Buckets:

download
raw
1.92 kB
import argparse
DEFAULT_MODEL_PATH = "GreenBitAI/Qwen-1.5-0.5B-layer-mix-bpw-2.2"
DEFAULT_SEQLEN = 2048
def setup_shared_arg_parser(parser_name="Shared argument parser for green-bit-llm scripts"):
"""Set up and return the argument parser with shared arguments."""
parser = argparse.ArgumentParser(description=parser_name)
parser.add_argument(
"--model",
type=str,
default=DEFAULT_MODEL_PATH,
help="The path to the local model directory or Hugging Face repo.",
)
parser.add_argument(
"--trust-remote-code",
action="store_true",
help="Enable trusting remote code for tokenizer",
)
parser.add_argument(
"--use-flash-attention-2",
action="store_true",
help="Enable using flash attention v2",
)
parser.add_argument(
"--seqlen",
type=int,
default=DEFAULT_SEQLEN,
help="Sequence length"
)
parser.add_argument(
"--save-dir",
type=str,
default="output/",
help="Specify save dir for eval results.",
)
parser.add_argument(
"--save-step",
type=int,
default=500,
help="Specify how many steps to save a checkpoint.",
)
parser.add_argument(
"--dtype",
type=str,
choices=["float", "half"],
default="half",
help="Dtype used in optimizer.",
)
parser.add_argument(
"--dataset",
type=str,
default="tatsu-lab/alpaca",
help="Dataset name for finetuning",
)
parser.add_argument(
"--optimizer",
type=str,
default="DiodeMix",
help="Optimizer to use: 1. DiodeMix, 2. AdamW8bit"
)
parser.add_argument(
"--batch-size",
type=int,
default=4,
help="Batch size"
)
parser.add_argument("--weight_decay", type=float, default=0.0)
return parser

Xet Storage Details

Size:
1.92 kB
·
Xet hash:
5aca0618e19a867c35ab6ae01a09b974bc87e78f750aadf05bf6c6ecf334ba21

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.