billju

import torch
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import LoraConfig, get_peft_model, prepare_model_for_int8_training
from trl import SFTTrainer
tokenizer = AutoTokenizer.from_pretrained('TinyPixel/Llama-2-7B-bf16-sharded')
tokenizer.pad_token_id = 0
tokenizer.padding_side = 'left'
train_dataset = load_dataset('json', data_files='alpaca_gpt4_data_zh.json', split='train')
model = AutoModelForCausalLM.from_pretrained(

billju

import os, json
from glob import glob
prompt_input = (
  "Below is an instruction that describes a task, paired with an input that provides further context. "
  "Write a response that appropriately completes the request.\n\n"
  "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:"
)
prompt_no_input = (
  "Below is an instruction that describes a task. "
  "Write a response that appropriately completes the request.\n\n"

billju

autotrain llm --train  \
	--data-path data \
	--epochs 3  \
	--block-size 1024 \
	--gradient-accumulation 4 \
	--logging_steps 500 \
	--lora-alpha 32  \
	--lora-dropout 0.05  \
	--lora-r 8  \
	--lr 2e-4 \

billju

import os, torch
from glob import glob
from peft import PeftModel
from transformers import PreTrainedTokenizerFast, AutoTokenizer, AutoModelForCausalLM, GenerationConfig, TextStreamer
pretrains = [os.path.basename(path).replace('models--','').replace('--','/') for path in glob(os.path.expanduser('~/.cache/huggingface/hub/models--*'))]
for i,pretrain in enumerate(pretrains): print(i,pretrain)
pretrain = pretrains[int(input('選擇預訓練'))]
tokenizer = AutoTokenizer.from_pretrained(pretrain)
model = AutoModelForCausalLM.from_pretrained(pretrain,device_map='auto',torch_dtype=torch.float16)
lora = input('LoRA路徑')

billju

from llama_cpp import Llama
from glob import glob
n_gpu_layers = 50 if input('使用GPU嗎(Y/n)').upper()=='Y' else 0
n_threads = None if n_gpu_layers else 4
ggmls = glob('*ggml*.bin')
for i,ggml in enumerate(ggmls): print(i, ggml)
ggml = ggmls[int(input('選擇模型'))]
llm = Llama(ggml, n_gpu_layers=n_gpu_layers, n_threads=n_threads, verbose=False)
while True:
	prompt = 'Q:{Q}?\nA:'.format(Q=input('\nyou>'))

billju

const fs = require('fs');
const { execSync } = require('child_process');
// 讀取版號
const package = JSON.parse(fs.readFileSync('package.json'));
let version = process.argv[2]; // 指令引數
if (!version) {
	// 自動增加版號
	version = parseInt(package.version.replace(/\./g, ''));
	version = (version + 1).toString();
	version = '0'.repeat(Math.max(3 - version.length)) + version;

billju

const puppeteer = require('puppeteer-core');
const fs = require('fs');
const path = require('path');
const glob = require('glob').sync;
const { execSync } = require('child_process');
const edgeExe = 'C:/Program Files (x86)/Microsoft/Edge/Application/msedge.exe';
/**
 * 全國電子公佈欄爬蟲
 * https://www.odbbs.gov.tw/odbbs/html/announce.jsp
 */

billju

# 下載 https://tesseract-ocr.github.io/tessdoc/Downloads.html
# 權重 https://github.com/tesseract-ocr/tessdata_best
# .box檔案格式：字 6 394 45 410 0
import os
import shutil
from PIL import Image
from glob import glob
lang = 'chi_tra'
font = 'ocrb'
dist = 'tessdata'

billju

import numpy as np
import pandas as pd
k = 2 # K個群集
max_iter = 10 # 最大迭代
data = np.array([1, 2, 3, 4, 11, 12])
last = np.random.choice(data, k, replace=False) # 初始化K個中心點
func = lambda x,y: np.abs(x-y) # 計算距離方法
steps = []
for _ in range(max_iter):
    # 對每筆資料求出最近的中心點

billju

import io
import pandas as pd
from itertools import permutations, combinations
CSV = """
交易紀錄 牛奶(A) 麵包(B) 餅乾(C) 柳橙汁(D) 汽水(E) 泡麵(F) 水果(G)
101 1 1 1 1 0 0 0
102 0 1 1 0 1 1 0
103 1 0 1 0 0 0 1
104 1 1 0 1 0 1 1
105 0 0 1 0 1 0 1