add chatglm3-6b model support huggingface model:
https://hf-mirror.com/THUDM/chatglm3-6b Signed-off-by: XingXing Qiao <qiaoxx@dingdao.com>
This commit is contained in:
XingXing Qiao
parent
b864b50ce5
commit
6630a2da48
6 changed files with 398 additions and 6 deletions
|
|
@ -79,7 +79,7 @@ class Model:
|
|||
if not self.is_safetensors:
|
||||
self.part_names = Model.get_model_part_names(self.dir_model, ".bin")
|
||||
self.hparams = Model.load_hparams(self.dir_model)
|
||||
self.block_count = self.find_hparam(["n_layers", "num_hidden_layers", "n_layer"])
|
||||
self.block_count = self.find_hparam(["n_layers", "num_hidden_layers", "n_layer", "num_layers"])
|
||||
self.tensor_map = gguf.get_tensor_name_map(self.model_arch, self.block_count)
|
||||
self.tensor_names = None
|
||||
if self.ftype == gguf.LlamaFileType.GUESSED:
|
||||
|
|
@ -2710,6 +2710,167 @@ class DeepseekV2Model(Model):
|
|||
raise ValueError(f"Unprocessed experts: {experts}")
|
||||
|
||||
|
||||
@Model.register("ChatGLMModel")
|
||||
class ChatGLMModel(Model):
|
||||
model_arch = gguf.MODEL_ARCH.CHATGLM
|
||||
|
||||
def set_vocab(self):
|
||||
dir_model = self.dir_model
|
||||
hparams = self.hparams
|
||||
tokens: list[bytearray] = []
|
||||
toktypes: list[int] = []
|
||||
scores: list[float] = []
|
||||
|
||||
from transformers import AutoTokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(dir_model, trust_remote_code=True)
|
||||
vocab_size = hparams.get("padded_vocab_size", len(tokenizer.get_vocab()))
|
||||
assert max(tokenizer.get_vocab().values()) < vocab_size
|
||||
|
||||
reverse_vocab = {id_: encoded_tok for encoded_tok, id_ in tokenizer.get_vocab().items()}
|
||||
|
||||
for token_id in range(vocab_size):
|
||||
piece = tokenizer._convert_id_to_token(token_id)
|
||||
if token_id == 0:
|
||||
piece = "<unk>"
|
||||
elif token_id == 1:
|
||||
piece = "<bos>"
|
||||
elif token_id == 2:
|
||||
piece = "<eos>"
|
||||
|
||||
text = piece.encode("utf-8")
|
||||
score = 0.0
|
||||
if len(piece) != 0 and token_id < 64789:
|
||||
score = tokenizer.tokenizer.sp_model.get_score(token_id)
|
||||
|
||||
if len(piece) == 0:
|
||||
text = f"[PAD{token_id}]".encode("utf-8")
|
||||
|
||||
if token_id >= 64789:
|
||||
toktype = SentencePieceTokenTypes.UNKNOWN
|
||||
tokens.append(text)
|
||||
scores.append(score)
|
||||
toktypes.append(toktype)
|
||||
continue
|
||||
|
||||
toktype = SentencePieceTokenTypes.NORMAL
|
||||
if tokenizer.tokenizer.sp_model.is_unknown(token_id):
|
||||
toktype = SentencePieceTokenTypes.UNKNOWN
|
||||
elif tokenizer.tokenizer.sp_model.is_control(token_id):
|
||||
toktype = SentencePieceTokenTypes.CONTROL
|
||||
elif tokenizer.tokenizer.sp_model.is_unused(token_id):
|
||||
toktype = SentencePieceTokenTypes.UNUSED
|
||||
elif tokenizer.tokenizer.sp_model.is_byte(token_id):
|
||||
toktype = SentencePieceTokenTypes.BYTE
|
||||
|
||||
tokens.append(text)
|
||||
scores.append(score)
|
||||
toktypes.append(toktype)
|
||||
|
||||
self.gguf_writer.add_tokenizer_model("llama")
|
||||
self.gguf_writer.add_token_list(tokens)
|
||||
self.gguf_writer.add_token_scores(scores)
|
||||
self.gguf_writer.add_token_types(toktypes)
|
||||
|
||||
special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
|
||||
special_vocab.add_to_gguf(self.gguf_writer)
|
||||
|
||||
def set_gguf_parameters(self):
|
||||
self.gguf_writer.add_name("ChatGLM-6b-chat")
|
||||
n_embed = self.hparams.get("hidden_size", self.hparams.get("n_embed"))
|
||||
n_head = self.hparams.get("n_head", self.hparams.get("num_attention_heads"))
|
||||
n_head_kv = self.hparams.get("multi_query_group_num", n_head)
|
||||
self.gguf_writer.add_context_length(self.hparams.get("seq_length", n_embed))
|
||||
self.gguf_writer.add_embedding_length(n_embed)
|
||||
self.gguf_writer.add_feed_forward_length(self.hparams.get("ffn_hidden_size", 4 * n_embed))
