react-native-executorch 0.5.1 → 0.5.3

package/third-party/ios/ExecutorchLib/ExecutorchLib/model/Model.cpp

@@ -1,11 +0,0 @@
-#include "Model.h"
-
-#include <executorch/extension/module/module.h>
-
-using ::executorch::extension::Module;
-
-Model::Model(const std::string &file_path) {
-  module_ = std::make_unique<Module>(file_path);
-}
-
-void Model::forward() { return; }

package/third-party/ios/ExecutorchLib/ExecutorchLib/model/Model.h

@@ -1,11 +0,0 @@
-#include <executorch/extension/module/module.h>
-
-class Model {
-public:
-  explicit Model(const std::string &file_path);
-
-  void forward();
-
-private:
-  std::unique_ptr<::executorch::extension::Module> module_;
-};

package/third-party/ios/ExecutorchLib/ExecutorchLib/tokenizer/base64.h

@@ -1,202 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-// @lint-ignore-every LICENSELINT
-/**************************************************************************
-   Copyright (c) 2023 sewenew
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
- *************************************************************************/
-
-#pragma once
-
-#include <cassert>
-#include <executorch/runtime/core/error.h>
-#include <executorch/runtime/core/result.h>
-#include <executorch/runtime/platform/assert.h>
-#include <string>
-#include <string_view>
-
-namespace executorch {
-namespace extension {
-namespace llm {
-using Error = executorch::runtime::Error;
-template <typename T> using Result = executorch::runtime::Result<T>;
-
-namespace base64 {
-
-Result<std::string> decode(const std::string_view &input);
-
-namespace detail {
-
-constexpr uint32_t DECODE_TABLE[] = {
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 62,  255,
-    255, 255, 63,  52,  53,  54,  55,  56,  57,  58,  59,  60,  61,  255, 255,
-    255, 255, 255, 255, 255, 0,   1,   2,   3,   4,   5,   6,   7,   8,   9,
-    10,  11,  12,  13,  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,
-    25,  255, 255, 255, 255, 255, 255, 26,  27,  28,  29,  30,  31,  32,  33,
-    34,  35,  36,  37,  38,  39,  40,  41,  42,  43,  44,  45,  46,  47,  48,
-    49,  50,  51,  255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-    255};
-
-inline Error validate(uint32_t v) {
-  ET_CHECK_OR_RETURN_ERROR(v != 255, InvalidArgument, "invalid char");
-  return Error::Ok;
-}
-
-inline Error decode(const std::string_view &input, std::string &output) {
-  ET_CHECK_OR_RETURN_ERROR(input.size() == 4, InvalidArgument,
-                           "input length must be 4, got %zu", input.size());
-
-  uint32_t val = 0;
-
-  uint8_t c = input[0];
-  auto v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = v;
-
-  c = input[1];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  c = input[2];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  c = input[3];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  output.push_back(static_cast<char>((val >> 16) & 0xFF));
-  output.push_back(static_cast<char>((val >> 8) & 0xFF));
-  output.push_back(static_cast<char>(val & 0xFF));
-  return Error::Ok;
-}
-
-inline Error decode_1_padding(const std::string_view &input,
-                              std::string &output) {
-  ET_CHECK_OR_RETURN_ERROR(input.size() == 3, InvalidArgument,
-                           "input length must be 3, got %zu", input.size());
-
-  uint32_t val = 0;
-
-  uint8_t c = input[0];
-  auto v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = v;
-
-  c = input[1];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  c = input[2];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  output.push_back(static_cast<char>((val >> 10) & 0xFF));
-  output.push_back(static_cast<char>((val >> 2) & 0xFF));
-  return Error::Ok;
-}
-
-inline Error decode_2_padding(const std::string_view &input,
-                              std::string &output) {
-  ET_CHECK_OR_RETURN_ERROR(input.size() == 2, InvalidArgument,
-                           "input length must be 2, got %zu", input.size());
-
-  uint32_t val = 0;
-
-  uint8_t c = input[0];
-  auto v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = v;
-
-  c = input[1];
-  v = DECODE_TABLE[c];
-  ET_CHECK_OK_OR_RETURN_ERROR(validate(v));
-  val = (val << 6) | v;
-
-  output.push_back(static_cast<char>((val >> 4) & 0xFF));
-  return Error::Ok;
-}
-
-} // namespace detail
-
-inline Result<std::string> decode(const std::string_view &input) {
-  ET_CHECK_OR_RETURN_ERROR(!input.empty(), InvalidArgument, "empty input");
-
-  // Faster than `input.size() % 4`.
-  ET_CHECK_OR_RETURN_ERROR(
-      (input.size() & 3) == 0 && input.size() >= 4, InvalidArgument,
-      "input length must be larger than 4 and is multiple of 4, got %zu",
-      input.size());
-
-  std::string output;
-  output.reserve(input.size() / 4 * 3);
-  auto idx = 0U;
-  for (; idx < input.size() - 4; idx += 4) {
-    ET_CHECK_OK_OR_RETURN_ERROR(detail::decode(input.substr(idx, 4), output));
-  }
-
-  // Last 4 bytes. Might contain paddings.
-  if (input[idx + 3] == '=') {
-    if (input[idx + 2] == '=') {
-      // Tow paddings.
-      ET_CHECK_OK_OR_RETURN_ERROR(
-          detail::decode_2_padding(input.substr(idx, 2), output));
-    } else {
-      // One padding.
-      ET_CHECK_OK_OR_RETURN_ERROR(
-          detail::decode_1_padding(input.substr(idx, 3), output));
-    }
-  } else {
-    // No padding.
-    ET_CHECK_OK_OR_RETURN_ERROR(detail::decode(input.substr(idx, 4), output));
-  }
-
-  return output;
-}
-
-} // namespace base64
-
-} // namespace llm
-} // namespace extension
-} // namespace executorch
-
-namespace torch {
-namespace executor {
-namespace base64 {
-// TODO(T197294990): Remove these deprecated aliases once all users have moved
-// to the new `::executorch` namespaces.
-using ::executorch::extension::llm::base64::decode;
-} // namespace base64
-} // namespace executor
-} // namespace torch

