cactus-react-native 0.2.10 → 1.0.1

package/lib/typescript/vlm.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"vlm.d.ts","sourceRoot":"","sources":["../../src/vlm.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,KAAK,aAAa,EAClB,KAAK,gBAAgB,EACrB,KAAK,0BAA0B,EAC/B,KAAK,sBAAsB,EAC5B,MAAM,SAAS,CAAA;AAIhB,OAAO,EAAE,0BAA0B,EAAE,MAAM,QAAQ,CAAA;AAEnD,UAAU,eAAe;IACvB,GAAG,EAAE,SAAS,GAAG,IAAI,CAAA;IACrB,KAAK,EAAE,KAAK,GAAG,IAAI,CAAA;CACpB;AAED,MAAM,MAAM,gBAAgB,GAAG,aAAa,GAAG;IAC7C,MAAM,EAAE,MAAM,CAAA;CACf,CAAA;AAED,MAAM,MAAM,mBAAmB,GAAG,IAAI,CAAC,gBAAgB,EAAE,QAAQ,CAAC,GAAG;IACnE,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,IAAI,CAAC,EAAE,MAAM,CAAA;CACd,CAAA;AAED,qBAAa,SAAS;IACpB,OAAO,CAAC,OAAO,CAAc;IAC7B,SAAS,CAAC,0BAA0B,EAAE,0BAA0B,CAAA;IAEhE,OAAO,CAAC,MAAM,CAAC,UAAU,CAAoD;IAE7E,OAAO,CAAC,MAAM,CAAC,WAAW;IAI1B,OAAO;WAKM,IAAI,CACf,MAAM,EAAE,gBAAgB,EACxB,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,EACvC,WAAW,CAAC,EAAE,MAAM,EACpB,YAAY,CAAC,EAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,GACvD,OAAO,CAAC,eAAe,CAAC;IA8ErB,UAAU,CACd,QAAQ,EAAE,0BAA0B,EAAE,EACtC,MAAM,GAAE,mBAAwB,EAChC,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,GAAG,KAAK,IAAI,GAC7B,OAAO,CAAC,sBAAsB,CAAC;IAuClC,OAAO,CAAC,sBAAsB,CAyC7B;YAEa,uBAAuB;IA+C/B,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;IAIvB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAIxB,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC;IAIrC,gBAAgB,IAAI,OAAO;CAG5B"}

package/scripts/postInstall.js

@@ -1,33 +0,0 @@
-const fs = require('fs');
-const path = require('path');
-const crypto = require('crypto');
-
-const projectRoot = process.env.INIT_CWD;
-
-if (!projectRoot || projectRoot === process.cwd()) {
-  return;
-}
-
-const rootIdPath = path.join(projectRoot, '.projectId.json');
-const internalIdPath = path.join(__dirname, '..', 'src', 'projectId.ts');
-
-let projectId;
-try {
-  if (fs.existsSync(rootIdPath)) {
-    projectId = JSON.parse(fs.readFileSync(rootIdPath, 'utf8'));
-  } else {
-    projectId = crypto.randomUUID();
-    fs.writeFileSync(rootIdPath, JSON.stringify(projectId), 'utf8');
-  }
-} catch (error) {
-  console.error('Failed to read or create project ID.', error);
-  return;
-}
-
-const fileContent = `// DO NOT EDIT. This file is automatically generated by a post-install script.\nexport const PROJECT_ID = '${projectId}';`;
-
-try {
-  fs.writeFileSync(internalIdPath, fileContent, 'utf8');
-} catch (error) {
-  console.error('Failed to sync project ID into the framework.', error);
-}

