From ff5a3f0c09dfa0a8e0bf76d1748df5c6dee0e8ff Mon Sep 17 00:00:00 2001 From: goerch Date: Tue, 3 Oct 2023 09:16:26 +0200 Subject: Work on the BPE tokenizer (#3252) * Work on the BPE tokenizer Tokenizer tests work for Falcon-7B * Try to fix build problem * Fix debug assertion failure * Fix MSVC Unicode BOM problem * Cleanup and an improvement * Fix compiler warning * Cleanup * Test doesn't work over the full range of Unicodes * Update .gitignore and Makefile * Another Makefile rule * Testing Aquila * Moving byte decoding back to `token_to_piece` ... ... because everyone is using it. * Guarding some unusable code pathes * Streamlining code and adding some more assertions Important change: I'm classifying added tokens as control tokens now for BPE. * Adding a comment * Adding another assertion * Fixed vocabulary guarding assertions * Fix PR for recent change * Fix PR for recent change * Fix for compiler warning * Fix PR for recent change * Fix PR for recent change * Fix PR for recent change * Fix for compiler warning * Fixes for more compiler warnings * Remove unused code * Fix initialization of static maps * Add scores and token types back, adapt gptneox * Update llama.cpp Co-authored-by: Georgi Gerganov * Update unicode.h Co-authored-by: Georgi Gerganov * Update unicode.h Co-authored-by: Georgi Gerganov * Ported Starcoder and added some assertions * Fix coding style * Apply @jploski 's fix for missing tokens --------- Co-authored-by: Georgi Gerganov --- convert-starcoder-hf-to-gguf.py | 47 ++++++----------------------------------- 1 file changed, 7 insertions(+), 40 deletions(-) (limited to 'convert-starcoder-hf-to-gguf.py') diff --git a/convert-starcoder-hf-to-gguf.py b/convert-starcoder-hf-to-gguf.py index f469beb8..90fa0c32 100755 --- a/convert-starcoder-hf-to-gguf.py +++ b/convert-starcoder-hf-to-gguf.py @@ -20,28 +20,6 @@ if 'NO_LOCAL_GGUF' not in os.environ: import gguf -def bytes_to_unicode(): - # ref: https://github.com/openai/gpt-2/blob/master/src/encoder.py - """ - Returns list of utf-8 byte and a corresponding list of unicode strings. - The reversible bpe codes work on unicode strings. - This means you need a large # of unicode characters in your vocab if you want to avoid UNKs. - When you're at something like a 10B token dataset you end up needing around 5K for decent coverage. - This is a significant percentage of your normal, say, 32K bpe vocab. - To avoid that, we want lookup tables between utf-8 bytes and unicode strings. - And avoids mapping to whitespace/control characters the bpe code barfs on. - """ - bs = list(range(ord("!"), ord("~")+1))+list(range(ord("¡"), ord("¬")+1))+list(range(ord("®"), ord("ÿ")+1)) - cs = bs[:] - n = 0 - for b in range(2**8): - if b not in bs: - bs.append(b) - cs.append(2**8+n) - n += 1 - return dict(zip(bs, (chr(n) for n in cs))) - - def count_model_parts(dir_model: Path) -> int: num_parts = 0 for filename in os.listdir(dir_model): @@ -117,6 +95,8 @@ gguf_writer.add_file_type(ftype) print("gguf: get tokenizer metadata") tokens: list[bytearray] = [] +scores: list[float] = [] +toktypes: list[int] = [] # gpt2 tokenizer gguf_writer.add_tokenizer_model("gpt2") @@ -132,28 +112,15 @@ vocab_size = hparams.get("vocab_size", len(tokenizer.vocab)) assert max(tokenizer.vocab.values()) < vocab_size reverse_vocab = {id: encoded_tok for encoded_tok, id in tokenizer.vocab.items()} -byte_encoder = bytes_to_unicode() -byte_decoder = {v: k for k, v in byte_encoder.items()} for i in range(vocab_size): - if i in reverse_vocab: - try: - text = bytearray([byte_decoder[c] for c in reverse_vocab[i]]) - except KeyError: - text = bytearray() - for c in reverse_vocab[i]: - if ord(c) < 256: # single byte character - text.append(byte_decoder[ord(c)]) - else: # multibyte special token character - text.extend(c.encode('utf-8')) - else: - print(f"Key {i} not in tokenizer vocabulary. Padding with an arbitrary token.") - pad_token = f"[PAD{i}]".encode("utf8") - text = bytearray(pad_token) - - tokens.append(text) + tokens.append(reverse_vocab[i] if i in reverse_vocab else f"[PAD{i}]") + scores.append(0.0) # dummy + toktypes.append(gguf.TokenType.NORMAL) gguf_writer.add_token_list(tokens) +gguf_writer.add_token_scores(scores) +gguf_writer.add_token_types(toktypes) special_vocab = gguf.SpecialVocab(dir_model, load_merges = True) special_vocab.add_to_gguf(gguf_writer) -- cgit v1.2.3