModelError({ model, provider }: { model: string; provider: string }) {
delete this.modelErrorToTimestamp[this.getKey({ model, provider })];
}
modelHasNoRecentError({
model,
provider,
}: {
model: string;
provider: string;
}): boolean {
const key = this.getKey({ model, provider });
const timestamp = this.modelErrorToTimestamp[key];
if (!timestamp) {
return true;
}
const oneHourAgo = Date.now() - 1000 * 60 * 60;
return timestamp < oneHourAgo;
}
private getKey({ model, provider }: { model: string; provider: string }) {
return `${provider}::${model}`;
}
}
export const llmErrorStore = new LlmErrorStore();
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论