kdf/python-docstring-generation

Basic info

model based Salesforce/codegen-350M-mono

fine-tuned with data codeparrot/github-code-clean

data filter by python

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_type = 'kdf/python-docstring-generation'
tokenizer = AutoTokenizer.from_pretrained(model_type)
model = AutoModelForCausalLM.from_pretrained(model_type)

inputs = tokenizer('''<|endoftext|>
def load_excel(path):
    return pd.read_excel(path)

# docstring
"""''', return_tensors='pt')

doc_max_length = 128

generated_ids = model.generate(
    **inputs,
    max_length=inputs.input_ids.shape[1] + doc_max_length,
    do_sample=False,
    return_dict_in_generate=True,
    num_return_sequences=1,
    output_scores=True,
    pad_token_id=50256,
    eos_token_id=50256  # <|endoftext|>
)

ret = tokenizer.decode(generated_ids.sequences[0], skip_special_tokens=False)
print(ret)

Prompt

You could give model a style or a specific language, for example:

inputs = tokenizer('''<|endoftext|>
def add(a, b):
    return a + b

# docstring
"""
    Calculate numbers add.

    Args:
        a: the first number to add
        b: the second number to add

    Return:
        The result of a + b
"""
<|endoftext|>
def load_excel(path):
    return pd.read_excel(path)

# docstring
"""''', return_tensors='pt')

doc_max_length = 128

generated_ids = model.generate(
    **inputs,
    max_length=inputs.input_ids.shape[1] + doc_max_length,
    do_sample=False,
    return_dict_in_generate=True,
    num_return_sequences=1,
    output_scores=True,
    pad_token_id=50256,
    eos_token_id=50256  # <|endoftext|>
)

ret = tokenizer.decode(generated_ids.sequences[0], skip_special_tokens=False)
print(ret)

inputs = tokenizer('''<|endoftext|>
def add(a, b):
    return a + b

# docstring
"""
    计算数字相加

    Args:
        a: 第一个加数
        b: 第二个加数

    Return:
        相加的结果
"""
<|endoftext|>
def load_excel(path):
    return pd.read_excel(path)

# docstring
"""''', return_tensors='pt')

doc_max_length = 128

generated_ids = model.generate(
    **inputs,
    max_length=inputs.input_ids.shape[1] + doc_max_length,
    do_sample=False,
    return_dict_in_generate=True,
    num_return_sequences=1,
    output_scores=True,
    pad_token_id=50256,
    eos_token_id=50256  # <|endoftext|>
)

ret = tokenizer.decode(generated_ids.sequences[0], skip_special_tokens=False)
print(ret)