package back

packages/008Q/package.json

@@ -16,7 +16,7 @@
     "watch": "webpack --watch"
   },
   "dependencies": {
-    "@mlc-ai/web-llm": "file:../../3Party/web-llm",
+    "@mlc-ai/web-llm": "file:../web-llm",
     "audiobuffer-to-wav": "^1.0.0",
     "onnxruntime-web": "^1.16.3",
     "whisper-webgpu": "^0.8.0"

packages/web-llm/lib/chat_module.d.ts

@@ -0,0 +1,65 @@
+import { AppConfig } from "./config";
+import { InitProgressCallback, ChatInterface, ChatOptions, GenerateProgressCallback, LogitProcessor } from "./types";
+/**
+ * This is the main interface to the chat module.
+ */
+export declare class ChatModule implements ChatInterface {
+    private logger;
+    private logitProcessorRegistry?;
+    private logitProcessor?;
+    private pipeline?;
+    private initProgressCallback?;
+    private interruptSignal;
+    private deviceLostIsError;
+    constructor(logitProcessorRegistry?: Map<string, LogitProcessor>);
+    setInitProgressCallback(initProgressCallback: InitProgressCallback): void;
+    reload(localId: string, chatOpts?: ChatOptions, appConfig?: AppConfig): Promise<void>;
+    generate(input: string, progressCallback?: GenerateProgressCallback, streamInterval?: number): Promise<string>;
+    interruptGenerate(): Promise<void>;
+    runtimeStatsText(): Promise<string>;
+    resetChat(keepStats?: boolean): Promise<void>;
+    unload(): Promise<void>;
+    getMaxStorageBufferBindingSize(): Promise<number>;
+    getGPUVendor(): Promise<string>;
+    forwardTokensAndSample(inputIds: Array<number>, curPos: number, isPrefill: boolean): Promise<number>;
+    /**
+     * @returns Whether the generation stopped.
+     */
+    stopped(): boolean;
+    /**
+     * Get the current generated response.
+     *
+     * @returns The current output message.
+     */
+    getMessage(): string;
+    /**
+     * Run a prefill step with a given input.
+     * @param input The input prompt.
+     */
+    prefill(input: string): Promise<void>;
+    /**
+     * Run a decode step to decode the next token.
+     */
+    decode(): Promise<void>;
+    private getPipeline;
+    private asyncLoadTokenizer;
+}
+/**
+ * This is the interface to the chat module that connects to the REST API.
+ */
+export declare class ChatRestModule implements ChatInterface {
+    private logger;
+    private initProgressCallback?;
+    setInitProgressCallback(initProgressCallback: InitProgressCallback): void;
+    reload(localId: string, chatOpts?: ChatOptions, appConfig?: AppConfig): Promise<void>;
+    getMaxStorageBufferBindingSize(): Promise<number>;
+    getGPUVendor(): Promise<string>;
+    unload(): Promise<void>;
+    interruptGenerate(): Promise<void>;
+    forwardTokensAndSample(inputIds: Array<number>, curPos: number, isPrefill: boolean): Promise<number>;
+    generate(input: string, progressCallback?: GenerateProgressCallback, streamInterval?: number): Promise<string>;
+    runtimeStatsText(): Promise<string>;
+    resetChat(keepStats?: boolean): Promise<void>;
+}
+export declare function hasModelInCache(localId: string, appConfig?: AppConfig): Promise<boolean>;
+//# sourceMappingURL=chat_module.d.ts.map

packages/web-llm/lib/config.d.ts

@@ -0,0 +1,58 @@
+/**
+ * Conversation template config
+ */
+export interface ConvTemplateConfig {
+    system: string;
+    roles: Array<string>;
+    seps: Array<string>;
+    separator_style: string;
+    offset: number;
+    stop_str: string;
+    add_bos: boolean;
+    stop_tokens: Array<number>;
+}
+/**
+ * Config of one chat model
+ */
+export interface ChatConfig {
+    tokenizer_files: Array<string>;
+    conv_config?: Partial<ConvTemplateConfig>;
+    conv_template: string;
+    mean_gen_len: number;
+    shift_fill_factor: number;
+    repetition_penalty: number;
+    top_p: number;
+    temperature: number;
+}
+/**
+ * Information for a model.
+ * @param model_url: the huggingface link to download the model weights.
+ * @param local_id: what we call the model.
+ * @param model_lib_url: link to the model library (wasm file) the model uses.
+ * @param vram_required_MB: amount of vram in MB required to run the model (can use
+ *    `utils/vram_requirements` to calculate).
+ * @param low_resource_required: whether the model can run on limited devices (e.g. Android phone).
+ * @param required_features: feature needed to run this model (e.g. shader-f16).
+ */
+export interface ModelRecord {
+    model_url: string;
+    local_id: string;
+    model_lib_url: string;
+    vram_required_MB?: number;
+    low_resource_required?: boolean;
+    required_features?: Array<string>;
+}
+/**
+ * Extra configuration that can be
+ * passed to the load.
+ *
+ * @param model_list: models to be used.
+ */
+export interface AppConfig {
+    model_list: Array<ModelRecord>;
+}
+/**
+ * Default models and model library mapping to be used if unspecified.
+ */
+export declare const prebuiltAppConfig: AppConfig;
+//# sourceMappingURL=config.d.ts.map

packages/web-llm/lib/conversation.d.ts

@@ -0,0 +1,33 @@
+import { ConvTemplateConfig } from "./config";
+/**
+ * Helper to keep track of history conversations.
+ */
+export declare class Conversation {
+    messages: Array<[string, string | undefined]>;
+    config: ConvTemplateConfig;
+    constructor(config: ConvTemplateConfig);
+    private getPromptArrayInternal;
+    /**
+     * Get prompt arrays with the first one as system.
+     *
+     * @returns The prompt array.
+     */
+    getPromptArray(): Array<string>;
+    /**
+     * Get the last round of prompt has not been fed as input.
+     *
+     * @note This function needs to be used with the assumption that
+     *       the caller call appendMessage then appendReplyHeader.
+     *
+     * @returns The prompt array.
+     */
+    getPrompArrayLastRound(): string[];
+    reset(): void;
+    getStopStr(): string;
+    getStopTokens(): number[];
+    appendMessage(role: string, message: string): void;
+    appendReplyHeader(role: string): void;
+    finishReply(message: string): void;
+}
+export declare function getConversation(conv_template: string, conv_config?: Partial<ConvTemplateConfig>): Conversation;
+//# sourceMappingURL=conversation.d.ts.map

packages/web-llm/lib/index.d.ts

@@ -0,0 +1,5 @@
+export { ModelRecord, AppConfig } from "./config";
+export { InitProgressCallback, InitProgressReport, ChatOptions, ChatInterface, LogitProcessor, } from "./types";
+export { ChatModule, ChatRestModule, hasModelInCache } from "./chat_module";
+export { ChatWorkerHandler, ChatWorkerClient, WorkerMessage, CustomRequestParams } from "./web_worker";
+//# sourceMappingURL=index.d.ts.map