Convert a raw UTF-8 text piece to GPT-2 byte-encoded form for BPE lookup.
| 352 | |
| 353 | // Convert a raw UTF-8 text piece to GPT-2 byte-encoded form for BPE lookup. |
| 354 | static std::string encode_gpt2_bpe(const std::string & text) { |
| 355 | std::string out; |
| 356 | out.reserve(text.size() * 2); // GPT-2 encoding may expand |
| 357 | for (uint8_t b : text) { |
| 358 | out += byte_to_gpt2_unicode(b); |
| 359 | } |
| 360 | return out; |
| 361 | } |
| 362 | |
| 363 | // Encode a single pre-tokenized piece using BPE merges. |
| 364 | std::vector<int32_t> Tokenizer::bpe_encode_piece(const std::string & piece) const { |
no test coverage detected