| GUA = [63, 0, 17, 34, 23, 58, 2, 16, 55, 59, 7, 56, 61, 47, 4, 8, 25, 38, 3, 48, 41, 37, 32, 1, 57, 39, 33, 30, 18, 45, 28, 14, 60, 15, 40, 5, 53, 43, 20, 10, 35, 49, 31, 62, 24, 6, 26, 22, 29, 46, 9, 36, 52, 11, 13, 44, 54, 27, 50, 19, 51, 12, 21, 42] # fmt: skip | |
| def bit2gua(bit6: int) -> str: | |
| return chr(0x4DC0 + GUA.index(bit6 & 0x3F)) | |
| def byte_segment(in_byte: bytes) -> bytearray: | |
| ret: bytearray = [] | |
| rem = 0 |
nvidia-smi said this required 11181MiB, at least to train on the sequence lengths of prompt that occurred initially in the alpaca dataset (~337 token long prompts).
You can get this down to about 10.9GB if (by modifying qlora.py) you run torch.cuda.empty_cache() after PEFT has been applied to your loaded model and before you begin training.
All instructions are written assuming your command-line shell is bash.
Clone repository:
| # original post: https://rentry.org/sd-loopback-wave | |
| # original author: https://rentry.org/AnimAnon | |
| import os | |
| import platform | |
| import numpy as np | |
| from tqdm import trange | |
| import math | |
| import subprocess as sp | |
| import string | |
| import random |
| # STEP 1: Load | |
| # Load documents using LangChain's DocumentLoaders | |
| # This is from https://langchain.readthedocs.io/en/latest/modules/document_loaders/examples/csv.html | |
| from langchain.document_loaders.csv_loader import CSVLoader | |
| loader = CSVLoader(file_path='./example_data/mlb_teams_2012.csv') | |
| data = loader.load() |
| import subprocess | |
| import openai | |
| import re | |
| from collections import defaultdict | |
| import textwrap | |
| import time | |
| import shlex | |
| openai.api_key = open("./openai_key.txt", "r").read().strip("\n") |