package/third-party/ios/ExecutorchLib/ExecutorchLib/tokenizer/bpe_tokenizer.cpp

@@ -1,313 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#include "bpe_tokenizer.h"
-
-#include <cstring>
-
-using ::executorch::runtime::Error;
-using ::executorch::runtime::Result;
-
-namespace executorch {
-namespace extension {
-namespace llm {
-
-static int compare_tokens(const void *a, const void *b) {
-  if (((TokenIndex *)a)->str == nullptr) {
-    return -1;
-  }
-  if (((TokenIndex *)b)->str == nullptr) {
-    return 1;
-  }
-  return strcmp(((TokenIndex *)a)->str, ((TokenIndex *)b)->str);
-}
-
-BPETokenizer::BPETokenizer() : Tokenizer() {
-  for (int i = 0; i < 256; i++) {
-    byte_pieces_[i * 2] = (unsigned char)i;
-    byte_pieces_[i * 2 + 1] = '\0';
-  }
-}
-
-/**
- * @brief Load the tokenizer from a file. The tokenizer file contains the
- * vocabulary and scores. The format is: the first integer is the maximum
- * token length, followed by a list of (word_len, word) pairs. Here we
- * are reading all the vocabulary into memory and keep it sorted for fast
- * lookup.
- *
- * @param tokenizer_path The path to the tokenizer file.
- * @return Error
- */
-Error BPETokenizer::load(const std::string &tokenizer_path) {
-  if (initialized_) {
-    ET_LOG(Info, "Tokenizer already initialized");
-    return Error::Ok;
-  }
-  // read in the file
-  FILE *file = fopen(tokenizer_path.c_str(), "rb");
-  if (!file) {
-    ET_LOG(Error, "couldn't load %s", tokenizer_path.c_str());
-    return Error::InvalidArgument;
-  }
-  int32_t metadata[4];
-  for (int i = 0; i < 4; i++) {
-    if (fread(metadata + i, sizeof(int32_t), 1, file) != 1) {
-      ET_LOG(Error,
-             "Failed to read the metadata at position %d, the tokenizer file "
-             "is not valid!",
-             i);
-      return Error::InvalidArgument;
-    }
-  }
-
-  // now we have two vocab_sizes one from the model and another from the
-  // tokenizer file.
-  int32_t tokenizer_vocab_size = metadata[0];
-  vocab_size_ = tokenizer_vocab_size;
-  bos_tok_ = metadata[1];
-  eos_tok_ = metadata[2];
-  max_token_length_ = metadata[3];
-
-  // allocate space for the vocabulary
-  vocab_ = std::make_unique<char *[]>(vocab_size_);
-  vocab_scores_ = std::make_unique<float[]>(vocab_size_);
-  sorted_vocab_ = std::make_unique<TokenIndex[]>(vocab_size_);
-
-  // read in the vocabulary
-  for (int i = 0; i < vocab_size_; i++) {
-    if (fread(vocab_scores_.get() + i, sizeof(float), 1, file) != 1) {
-      // This is allowed, we just pad the rest of the vocab with <pad> strings
-      std::string padding = "<pad>";
-      vocab_[i] = new char[padding.length() + 1];
-      strcpy(vocab_[i], padding.c_str());
-      vocab_[i][padding.length()] = '\0';
-      continue;
-    }
-    int32_t len;
-    if (fread(&len, sizeof(int32_t), 1, file) != 1) {
-      ET_LOG(Error, "Failed to read the length of the word at index %d", i);
-      return Error::InvalidArgument;
-    }
-    vocab_[i] = new char[len + 1];
-    if (fread(vocab_[i], len, 1, file) != 1) {
-      ET_LOG(Error, "Failed to read the word, total length %d, index %d\n", len,
-             i);
-      return Error::InvalidArgument;
-    }
-    vocab_[i][len] = '\0'; // add the string terminating token
-  }
-  fclose(file);
-
-  for (int32_t i = 0; i < vocab_size_; i++) {
-    sorted_vocab_[i].str = vocab_[i];
-    sorted_vocab_[i].id = i;
-  }
-  qsort(sorted_vocab_.get(), vocab_size_, sizeof(TokenIndex), compare_tokens);
-
-  initialized_ = true;
-  return Error::Ok;
-}
-
-BPETokenizer::~BPETokenizer() {
-  for (int i = 0; i < vocab_size_; i++) {
-    delete[] vocab_[i];
-  }
-}
-
-/**
- * @brief Decode a token into string.
- *
- * @param prev_token The previous token.
- * @param token The current token.
- * @return Result<std::string> A pointer to the string representation of the
- * token.
- */
-Result<std::string> BPETokenizer::decode(uint64_t prev_token,
-                                         uint64_t token) const {
-  ET_CHECK_OK_OR_RETURN_ERROR(Tokenizer::decode_verify(token));
-  const char *piece = vocab_[token];
-  // following BOS token, sentencepiece decoder strips any leading
-  // whitespace
-  if (prev_token == bos_tok_ && piece[0] == ' ') {
-    piece++;
-  }
-  // careful, some tokens designate raw bytes, and look like e.g. '<0x01>'
-  // parse this and convert and return the actual byte
-  unsigned char byte_val;
-  if (sscanf(piece, "<0x%02hhX>", &byte_val) == 1) {
-    piece = (char *)byte_pieces_ + byte_val * 2;
-  }
-  std::string res(piece);
-  return res;
-}
-
-static int32_t str_lookup(const char *str, TokenIndex *sorted_vocab,
-                          int32_t vocab_size) {
-  // efficiently find the perfect match for str in vocab, return its index or -1
-  // if not found
-  TokenIndex tok = {.str = str}; // acts as the key to search for
-  TokenIndex *res = (TokenIndex *)bsearch(&tok, sorted_vocab, vocab_size,
-                                          sizeof(TokenIndex), compare_tokens);
-  return res != nullptr ? res->id : -1;
-}
-
-/**
- * @brief Encode a string into a sequence of tokens.
- *
- * @param text The string to be encoded.
- * @param bos The number of BOS to prepend to the token list.
- * @param eos The number of EOS to append to the token list.
- * @param tokens The output tokens.
- * @param n_tokens The number of tokens.
- * @return Result<std::vector<uint64_t>>
- */
-Result<std::vector<uint64_t>>
-BPETokenizer::encode(const std::string &text, int8_t bos, int8_t eos) const {
-  if (!initialized_) {
-    ET_LOG(Error, "Tokenizer not initialized");
-    return Error::NotSupported;
-  }
-  // encode the string text (input) into an upper-bound preallocated tokens[]
-  // array bos != 0 means prepend the BOS token (=1), eos != 0 means append the
-  // EOS token (=2)
-  if (text.empty()) {
-    ET_LOG(Error, "cannot encode empty text");
-    return Error::InvalidArgument;
-  }
-
-  // create a temporary buffer that will store merge candidates of always two
-  // consecutive tokens *2 for concat, +1 for null terminator +2 for UTF8 (in
-  // case max_token_length is 1)
-  char *str_buffer = new char[max_token_length_ * 2 + 1 + 2];
-  size_t str_len = 0;
-
-  // start at 0 tokens
-  std::vector<uint64_t> tokens;
-
-  // add optional BOS token, if desired
-  if (bos >= 0) {
-    while (bos--) {
-      tokens.push_back(bos_tok_);
-    }
-  } else {
-    ET_LOG(Error, "bos %d should be >= 0", bos);
-    return Error::InvalidArgument;
-  }
-
-  // add_dummy_prefix is true by default
-  // so prepend a dummy prefix token to the input string, but only if text != ""
-  // TODO: pretty sure this isn't correct in the general case but I don't have
-  // the energy to read more of the sentencepiece code to figure out what it's
-  // doing
-  const char *space = " ";
-  if (text[0] != '\0') {
-    int dummy_prefix = str_lookup(space, sorted_vocab_.get(), vocab_size_);
-    tokens.push_back(dummy_prefix);
-  }
-
-  // Okay UTF-8 time. This will get messy. Here is the reference from Wikipedia:
-  // Code point ↔ UTF-8 conversion
-  // First code point  Last code point  Byte 1  Byte 2  Byte 3  Byte 4
-  // U+0000  U+007F      0xxxxxxx
-  // U+0080  U+07FF      110xxxxx  10xxxxxx
-  // U+0800  U+FFFF      1110xxxx  10xxxxxx  10xxxxxx
-  // U+10000  U+10FFFF    11110xxx  10xxxxxx  10xxxxxx  10xxxxxx
-
-  // process the raw (UTF-8) byte sequence of the input string
-  for (const char *c = text.c_str(); *c != '\0'; c++) {
-    // reset buffer if the current byte is ASCII or a leading byte
-    // 0xC0 is 11000000, so (*c & 0xC0) keeps the first 2 bits and zeros the
-    // rest 0x80 is 10000000 in UTF-8, all continuation bytes start with "10" in
-    // first two bits so in English this is: "if this byte is not a continuation
-    // byte"
-    if ((*c & 0xC0) != 0x80) {
-      // this byte must be either a leading byte (11...) or an ASCII char
-      // (0x...)
-      // => reset our location, as we're starting a new UTF-8 codepoint
-      str_len = 0;
-    }
-
-    // append the current byte to the buffer
-    str_buffer[str_len++] =
-        *c; // ++ is post-increment, incremented after this line
-    str_buffer[str_len] = '\0';
-
-    // while the next character is a continuation byte, continue appending
-    // but if there are too many of them, just stop to avoid overruning
-    // str_buffer size.
-    if ((*(c + 1) & 0xC0) == 0x80 && str_len < 4) {
-      continue;
-    }
-
-    // ok c+1 is not a continuation byte, so we've read in a full codepoint
-    int id = str_lookup(str_buffer, sorted_vocab_.get(), vocab_size_);
-    if (id != -1) {
-      // we found this codepoint in vocab, add it as a token
-      tokens.push_back(id);
-    } else {
-      // byte_fallback encoding: just encode each byte as a token
-      // +3 is here because the first 3 vocab elements are <unk>, <s>, </s>
-      // so the individual bytes only start at index 3
-      for (int i = 0; i < str_len; i++) {
-        tokens.push_back((unsigned char)str_buffer[i] + 3);
-      }
-    }
-    str_len = 0; // protect against a sequence of stray UTF8 continuation bytes
-  }
-
-  // merge the best consecutive pair each iteration, according the scores in
-  // vocab_scores
-  while (1) {
-    float best_score = -1e10;
-    int best_id = -1;
-    int best_idx = -1;
-
-    for (int i = 0; i < tokens.size() - 1; i++) {
-      // check if we can merge the pair (tokens[i], tokens[i+1])
-      snprintf(str_buffer, max_token_length_ * 2 + 3, "%s%s", vocab_[tokens[i]],
-               vocab_[tokens[i + 1]]);
-      int id = str_lookup(str_buffer, sorted_vocab_.get(), vocab_size_);
-      if (id != -1 && vocab_scores_[id] > best_score) {
-        // this merge pair exists in vocab! record its score and position
-        best_score = vocab_scores_[id];
-        best_id = id;
-        best_idx = i;
-      }
-    }
-
-    if (best_idx == -1) {
-      break; // we couldn't find any more pairs to merge, so we're done
-    }
-
-    // merge the consecutive pair (best_idx, best_idx+1) into new token best_id
-    tokens[best_idx] = best_id;
-    // delete token at position best_idx+1, shift the entire sequence back 1
-    for (int i = best_idx + 1; i < tokens.size() - 1; i++) {
-      tokens[i] = tokens[i + 1];
-    }
-    tokens.pop_back(); // token length decreased
-  }
-
-  // add optional EOS (=2) token, if desired
-  if (eos >= 0) {
-    while (eos--) {
-      tokens.push_back(eos_tok_);
-    }
-  } else {
-    ET_LOG(Error, "eos %d should be >= 0", eos);
-    return Error::InvalidArgument;
-  }
-
-  delete[] str_buffer;
-  return Result(tokens);
-}
-
-} // namespace llm
-} // namespace extension
-} // namespace executorch