package/src/NativeCactus.ts

@@ -1,317 +0,0 @@
-import type { TurboModule } from 'react-native'
-import { TurboModuleRegistry } from 'react-native'
-
-export type NativeEmbeddingParams = {
-  embd_normalize?: number
-}
-
-export type NativeContextParams = {
-  model: string
-  chat_template?: string
-  reasoning_format?: string
-  is_model_asset?: boolean
-  use_progress_callback?: boolean
-  n_ctx?: number
-  n_batch?: number
-  n_ubatch?: number
-  n_threads?: number
-  n_gpu_layers?: number
-  no_gpu_devices?: boolean
-  flash_attn?: boolean
-  cache_type_k?: string
-  cache_type_v?: string
-  use_mlock?: boolean
-  use_mmap?: boolean
-  vocab_only?: boolean
-  lora?: string
-  lora_scaled?: number
-  lora_list?: Array<{ path: string; scaled?: number }>
-  rope_freq_base?: number
-  rope_freq_scale?: number
-  pooling_type?: number
-  embedding?: boolean
-  embd_normalize?: number
-}
-
-export type NativeCompletionParams = {
-  prompt: string
-  n_threads?: number
-  json_schema?: string
-  grammar?: string
-  grammar_lazy?: boolean
-  grammar_triggers?: Array<{
-    type: number
-    value: string
-    token: number
-  }>
-  preserved_tokens?: Array<string>
-  chat_format?: number
-  stop?: Array<string>
-  n_predict?: number
-  n_probs?: number
-  top_k?: number
-  top_p?: number
-  min_p?: number
-  xtc_probability?: number
-  xtc_threshold?: number
-  typical_p?: number
-  temperature?: number
-  penalty_last_n?: number
-  penalty_repeat?: number
-  penalty_freq?: number
-  penalty_present?: number
-  mirostat?: number
-  mirostat_tau?: number
-  mirostat_eta?: number
-  dry_multiplier?: number
-  dry_base?: number
-  dry_allowed_length?: number
-  dry_penalty_last_n?: number
-  dry_sequence_breakers?: Array<string>
-  top_n_sigma?: number
-  ignore_eos?: boolean
-  logit_bias?: Array<Array<number>>
-  seed?: number
-  emit_partial_completion: boolean
-}
-
-export type NativeCompletionTokenProbItem = {
-  tok_str: string
-  prob: number
-}
-
-export type NativeCompletionTokenProb = {
-  content: string
-  probs: Array<NativeCompletionTokenProbItem>
-}
-
-export type NativeCompletionResultTimings = {
-  prompt_n: number
-  prompt_ms: number
-  prompt_per_token_ms: number
-  prompt_per_second: number
-  predicted_n: number
-  predicted_ms: number
-  predicted_per_token_ms: number
-  predicted_per_second: number
-}
-
-export type NativeCompletionResult = {
-  text: string
-  reasoning_content: string
-  tool_calls: Array<{
-    type: 'function'
-    function: {
-      name: string
-      arguments: string
-    }
-    id?: string
-  }>
-  content: string
-  tokens_predicted: number
-  tokens_evaluated: number
-  truncated: boolean
-  stopped_eos: boolean
-  stopped_word: string
-  stopped_limit: number
-  stopping_word: string
-  tokens_cached: number
-  timings: NativeCompletionResultTimings
-  completion_probabilities?: Array<NativeCompletionTokenProb>
-}
-
-export type NativeTokenizeResult = {
-  tokens: Array<number>
-  has_media?: boolean
-  bitmap_hashes?: Array<string>
-  chunk_pos?: Array<number>
-  chunk_pos_media?: Array<number>
-}
-
-export type NativeEmbeddingResult = {
-  embedding: Array<number>
-}
-
-export type NativeTTSType = {
-  type: number
-}
-
-export type NativeAudioCompletionResult = {
-  formatted_prompt: string
-}
-
-export type NativeAudioTokensResult = {
-  tokens: Array<number>
-}
-
-export type NativeAudioDecodeResult = {
-  audio_data: Array<number>
-}
-
-export type NativeDeviceInfo = {
-  deviceId: string
-  model: string
-  make: string
-  os: string
-}
-
-export type NativeLlamaContext = {
-  contextId: number
-  model: {
-    desc: string
-    size: number
-    nEmbd: number
-    nParams: number
-    chatTemplates: {
-      llamaChat: boolean
-      minja: {
-        default: boolean
-        defaultCaps: {
-          tools: boolean
-          toolCalls: boolean
-          toolResponses: boolean
-          systemRole: boolean
-          parallelToolCalls: boolean
-          toolCallId: boolean
-        }
