vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

convert : upgrade to sentencepiece v0.2.0

Browse source
This commit is contained in:
Francis Couture-Harpin 2024-04-30 15:01:23 -04:00
parent 0d720acb91
commit c33775bcc7
2 changed files with 13 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

20
convert.py
View file

@ -281,6 +281,7 @@ class Params:
n_experts = None n_experts = None
n_experts_used = None n_experts_used = None
f_rope_freq_base = None f_rope_freq_base = None
n_ff = None
# hack to determine LLaMA v1 vs v2 vs CodeLlama # hack to determine LLaMA v1 vs v2 vs CodeLlama
if config.get("moe"): if config.get("moe"):
@ -305,6 +306,8 @@ class Params:
n_experts_used = config["moe"]["num_experts_per_tok"] n_experts_used = config["moe"]["num_experts_per_tok"]
f_rope_freq_base = 1e6 f_rope_freq_base = 1e6
assert n_ff is not None
return Params( return Params(
n_vocab = model["tok_embeddings.weight"].shape[0], n_vocab = model["tok_embeddings.weight"].shape[0],
n_embd = config["dim"], n_embd = config["dim"],
@ -459,7 +462,8 @@ class SentencePieceVocab(Vocab):
# not found in alternate location either # not found in alternate location either
raise FileNotFoundError('Cannot find tokenizer.model') raise FileNotFoundError('Cannot find tokenizer.model')
self.sentencepiece_tokenizer = SentencePieceProcessor(str(fname_tokenizer)) self.sentencepiece_tokenizer = SentencePieceProcessor()
self.sentencepiece_tokenizer.LoadFromFile(fname_tokenizer)
vocab_size = self.sentencepiece_tokenizer.vocab_size() vocab_size = self.sentencepiece_tokenizer.vocab_size()
new_tokens = {id: piece for piece, id in added_tokens.items() if id >= vocab_size} new_tokens = {id: piece for piece, id in added_tokens.items() if id >= vocab_size}
@ -479,23 +483,23 @@ class SentencePieceVocab(Vocab):
def sentencepiece_tokens(self) -> Iterable[tuple[bytes, float, gguf.TokenType]]: def sentencepiece_tokens(self) -> Iterable[tuple[bytes, float, gguf.TokenType]]:
tokenizer = self.sentencepiece_tokenizer tokenizer = self.sentencepiece_tokenizer
for i in range(tokenizer.vocab_size()): for i in range(tokenizer.vocab_size()):
piece = tokenizer.id_to_piece(i) piece = tokenizer.IdToPiece(i)
text = piece.encode("utf-8") text = piece.encode("utf-8")
score: float = tokenizer.get_score(i) score: float = tokenizer.GetScore(i)
toktype = gguf.TokenType.NORMAL toktype = gguf.TokenType.NORMAL
if tokenizer.is_unknown(i): if tokenizer.IsUnknown(i):
toktype = gguf.TokenType.UNKNOWN toktype = gguf.TokenType.UNKNOWN
if tokenizer.is_control(i): if tokenizer.IsControl(i):
toktype = gguf.TokenType.CONTROL toktype = gguf.TokenType.CONTROL
# NOTE: I think added_tokens are user defined. # NOTE: I think added_tokens are user defined.
# ref: https://github.com/google/sentencepiece/blob/master/src/sentencepiece_model.proto # ref: https://github.com/google/sentencepiece/blob/master/src/sentencepiece_model.proto
# if tokenizer.is_user_defined(i): toktype = gguf.TokenType.USER_DEFINED # if tokenizer.is_user_defined(i): toktype = gguf.TokenType.USER_DEFINED
if tokenizer.is_unused(i): if tokenizer.IsUnused(i):
toktype = gguf.TokenType.UNUSED toktype = gguf.TokenType.UNUSED
if tokenizer.is_byte(i): if tokenizer.IsByte(i):
toktype = gguf.TokenType.BYTE toktype = gguf.TokenType.BYTE
yield text, score, toktype yield text, score, toktype
@ -904,7 +908,7 @@ class LazyUnpickler(pickle.Unpickler):
def rebuild_from_type_v2(func, new_type, args, state): def rebuild_from_type_v2(func, new_type, args, state):
return func(*args) return func(*args)
CLASSES = { CLASSES: dict[tuple[str, str], type[LazyTensor] | LazyStorageKind] = {
# getattr used here as a workaround for mypy not being smart enough to determine # getattr used here as a workaround for mypy not being smart enough to determine
# the staticmethods have a __func__ attribute. # the staticmethods have a __func__ attribute.
('torch._tensor', '_rebuild_from_type_v2'): getattr(rebuild_from_type_v2, '__func__'), ('torch._tensor', '_rebuild_from_type_v2'): getattr(rebuild_from_type_v2, '__func__'),

2
requirements/requirements-convert.txt
View file

@ -1,5 +1,5 @@
numpy~=1.24.4 numpy~=1.24.4
sentencepiece~=0.1.98 sentencepiece~=0.2.0
transformers>=4.35.2,<5.0.0 transformers>=4.35.2,<5.0.0
gguf>=0.1.0 gguf>=0.1.0
protobuf>=4.21.0,<5.0.0 protobuf>=4.21.0,<5.0.0