|
||||
self.gguf_writer.add_block_count(self.hparams["num_layers"])
|
||||
self.gguf_writer.add_head_count(n_head)
|
||||
self.gguf_writer.add_head_count_kv(n_head_kv)
|
||||
self.gguf_writer.add_layer_norm_rms_eps(self.hparams["layernorm_epsilon"])
|
||||
self.gguf_writer.add_file_type(self.ftype)
|
||||
self.gguf_writer.add_rope_dimension_count(64)
|
||||
self.gguf_writer.add_add_bos_token(False)
|
||||
|
||||
def write_tensors(self):
|
||||
block_count = self.hparams["num_layers"]
|
||||
tensors = dict(self.get_tensors())
|
||||
tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
|
||||
has_lm_head = True
|
||||
n_head = self.hparams.get("n_head", self.hparams.get("num_attention_heads"))
|
||||
n_embed = self.hparams.get("hidden_size", self.hparams.get("n_embed"))
|
||||
|
||||
for name, data_torch in tensors.items():
|
||||
if name.endswith(".rotary_pos_emb.inv_freq"):
|
||||
continue
|
||||
|
||||
if "lm_head.weight" not in tensors.keys() and "output.weight" not in tensors.keys():
|
||||
has_lm_head = False
|
||||
|
||||
name = re.sub(r'transformer\.', '', name)
|
||||
|
||||
old_dtype = data_torch.dtype
|
||||
|
||||
# convert any unsupported data types to float32
|
||||
if data_torch.dtype not in (torch.float16, torch.float32):
|
||||
data_torch = data_torch.to(torch.float32)
|
||||
|
||||
data = data_torch.squeeze().numpy()
|
||||
|
||||
if re.match(r"h\.\d+\.self_attention\.query_key_value\.weight", name):
|
||||
# Map bloom-style qkv_linear to gpt-style qkv_linear
|
||||
# bloom: https://github.com/huggingface/transformers/blob/main/src/transformers/models/bloom/modeling_bloom.py#L238-L252 # noqa
|
||||
# gpt-2: https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt2/modeling_gpt2.py#L312 # noqa
|
||||
qkv_weights = data.reshape((n_head, 3, n_embed // n_head, n_embed))
|
||||
data = np.concatenate(
|
||||
(
|
||||
qkv_weights[:, 0, :, :].reshape((-1, n_embed)),
|
||||
qkv_weights[:, 1, :, :].reshape((-1, n_embed)),
|
||||
qkv_weights[:, 2, :, :].reshape((-1, n_embed)),
|
||||
),
|
||||
axis=0,
|
||||
)
|
||||
print("re-format attention.linear_qkv.weight")
|
||||
elif re.match(r"h\.\d+\.self_attention\.query_key_value\.bias", name):
|
||||
qkv_bias = data.reshape((n_head, 3, n_embed // n_head))
|
||||
data = np.concatenate(
|
||||
(
|
||||
qkv_bias[:, 0, :].reshape((n_embed,)),
|
||||
qkv_bias[:, 1, :].reshape((n_embed,)),
|
||||
qkv_bias[:, 2, :].reshape((n_embed,)),
|
||||
),
|
||||
axis=0,
|
||||
)
|
||||
print("re-format attention.linear_qkv.bias")
|
||||
|
||||
# map tensor names
|
||||
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
|
||||
if new_name is None:
|
||||
print(f"Can not map tensor {name!r}")
|
||||
sys.exit()
|
||||
|
||||
n_dims = len(data.shape)
|
||||
data_dtype = data.dtype
|
||||
|
||||
# if f32 desired, convert any float16 to float32
|
||||
if self.ftype == 0 and data_dtype == np.float16:
|
||||
data = data.astype(np.float32)
|
||||
|
||||
# TODO: Why cant we use these float16 as-is? There should be not reason to store float16 as float32
|
||||
if self.ftype == 1 and data_dtype == np.float16 and n_dims == 1:
|
||||
data = data.astype(np.float32)
|
||||
|
||||
# if f16 desired, convert any float32 2-dim weight tensors to float16
|
||||
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and n_dims == 2:
|
||||
data = data.astype(np.float16)
|
||||
|
||||
print(f"=> {new_name}, shape = {data.shape}, {old_dtype} --> {data.dtype}")
|
||||
|
||||
self.gguf_writer.add_tensor(new_name, data)
|
||||
|
||||
if not has_lm_head and name == "word_embeddings.weight":
|
||||
self.gguf_writer.add_tensor("output.weight", data)
|
||||
print(name, f"=> output.weight, shape = {data.shape}, {old_dtype} --> {data.dtype}")
|
||||
|
||||
|
||||
###### CONVERSION LOGIC ######
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue