Question

我在调试模式下使用以下脚本来更好地了解Transformers的model.generate（）函数的内部工作原理。它是我为客户端构建的API的一部分，因此请忽略Flask代码-此处的关键问题是使调试器正常工作，以便我可以在模型的生成过程中遵循标记化文本。变形金刚库会在调试时中断吗？为什么会这样呢？：

import os
import shutil
import subprocess

import numpy as np
import torch
from flask import Flask, request
from flask_restful import Api, Resource, reqparse

import transformers
from transformers import CONFIG_NAME, WEIGHTS_NAME, GPT2LMHeadModel, GPT2Tokenizer


'''
Model Imports
'''


app = Flask(__name__)
api = Api(app)


def get_model(model_dir):
    if not os.path.exists(model_dir):
        print(f'Building model directory at {model_dir}')
        os.mkdir(model_dir)
    try:
        command = f'aws s3 sync AWS_BUCKET {model_dir}'
        subprocess.call(command.split())
    except:
        print('AWS commandline call failed. Have you configured the AWS cli yet?')

MODEL_DIR = "./model"
if not os.path.exists(MODEL_DIR):
    get_model(MODEL_DIR)

NUM_PATTERN = r'\s\d+[A-Za-z]*'

output_model_file = os.path.join(MODEL_DIR, WEIGHTS_NAME)
output_config_file = os.path.join(MODEL_DIR, CONFIG_NAME)

# Re-load the saved model and vocabulary
print('Loading model!')
model = GPT2LMHeadModel.from_pretrained(MODEL_DIR)
tokenizer = GPT2Tokenizer.from_pretrained(MODEL_DIR)

'''
Arg Parser
'''
parser = reqparse.RequestParser()
parser.add_argument('prompt', type=str, help='Main input string to be transformed. REQUIRED.', required=True)
parser.add_argument('max_length', type=int, help='Max length for generation.', default=20)
parser.add_argument('repetition_penalty', type=int, help='Penalty for word repetition. Higher = fewer repetitions.', default=5)
parser.add_argument('length_penalty', type=int, help='Exponential penalty for length. Higher = shorter sentences.', default=1)
parser.add_argument('num_beams', type=int, help='# Beams to use for beam search.', default=5)
parser.add_argument('temperature', type=float, help='Temperature of the softmax operation used in generation.', default=3)
parser.add_argument('top_k', type=int, help='Top words to select from during text generation.', default=50)
parser.add_argument('top_p', type=float, help='Top-P for Nucleus Sampling. Lower = more restrictive search.', default=0.8)
parser.add_argument('num_return_sequences', type=int, help='Number of sequences to generate.', default=1)

def decode(output):
    return str(tokenizer.decode(output, skip_special_tokens=True))

class TransformerAPI(Resource):
    def get(self):
        args = parser.parse_args()

        app.logger.info(f'Using model loaded from {MODEL_DIR}.')
        ids = tokenizer.encode(args['prompt'])
        inp = torch.tensor(np.array(ids)[np.newaxis, :])
        
        #Account for generation limits < input value
        if inp.shape[1] >= args['max_length']:
            print(inp.shape[1])
            print(args['max_length'])
            result = inp[:, :args['max_length']]
            print(result)
            decoded = [decode(result.tolist()[0])] * args['num_return_sequences']
            return {'completion': decoded,
                    'model_used': MODEL_DIR}
        else:
            result = model.generate(input_ids=inp,
                                    max_length=args['max_length'],
                                    repetition_penalty=args['repetition_penalty'],
                                    length_penalty=args['length_penalty'],
                                    do_sample=True,
                                    num_beams=args['num_beams'],
                                    temperature=args['temperature'],
                                    top_k=args['top_k'],
                                    top_p=args['top_p'],
                                    num_return_sequences=args['num_return_sequences'])

            decoded = [decode(l.tolist()) for l in result]

            return {'completion': decoded,
                    'model_used': MODEL_DIR}

api.add_resource(TransformerAPI, '/api/v1')

if __name__ == '__main__':
    #app.run(debug=True)
    ids = tokenizer.encode('The present invention')
    inp = torch.tensor(np.array(ids)[np.newaxis, :])
    result = model.generate(input_ids=inp,
                            max_length=15,
                            repetition_penalty=5,
                            length_penalty=1,
                            do_sample=True,
                            num_beams=5,
                            temperature=3,
                            num_return_sequences=1)
    print(result)

python app.py执行得很好，但是在调试模式下（在PyCharm上）运行相同会遇到错误：

Traceback (most recent call last):
  File "/Applications/PyCharm.app/Contents/plugins/python/helpers/pydev/pydevd.py", line 1438, in _exec
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "/Applications/PyCharm.app/Contents/plugins/python/helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "/Users/mgb/Desktop/Work/Apteryx_Clients_2/bao/bao-ai/apteryx_apis/patformer/app.py", line 45, in <module>
    tokenizer = GPT2Tokenizer.from_pretrained(MODEL_DIR)
  File "/Users/mgb/opt/anaconda3/envs/transformers/lib/python3.8/site-packages/transformers/tokenization_utils.py", line 282, in from_pretrained
    return cls._from_pretrained(*inputs, **kwargs)
  File "/Users/mgb/opt/anaconda3/envs/transformers/lib/python3.8/site-packages/transformers/tokenization_utils.py", line 411, in _from_pretrained
    tokenizer = cls(*init_inputs, **init_kwargs)
  File "/Users/mgb/opt/anaconda3/envs/transformers/lib/python3.8/site-packages/transformers/tokenization_gpt2.py", line 118, in __init__
    super(GPT2Tokenizer, self).__init__(bos_token=bos_token, eos_token=eos_token, unk_token=unk_token, **kwargs)
  File "/Users/mgb/opt/anaconda3/envs/transformers/lib/python3.8/site-packages/transformers/tokenization_utils.py", line 232, in __init__
    assert isinstance(value, str) or (six.PY2 and isinstance(value, unicode))
AssertionError

这似乎与GPT2Tokenizer对象有关。

变形金刚脚本运行，但在PyCharm调试器中中断

0 个答案: