Skip to content

Config

helical.models.scgpt.scGPTConfig

Configuration class to use the scGPT Model.

Parameters:

Name Type Description Default
pad_token str

The padding token

 "<pad>"
batch_size int

The batch size

 24
emb_mode Literal['cls', 'cell', 'gene']

The embedding mode to use. "cls" uses the cls token, "cell" uses the cell token, "gene" uses the gene token.

 "cell"
nlayers int

The number of layers

 12
nheads int

The number of heads

 8
embsize int

The embedding size

 512
d_hid int

The hidden dimension

 512
dropout float

The dropout rate

 0.2
n_layers_cls int

The number of classification layers

 3
mask_value int

The mask value

 -1
pad_value int

The padding value

 -2
world_size int

The world size

 8
accelerator bool

The accelerator configuration. By default same device as model.

 False
device Literal['cpu', 'cuda']

The device to use. Either use "cuda" or "cpu".

 "cpu"
binning_seed int

The seed for the binning

 123

Returns:

Type Description
scGPTConfig

The scGPT configuration object
Notes

This configuration contains all the default parameteres that have been used in the original scGPT repository.

Source code in helical/models/scgpt/scgpt_config.py 
 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
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
class scGPTConfig:
    """
    Configuration class to use the scGPT Model.

    Parameters
    ----------
    pad_token : str, optional, default="<pad>"
        The padding token
    batch_size : int, optional, default=24
        The batch size
    emb_mode : Literal["cls", "cell", "gene"], optional, default="cell"
        The embedding mode to use. "cls" uses the cls token, "cell" uses the cell token, "gene" uses the gene token.
    nlayers : int, optional, default=12
        The number of layers
    nheads : int, optional, default=8
        The number of heads
    embsize : int, optional, default=512
        The embedding size
    d_hid : int, optional, default=512
        The hidden dimension
    dropout : float, optional, default=0.2
        The dropout rate
    n_layers_cls : int, optional, default=3
        The number of classification layers
    mask_value : int, optional, default=-1
        The mask value
    pad_value : int, optional, default=-2
        The padding value
    world_size : int, optional, default=8
        The world size
    accelerator : bool, optional, default=False
        The accelerator configuration. By default same device as model.
    device : Literal["cpu", "cuda"], optional, default="cpu"
        The device to use. Either use "cuda" or "cpu".
    binning_seed : int, optional, default=123
        The seed for the binning

    Returns
    -------
    scGPTConfig
        The scGPT configuration object

    Notes
    -----
    This configuration contains all the default parameteres that have been used in the original scGPT repository.

    """

    def __init__(
        self,
        pad_token: str = "<pad>",
        batch_size: int = 24,
        emb_mode: Literal["cls", "cell", "gene"] = "cell",
        nlayers: int = 12,
        nheads: int = 8,
        embsize: int = 512,
        d_hid: int = 512,
        dropout: float = 0.2,
        n_layers_cls: int = 3,
        mask_value: int = -1,
        pad_value: int = -2,
        world_size: int = 8,
        accelerator: Optional[bool] = False,
        device: Literal["cpu", "cuda"] = "cpu",
        binning_seed: int = 123,
    ):

        model_name = "best_model"  # TODO: Include more models
        list_of_files_to_download = [
            "scgpt/scGPT_CP/vocab.json",
            f"scgpt/scGPT_CP/{model_name}.pt",
        ]
        model_path = Path(CACHE_DIR_HELICAL, "scgpt/scGPT_CP", f"{model_name}.pt")

        self.config = {
            "model_path": model_path,
            "list_of_files_to_download": list_of_files_to_download,
            "pad_token": pad_token,
            "batch_size": batch_size,
            "emb_mode": emb_mode,
            "nlayers": nlayers,
            "nheads": nheads,
            "embsize": embsize,
            "d_hid": d_hid,
            "dropout": dropout,
            "n_layers_cls": n_layers_cls,
            "mask_value": mask_value,
            "pad_value": pad_value,
            "world_size": world_size,
            "accelerator": accelerator,
            "device": device,
            "MAX_LENGTH": 1200,
            "binning_seed": binning_seed,
        }