qhduan's picture
Update README.md
35cc9cd
|
raw
history blame
1.15 kB
metadata
license: apache-2.0
widget:
  - text: |-
      <|endoftext|>
      def load_excel(path):
          return pd.read_excel(path)
      # docstring
      """

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)