yuchenlin

from transformers import GPT2LMHeadModel, GPT2Tokenizer
import torch 

# Test the model before saving
def chat_with_model(model, tokenizer, instruction, max_length=50):
    prompt = f"### User: {instruction}\n\n### AI: "
    inputs = tokenizer.encode(prompt, return_tensors='pt').to(device)
    outputs = model.generate(inputs, max_length=max_length, num_return_sequences=1, top_k=50)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

yuchenlin

ZebraLogic: Benchmarking the Logical Reasoning Ability of Language Models

yuchenlin

[
  {
    "session_id": "7f4abf64593c439f8b085585aeeda566",
    "model_input": "<|im_start|>user\nGive me an introduction over 200 words for ShangHai BMG Chemical Co., Ltd, a chemical company in Room 602, no 291 sikai road shanghai Shanghai,China<|im_end|>\n<|im_start|>assistant\n",
    "output": [
      "ShangHai BMG Chemical Co., Ltd. is a leading chemical company located in Room 602, No. 291 Sikai Road, Shanghai, China. Established in 2010, the company has since been dedicated to providing high-quality chemical products and services to its customers. With a strong R&D team and advanced production facilities, ShangHai BMG Chemical Co., Ltd. has become a trusted partner in the chemical industry, serving a wide range of clients across various sectors. The company's commitment to innovation, quality, and customer satisfaction has enabled it to maintain a competitive edge in the market.",
      "ShangHai BMG Chemical Co., Ltd. is a leading chemical company located in Room 602, No. 291 Sikai Road, Shanghai, Chin

yuchenlin

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import StoppingCriteria, StoppingCriteriaList, LogitsProcessor, LogitsProcessorList

model_path = "./qlora-out-hkg_300B/merged/"

tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto", trust_remote_code=True)

yuchenlin

"""
Decompose and memorize by program, only asking LLMs to do low-level computations. 
"""

import openai 
import re
import math  

from tenacity import (
    retry,

yuchenlin

import sys

acc = float(sys.argv[1]) # mnist acc
ta_acc = float(sys.argv[2]) # ta acc

t1=50.00
t2=90.00

tt1=30.0
tt2=60.0

yuchenlin

#!/bin/bash

echo "HW3 Report" > $vocareumReportFile
echo "Programming language..." >> $vocareumReportFile

filename=$(ls|grep NeuralNetwork)
if [[ $filename =~ (^|[[:space:]])"NeuralNetwork.py"($|[[:space:]]) ]]; then
	cmd="python NeuralNetwork.py train_image.csv train_label.csv test_image.csv"
    mnist_cmd="python NeuralNetwork.py grading_train_image.csv grading_train_label.csv grading_test_image.csv"
    ta_cmd="python NeuralNetwork.py grading_train_image.csv grading_train_label.csv additional_test_image.csv"

yuchenlin

"""
# stanza.download()
# http://nlp.stanford.edu/software/stanza/1.0.0/en/default.zip

Example usage:
CUDA_VISIBLE_DEVICES=1 \
python parsing.py \
    --input_corpus_path ./corpora/gkb_best_sent.txt \
    --output_json_path ./parses/gkb_best.parses.jsonl \
    --prefix gkb_best --num_shards 10000 --shard_id 0

yuchenlin

from argparse import Namespace

from numpy.core.defchararray import index
from semanticdebugger.debug_algs.cl_simple_alg import ContinualFinetuning
from tqdm import tqdm 
import torch 
from semanticdebugger.models.utils import trim_batch
import json
from semanticdebugger.debug_algs import run_lifelong_finetune

yuchenlin

#!/bin/bash
# Usage: bash mv.sh /path/to/src/ /path/to/target/

sourcedir=$1
targetdir=$2

filecount=$(find $sourcedir | wc -l)
echo $filecount # Print the number of the total files.
mkdir $targetdir
mv -v $sourcedir $targetdir | pv -l -s $filecount > /tmp/mv_log.txt