package/third-party/ios/ExecutorchLib/ExecutorchLib/tokenizer/bpe_tokenizer.h

@@ -1,57 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#pragma once
-
-#include "tokenizer.h"
-#include <memory>
-
-namespace executorch {
-namespace extension {
-namespace llm {
-
-struct TokenIndex {
-  const char *str;
-  int32_t id;
-};
-
-// A simple Byte Pair Encoding (BPE) Tokenizer. Note that the current C++ code
-// won't work with this class, it needs to go through tokenizer.py first.
-class BPETokenizer : public Tokenizer {
-public:
-  explicit BPETokenizer();
-  ~BPETokenizer() override;
-
-  ::executorch::runtime::Error load(const std::string &tokenizer_path) override;
-
-  ::executorch::runtime::Result<std::vector<uint64_t>>
-  encode(const std::string &input, int8_t bos, int8_t eos) const override;
-
-  ::executorch::runtime::Result<std::string>
-  decode(uint64_t prev_token, uint64_t token) const override;
-
-private:
-  std::unique_ptr<char *[]> vocab_ = nullptr;
-  std::unique_ptr<float[]> vocab_scores_ = nullptr;
-  std::unique_ptr<TokenIndex[]> sorted_vocab_ = nullptr;
-  unsigned int max_token_length_ = 0;
-  unsigned char byte_pieces_[512]; // stores all single-byte strings
-};
-
-} // namespace llm
-} // namespace extension
-} // namespace executorch
-
-namespace torch {
-namespace executor {
-// TODO(T197294990): Remove these deprecated aliases once all users have moved
-// to the new `::executorch` namespaces.
-using ::executorch::extension::llm::BPETokenizer;
-using ::executorch::extension::llm::TokenIndex;
-} // namespace executor
-} // namespace torch