-        toolUse: boolean
-        toolUseCaps: {
-          tools: boolean
-          toolCalls: boolean
-          toolResponses: boolean
-          systemRole: boolean
-          parallelToolCalls: boolean
-          toolCallId: boolean
-        }
-      }
-    }
-    metadata: Object
-    isChatTemplateSupported: boolean
-  }
-  androidLib?: string
-  gpu: boolean
-  reasonNoGPU: string
-}
-
-export type NativeSessionLoadResult = {
-  tokens_loaded: number
-  prompt: string
-}
-
-export type NativeLlamaChatMessage = {
-  role: string
-  content: string
-}
-
-export type JinjaFormattedChatResult = {
-  prompt: string
-  chat_format?: number
-  grammar?: string
-  grammar_lazy?: boolean
-  grammar_triggers?: Array<{
-    type: number
-    value: string
-    token: number
-  }>
-  preserved_tokens?: Array<string>
-  additional_stops?: Array<string>
-}
-
-export interface Spec extends TurboModule {
-  toggleNativeLog(enabled: boolean): Promise<void>
-  setContextLimit(limit: number): Promise<void>
-  modelInfo(path: string, skip?: string[]): Promise<Object>
-  initContext(
-    contextId: number,
-    params: NativeContextParams,
-  ): Promise<NativeLlamaContext>
-  getFormattedChat(
-    contextId: number,
-    messages: string,
-    chatTemplate?: string,
-    params?: {
-      jinja?: boolean
-      json_schema?: string
-      tools?: string
-      parallel_tool_calls?: string
-      tool_choice?: string
-    },
-  ): Promise<JinjaFormattedChatResult | string>
-  loadSession(
-    contextId: number,
-    filepath: string,
-  ): Promise<NativeSessionLoadResult>
-  saveSession(
-    contextId: number,
-    filepath: string,
-    size: number,
-  ): Promise<number>
-  completion(
-    contextId: number,
-    params: NativeCompletionParams,
-  ): Promise<NativeCompletionResult>
-  multimodalCompletion(
-    contextId: number,
-    prompt: string,
-    mediaPaths: string[],
-    params: NativeCompletionParams,
-  ): Promise<NativeCompletionResult>
-  stopCompletion(contextId: number): Promise<void>
-  tokenize(contextId: number, text: string, mediaPaths?: string[]): Promise<NativeTokenizeResult>
-  detokenize(contextId: number, tokens: number[]): Promise<string>
-  embedding(
-    contextId: number,
-    text: string,
-    params: NativeEmbeddingParams,
-  ): Promise<NativeEmbeddingResult>
-  bench(
-    contextId: number,
-    pp: number,
-    tg: number,
-    pl: number,
-    nr: number,
-  ): Promise<string>
-  applyLoraAdapters(
-    contextId: number,
-    loraAdapters: Array<{ path: string; scaled?: number }>,
-  ): Promise<void>
-  removeLoraAdapters(contextId: number): Promise<void>
-  getLoadedLoraAdapters(
-    contextId: number,
-  ): Promise<Array<{ path: string; scaled?: number }>>
-  initMultimodal(
-    contextId: number,
-    mmprojPath: string,
-    useGpu?: boolean,
-  ): Promise<boolean>
-  isMultimodalEnabled(contextId: number): Promise<boolean>
-  isMultimodalSupportVision(contextId: number): Promise<boolean>
-  isMultimodalSupportAudio(contextId: number): Promise<boolean>
-  releaseMultimodal(contextId: number): Promise<void>
-  initVocoder(
-    contextId: number,
-    vocoderModelPath: string,
-  ): Promise<boolean>
-  isVocoderEnabled(contextId: number): Promise<boolean>
-  getTTSType(contextId: number): Promise<NativeTTSType>
-  getFormattedAudioCompletion(
-    contextId: number,
-    speakerJsonStr: string,
-    textToSpeak: string,
-  ): Promise<NativeAudioCompletionResult>
-  getAudioCompletionGuideTokens(
-    contextId: number,
-    textToSpeak: string,
-  ): Promise<NativeAudioTokensResult>
-  decodeAudioTokens(
-    contextId: number,
-    tokens: number[],
-  ): Promise<NativeAudioDecodeResult>
-  getDeviceInfo(contextId: number): Promise<NativeDeviceInfo>
-  releaseVocoder(contextId: number): Promise<void>
-  rewind(contextId: number): Promise<void>
-  releaseContext(contextId: number): Promise<void>
-  releaseAllContexts(): Promise<void>
-}
-
-export default TurboModuleRegistry.get<Spec>('Cactus') as Spec

