metadata
license: mit
language:
- ru
- en
pipeline_tag: text-generation
inference: false
tags:
- gpt3
- transformers
- pytorch
This is a generative model converted to fp16 format based on ai-forever/ruGPT-3.5-13B
Examples of usage
from transformers import GPT2LMHeadModel, AutoTokenizer
model = GPT2LMHeadModel.from_pretrained('Gaivoronsky/ruGPT-3.5-13B-fp16')
tokenizer = AutoTokenizer.from_pretrained('Gaivoronsky/ruGPT-3.5-13B-fp16')
model = model.half()
model = model.to('cuda')
request = "Человек: Сколько весит жираф? Помощник: "
encoded_input = tokenizer(request, return_tensors='pt', \
add_special_tokens=False).to('cuda')
output = model.generate(
**encoded_input,
num_beams=2,
do_sample=True,
max_new_tokens=100
)
print(tokenizer.decode(output[0], skip_special_tokens=True))