35 lines
928 B
Markdown
35 lines
928 B
Markdown
|
---
|
||
|
|
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](https://huggingface.co/ai-forever/ruGPT-3.5-13B)
|
||
|
|
|
||
|
|
## Examples of usage
|
||
|
|
|
||
|
|
```python
|
||
|
|
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))
|
||
|
|
```
|