package/third-party/ios/ExecutorchLib/ExecutorchLib/tokenizer/llama_tiktoken.cpp

@@ -1,78 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#include "llama_tiktoken.h"
-
-namespace example {
-
-using ::executorch::extension::llm::Tiktoken;
-
-namespace {
-static constexpr int32_t kSpecialTokensSize = 256;
-static constexpr size_t kBOSTokenIndex = 0;
-static constexpr size_t kEOSTokenIndex = 1;
-
-static inline std::unique_ptr<std::vector<std::string>>
-_get_default_special_tokens() {
-  auto special_tokens =
-      std::make_unique<std::vector<std::string>>(std::vector<std::string>{
-          "<|begin_of_text|>", "<|end_of_text|>",
-          "<|reserved_special_token_0|>", "<|reserved_special_token_1|>",
-          "<|finetune_right_pad_id|>", "<|step_id|>", "<|start_header_id|>",
-          "<|end_header_id|>", "<|eom_id|>", "<|eot_id|>", "<|python_tag|>"});
-  // pad the rest of the special tokens with reserved tokens
-  ssize_t reserved_special_token_num = 2;
-  while (special_tokens->size() < kSpecialTokensSize) {
-    special_tokens->emplace_back("<|reserved_special_token_" +
-                                 std::to_string(reserved_special_token_num++) +
-                                 "|>");
-  }
-  return special_tokens;
-}
-
-static inline std::unique_ptr<std::vector<std::string>>
-_get_multimodal_special_tokens() {
-  auto special_tokens =
-      std::make_unique<std::vector<std::string>>(std::vector<std::string>{
-          "<|begin_of_text|>", "<|end_of_text|>",
-          "<|reserved_special_token_0|>", "<|reserved_special_token_1|>",
-          "<|reserved_special_token_2|>", "<|reserved_special_token_3|>",
-          "<|start_header_id|>", "<|end_header_id|>", "<|eom_id|>",
-          "<|eot_id|>", "<|image|>"});
-
-  // pad the rest of the special tokens with reserved tokens except the last
-  // one
-  ssize_t reserved_special_token_num = 4;
-  while (special_tokens->size() < kSpecialTokensSize - 1) {
-    special_tokens->emplace_back("<|reserved_special_token_" +
-                                 std::to_string(reserved_special_token_num++) +
-                                 "|>");
-  }
-
-  special_tokens->emplace_back("<|python_tag|>");
-
-  return special_tokens;
-}
-
-std::unique_ptr<std::vector<std::string>> _get_special_tokens(Version version) {
-  switch (version) {
-  case Version::Multimodal:
-    return _get_multimodal_special_tokens();
-  default:
-    return _get_default_special_tokens();
-  }
-}
-
-} // namespace
-
-std::unique_ptr<Tiktoken> get_tiktoken_for_llama(Version version) {
-  return std::make_unique<Tiktoken>(_get_special_tokens(version),
-                                    kBOSTokenIndex, kEOSTokenIndex);
-}
-
-} // namespace example

package/third-party/ios/ExecutorchLib/ExecutorchLib/tokenizer/llama_tiktoken.h

@@ -1,23 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#pragma once
-
-#include "tiktoken.h"
-
-namespace example {
-
-enum class Version {
-  Default,
-  Multimodal,
-};
-
-std::unique_ptr<::executorch::extension::llm::Tiktoken>
-get_tiktoken_for_llama(Version version = Version::Default);
-
-} // namespace example