Datasets:

Modalities:
Text
Formats:
json
ArXiv:
Libraries:
Datasets
pandas
License:
BLEnD / model_inference.py
nayeon212's picture
ADD: Data & Code
605bdce verified
from utils import *
parser = argparse.ArgumentParser(description='Choose your model(s) & language(s)')
parser.add_argument('--model',type=str,
help='Provide the model you want to use. Check and choose from the key values of the MODEL_PATHS variable. If you want to test on multiple models, provide multiple model names with ", " between each (e.g., "gpt-4-0125-preview, aya-101").')
parser.add_argument('--language',type=str,default=None,
help='Provide the language you want to test on. Check and choose from the first values of the LANG_COUNTRY variable. If you want to test on multiple languages, provide multiple languages with ", " between each (e.g., "English, Korean").')
parser.add_argument('--country',type=str,default=None,
help='Provide the country you want to test on. Check and choose from the second values of the LANG_COUNTRY variable. If you want to test on multiple countries, provide multiple countries with ", " between each (e.g., "UK, South Korea"). Make sure you have the same number of countries and languages provided. The language-country pair do not have to be identical with the pairs within the LANG_COUNTRY variable.')
parser.add_argument('--question_dir',type=str,default=None,
help='Provide the directory name with (translated) questions.')
parser.add_argument('--question_file',type=str,default=None,
help='Provide the csv file name with (translated) questions.')
parser.add_argument('--question_col',type=str,default=None,
help='Provide the column name from the given csv file name with (translated) questions.')
parser.add_argument('--prompt_dir',type=str,default=None,
help='Provide the directory where the propmts are saved.')
parser.add_argument('--prompt_file',type=str,default=None,
help='Provide the name of the csv file where the propmts are saved.')
parser.add_argument('--prompt_no',type=str,default=None,
help='Provide the propmt id (ex. inst-1, inst-2, pers-1, etc.)')
parser.add_argument('--id_col',type=str,default="ID",
help='Provide the column name from the given csv file name with question IDs.')
parser.add_argument('--output_dir',type=str,default='./model_inference_results',
help='Provide the directory for the output files to be saved.')
parser.add.argument('--output_file',type=str,default=None,
help='Provide the name of the output file.')
parser.add_argument('--model_cache_dir',type=str,default='.cache',
help='Provide the directory saving model caches.')
parser.add_argument("--gpt_azure", type=str2bool, nargs='?',
const=True, default=False,
help="Whether you are using the AzureOpenAI for GPT-models' response generation.")
parser.add_argument('--temperature',type=int,default=0,
help='Provide generation temperature for GPT models.')
parser.add_argument('--top_p',type=int,default=0,
help='Provide generation top_p for GPT models.')
args = parser.parse_args()
def make_prompt(question,prompt_no,language,country,prompt_sheet):
prompt = prompt_sheet[prompt_sheet['id']==prompt_no]
if language == 'English':
prompt = prompt['English'].values[0]
else:
prompt = prompt['Translation'].values[0]
return prompt.replace('{q}',question)
def generate_response(model_name,model_path,tokenizer,model,language,country,q_df,q_col,id_col,output_dir,prompt_no=None):
replace_country_flag = False
if language != COUNTRY_LANG[country] and language == 'English':
replace_country_flag = True
if q_col == None:
if language == COUNTRY_LANG[country]:
q_col = 'Translation'
elif language == 'English':
q_col = 'Question'
if prompt_no is not None:
prompt_sheet = import_google_sheet(PROMPT_SHEET_ID,PROMPT_COUNTRY_SHEET[country])
output_filename = os.path.join(output_dir,f"{model_name}-{country}_{language}_{prompt_no}_result.csv")
else:
output_filename = os.path.join(output_dir,f"{model_name}-{country}_{language}_result.csv")
print(q_df[[id_col,q_col]])
guid_list = set()
if os.path.exists(output_filename):
already = pd.read_csv(output_filename)
guid_list = set(already[id_col])
print(already)
else:
write_csv_row([id_col,q_col,'prompt','response','prompt_no'],output_filename)
pb = tqdm(q_df.iterrows(),desc=model_name,total=len(q_df))
for _,d in pb:
q = d[q_col]
guid = d[id_col]
pb.set_postfix({'ID':guid})
if guid in guid_list:
continue
if replace_country_flag:
q = replace_country_name(q,country.replace('_',' '))
if prompt_no is not None:
prompt = make_prompt(q,prompt_no,language,country,prompt_sheet)
else:
prompt = q
print(prompt)
response = get_model_response(model_path,prompt,model,tokenizer,temperature=args.temperature,top_p=args.top_p,gpt_azure=args.gpt_azure)
print(response)
write_csv_row([guid,q,prompt,response,prompt_no],output_filename)
del guid_list
def get_response_from_all():
models = args.model
languages = args.language
countries = args.country
question_dir = args.question_dir
question_file = args.question_file
question_col = args.question_col
prompt_no = args.prompt_no
id_col = args.id_col
output_dir = args.output_dir
azure = args.gpt_azure
if not os.path.exists(output_dir):
os.mkdir(output_dir)
if args.gpus:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpus
if ',' in languages:
languages = languages.split(',')
if ',' in countries:
countries = countries.split(',')
if ', ' in models:
models = models.split(',')
if type(languages) == type(countries) and isinstance(languages,list):
if len(languages) != len(countries):
print("ERROR: Same number of languages and countries necessary. If multiple languages and countries are given, each element of the two lists should be in pairs.")
exit()
def get_questions(language,country):
questions_df = pd.read_csv(os.path.join(question_dir,f'{country}_full_final_questions.csv'),encoding='utf-8')
return questions_df
def generate_response_per_model(model_name):
model_path = MODEL_PATHS[model_name]
tokenizer,model = get_tokenizer_model(model_name,model_path,args.model_cache_dir)
if isinstance(languages,str):
questions = get_questions(languages,countries)
generate_response(model_name,model_path,tokenizer,model,languages,countries,questions,question_col,id_col,output_dir,prompt_no=prompt_no)
else:
for l,c in zip(languages,countries):
questions = get_questions(l,c)
generate_response(model_name,model_path,tokenizer,model,l,c,questions,question_col,id_col,output_dir,prompt_no=prompt_no)
if isinstance(models,str):
generate_response_per_model(models)
else:
for m in models:
generate_response_per_model(m)
if __name__ == "__main__":
get_response_from_all()