convert : add support of codeqwen due to tokenizer (#6707)

* add support of codeqwen due to tokenizer

* override load_hparams

* fix typo

* fix load_params

* convert : fix whitespace

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Junyang Lin 2024-04-24 15:16:21 +08:00 committed by GitHub
parent c8297c6af5
commit 3fec68be4e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -363,6 +363,16 @@ class Model(ABC):
scores.append(-1000.0)
toktypes.append(SentencePieceTokenTypes.USER_DEFINED)
if vocab_size > len(tokens):
pad_count = vocab_size - len(tokens)
print(
f"Padding vocab with {pad_count} token(s) - [PAD1] through [PAD{pad_count}]"
)
for i in range(1, pad_count + 1):
tokens.append(f"[PAD{i}]")
scores.append(-1000.0)
toktypes.append(SentencePieceTokenTypes.UNUSED)
assert len(tokens) == vocab_size
self.gguf_writer.add_tokenizer_model("llama")
@ -1789,6 +1799,12 @@ class QwenModel(Model):
class Qwen2Model(Model):
model_arch = gguf.MODEL_ARCH.QWEN2
def set_vocab(self):
try:
self._set_vocab_sentencepiece()
except FileNotFoundError:
self._set_vocab_gpt2()
@Model.register("Qwen2MoeForCausalLM")
class Qwen2MoeModel(Model):