I'm pursuing ai course and using the following script in debug mode to get a better understanding of the internal workings of the Transformers' model.generate() function. It's part of an API I'm building for a client, so ignore the Flask code - the key issue here is getting the debugger to work so I can follow the tokenized text through the model's generation process. Does the Transformers library break on debugging? Why might this be the case?:

import os

import shutilimport subprocessimport numpy as npimport torchfrom flask import Flask, requestfrom flask_restful import Api, Resource, reqparseimport transformersfrom 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 vocabularyprint('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)

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