File size: 1,584 Bytes
d73e1e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
# coding=utf-8
""" Hgrn configuration"""

from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)

LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP = {}


class HgrnConfig(PretrainedConfig):
    model_type = "hgrn"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        pad_token_id=1,
        bos_token_id=0,
        eos_token_id=2,
        vocab_size=50272,
        use_cache=True,
        init_std=0.02,
        # model config
        decoder_embed_dim=1024,
        decoder_layers=24,
        add_bos_token=False,
        act_fun="swish",
        causal=True,
        use_triton=False,
        glu_act="swish",
        glu_dim=2816,
        bias=False,
        norm_type="layernorm",
        no_scale_embedding=False,
        **kwargs,
    ):
        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            **kwargs,
        )
        # hf origin
        self.vocab_size = vocab_size
        self.use_cache = use_cache
        self.init_std = init_std
        # add 
        self.decoder_embed_dim = decoder_embed_dim
        self.decoder_layers = decoder_layers
        self.add_bos_token = add_bos_token
        self.act_fun = act_fun
        self.causal = causal
        self.use_triton = use_triton
        self.glu_act = glu_act
        self.glu_dim = glu_dim
        self.bias = bias
        self.norm_type = norm_type
        self.no_scale_embedding = no_scale_embedding