Download kdflow/arguments/model_args.py from codemaivanngu/ctopd: direct link, hf CLI and curl.
- Browser
- Download file 1.16 kB
-
https://huggingface.co/codemaivanngu/ctopd/resolve/main/kdflow/arguments/model_args.py
- Command line
-
hf download hf://codemaivanngu/ctopd/kdflow/arguments/model_args.py
-
curl -L -o model_args.py https://huggingface.co/codemaivanngu/ctopd/resolve/main/kdflow/arguments/model_args.py
1.16 kB
| from dataclasses import dataclass, field | |
| from typing import Optional | |
| class ModelArguments: | |
| """ Arguments for model.""" | |
| student_name_or_path: str = field( | |
| default=None, | |
| metadata={"help": "Student model name or path."} | |
| ) | |
| teacher_name_or_path: str = field( | |
| default=None, | |
| metadata={"help": "Teacher model name or path."} | |
| ) | |
| attn_implementation: str = field( | |
| default="flash_attention_2", | |
| metadata={"help": "Attention implementation (e.g., eager, flash_attention_2, flash_attention_3, kernels-community/vllm-flash-attn3)."} | |
| ) | |
| use_liger_kernel: bool = field( | |
| default=False, | |
| metadata={"help": "Use Liger Kernel in LLM."} | |
| ) | |
| lora_rank: int = field( | |
| default=0 | |
| ) | |
| lora_alpha: int = field( | |
| default=16 | |
| ) | |
| target_modules: str = field( | |
| default="all-linear", | |
| ) | |
| lora_dropout: float = field( | |
| default=0.0, | |
| ) | |
| disable_fast_tokenizer: bool = field( | |
| default=False | |
| ) | |
| enable_thinking: bool = field( | |
| default=False, | |
| ) | |
| ring_attn_size: int = field( | |
| default=1 | |
| ) |