MCPcopy Create free account
hub / github.com/Tele-AI/Telechat / parse_args

Function parse_args

deepspeed-telechat/sft/main.py:40–182  ·  view source on GitHub ↗
()

Source from the content-addressed store, hash-verified

38
39
40def parse_args():
41 parser = argparse.ArgumentParser(
42 description=
43 "Finetune a transformers model on a causal language modeling task")
44 parser.add_argument('--data_path',
45 type=str,
46 required=True,
47 help='Path to the training dataset.')
48 parser.add_argument(
49 "--model_name_or_path",
50 type=str,
51 help=
52 "Path to pretrained model or model identifier from huggingface.co/models.",
53 required=True,
54 )
55 parser.add_argument(
56 "--per_device_train_batch_size",
57 type=int,
58 default=1,
59 help="Batch size (per device) for the training dataloader.",
60 )
61 parser.add_argument(
62 "--max_seq_len",
63 type=int,
64 default=512,
65 help="The maximum sequence length.",
66 )
67 parser.add_argument(
68 "--learning_rate",
69 type=float,
70 default=1e-5,
71 help=
72 "Initial learning rate (after the potential warmup period) to use.",
73 )
74 parser.add_argument("--weight_decay",
75 type=float,
76 default=0.,
77 help="Weight decay to use.")
78 parser.add_argument("--num_train_epochs",
79 type=int,
80 default=1,
81 help="Total number of training epochs to perform.")
82 parser.add_argument(
83 "--gradient_accumulation_steps",
84 type=int,
85 default=1,
86 help=
87 "Number of updates steps to accumulate before performing a backward/update pass.",
88 )
89 parser.add_argument(
90 "--lr_scheduler_type",
91 type=SchedulerType,
92 default="cosine",
93 help="The scheduler type to use.",
94 choices=[
95 "linear", "cosine", "cosine_with_restarts", "polynomial",
96 "constant", "constant_with_warmup"
97 ],

Callers 1

mainFunction · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected