File size: 583 Bytes
d8bc908
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
"""Fine-tuning utilities for ARB model with LoRA adapters.

Allows memory-efficient fine-tuning of 1.5B ternary model on 8GB VRAM
by freezing base weights and training only small float LoRA adapters.

Scripts:
    python -m training.finetuning.text        # Text/coding tasks
    python -m training.finetuning.vision      # Image understanding
    python -m training.finetuning.audio       # Speech/audio tasks
    python -m training.finetuning.diffusion   # Video/latent diffusion
"""
from .lora import TernaryLoRALayer, apply_lora_to_model, count_lora_params, save_lora, load_lora