alexeymosc/ai_stalker_ru_gpt_3_medium

The ruGPT-3 medium model finetuned for 3 epochs on the corpus of 300 electronic books in the universe of the Zona.

Usage

## Load NN locally

from transformers import GPT2LMHeadModel, GPT2Tokenizer
import torch
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def load_tokenizer_and_model(model_name_or_path):
  return GPT2Tokenizer.from_pretrained(model_name_or_path), \
    GPT2LMHeadModel.from_pretrained(model_name_or_path).to(DEVICE)

# Load model from local dir

tokenizer, model = load_tokenizer_and_model("./models/path/")

model.eval() #freeze gradient calc

print('Model was loaded')

# Define generator
def generate_story_actions(
    model, 
    tok, 
    text,
    max_length = 500,
    top_k = 5,
    top_p = 0.95,
    do_sample = True,
    temperature = 1.2,
    num_beams = 3,
    no_repeat_ngram_size = 3,
    repetition_penalty = 2.,
    last_text = None,
    num_sentences = 3
    ):

    input_ids = tok.encode(text, return_tensors="pt").to(DEVICE)

    out = model.generate(
      input_ids,
      max_length=max_length,
      repetition_penalty=repetition_penalty,
      do_sample=do_sample,
      top_k=top_k, top_p=top_p, temperature=temperature,
      num_beams=num_beams, no_repeat_ngram_size=no_repeat_ngram_size
      )

    generated_content = list(map(tok.decode, out))[0]

    return ' '.join(generated_content)

# Generate
story_action = generate_story_actions(model, tokenizer, text = player_action_promt, last_text = last_text)