package/src/agent.ts

@@ -1,112 +0,0 @@
-import { CactusLM } from './lm'
-
-import type {
-  ContextParams,
-  CompletionParams,
-  CactusOAICompatibleMessage,
-  NativeCompletionResult,
-  TokenData,
-} from './index'
-
-import { Tools } from './tools'
-
-interface Parameter {
-  type: string
-  description: string
-  required?: boolean
-}
-
-interface CactusAgentReturn {
-  agent: CactusAgent | null
-  error: Error | null
-  lm: CactusAgent | null
-}
-
-export interface AgentCompletionParams extends CompletionParams {
-  tools?: Tools
-}
-
-export class CactusAgent extends CactusLM {
-  private tools!: Tools
-
-  static async init(
-    params: ContextParams,
-    onProgress?: (progress: number) => void,
-    cactusToken?: string,
-    retryOptions?: { maxRetries?: number; delayMs?: number },
-  ): Promise<CactusAgentReturn> {
-    const result = await CactusLM.init(params, onProgress, cactusToken, retryOptions)
-    
-    if (result.error || !result.lm) {
-      return { agent: null, error: result.error, lm: null }
-    }
-
-    const agent = Object.setPrototypeOf(result.lm, CactusAgent.prototype) as CactusAgent
-    agent.tools = new Tools()
-    return { agent, error: null, lm: agent }
-  }
-
-  addTool(
-    func: Function,
-    description: string,
-    parameters: { [key: string]: Parameter }
-  ): Function {
-    return this.tools.add(func, description, parameters)
-  }
-
-  getTools(): Tools {
-    return this.tools
-  }
-
-  async completionWithTools(
-    messages: CactusOAICompatibleMessage[],
-    params: AgentCompletionParams = {},
-    callback?: (data: TokenData) => void,
-    recursionCount: number = 0,
-    recursionLimit: number = 3
-  ): Promise<NativeCompletionResult> {
-    const tools = params.tools || this.tools
-    
-    if (!messages?.length) {
-      return this.completion([], params, callback)
-    }
-    if (!tools || tools.getSchemas().length === 0) {
-      return this.completion(messages, params, callback)
-    }
-    if (recursionCount >= recursionLimit) {
-      return this.completion(messages, {
-        ...params,
-        jinja: true,
-        tools: tools.getSchemas()
-      }, callback)
-    }
-
-    const { newMessages, requiresReset } =
-      this.conversationHistoryManager.processNewMessages(messages)
-
-    if (requiresReset) {
-      this.context?.rewind()
-      this.conversationHistoryManager.reset()
-    }
-
-    const result = await this.context.completionWithTools({
-      ...params,
-      messages: newMessages.length > 0 ? newMessages : messages,
-      tools: tools
-    }, callback, recursionCount, recursionLimit)
-
-    this.conversationHistoryManager.update(
-      newMessages.length > 0 ? newMessages : messages,
-      {
-        role: 'assistant',
-        content: result.content,
-        tool_calls: result.tool_calls
-      }
-    )
-
-    return result
-  }
-  isJinjaSupported(): boolean {
-    return this.context.isJinjaSupported();
-  }
-}

package/src/chat.ts

@@ -1,91 +0,0 @@
-import type { NativeLlamaChatMessage } from './NativeCactus'
-
-export type CactusMessagePart = {
-  text?: string
-}
-
-export type CactusOAICompatibleMessage = {
-  role: string
-  content?: string | CactusMessagePart[] | any
-  tool_calls?: any[]
-  tool_call_id?: string
-}
-
-export function formatChat(
-  messages: CactusOAICompatibleMessage[],
-): NativeLlamaChatMessage[] {
-  const chat: NativeLlamaChatMessage[] = []
-
-  messages.forEach((currMsg) => {
-    const role: string = currMsg.role || ''
-
-    let content: string = ''
-    if ('content' in currMsg) {
-      if (typeof currMsg.content === 'string') {
-        ;({ content } = currMsg)
-      } else if (Array.isArray(currMsg.content)) {
-        currMsg.content.forEach((part) => {
-          if ('text' in part) {
-            content += `${content ? '\n' : ''}${part.text}`
-          }
-        })
-      } else {
-        throw new TypeError(
-          "Invalid 'content' type (ref: https://github.com/ggerganov/llama.cpp/issues/8367)",
-        )
-      }
-    } else {
-      throw new Error(
-        "Missing 'content' (ref: https://github.com/ggerganov/llama.cpp/issues/8367)",
-      )
-    }
-
-    chat.push({ role, content })
-  })
-  return chat
-}
-
-export interface ProcessedMessages {
-  newMessages: CactusOAICompatibleMessage[];
-  requiresReset: boolean;
-}
-
-export class ConversationHistoryManager {
-  private history: CactusOAICompatibleMessage[] = [];
-
-  public processNewMessages(
-    fullMessageHistory: CactusOAICompatibleMessage[]
-  ): ProcessedMessages {
-    let divergent = fullMessageHistory.length < this.history.length;
-    if (!divergent) {
-      for (let i = 0; i < this.history.length; i++) {
-        if (JSON.stringify(this.history[i]) !== JSON.stringify(fullMessageHistory[i])) {
-          divergent = true;
-          break;
-        }
-      }
-    }
-
-    if (divergent) {
-      return { newMessages: fullMessageHistory, requiresReset: true };
-    }
-
-    const newMessages = fullMessageHistory.slice(this.history.length);
-    return { newMessages, requiresReset: false };
-  }
-
-  public update(
-    newMessages: CactusOAICompatibleMessage[],
-    assistantResponse: CactusOAICompatibleMessage
-  ) {
-    this.history.push(...newMessages, assistantResponse);
-  }
-
-  public reset() {
-    this.history = [];
-  }
-
-  public getMessages(): CactusOAICompatibleMessage[] {
-    return this.history;
-  }
-}