|
--- |
|
language: |
|
- bo |
|
tags: |
|
- tibetan |
|
- pretrained causal language model |
|
- roberta |
|
widget: |
|
- text: "རིན་" |
|
- text: "རྫོགས་པའི་" |
|
- text: "ཆོས་ཀྱི་" |
|
- text: "གངས་རིའི་" |
|
- text: "བོད་ཀྱི་སྨན་" |
|
license: "mit" |
|
--- |
|
|
|
# A demo for generating text using `Tibetan Roberta Causal Language Model` |
|
|
|
``` |
|
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline |
|
|
|
model_name = 'sangjeedondrub/tibetan-roberta-causal-base' |
|
model = AutoModelForCausalLM.from_pretrained(model_name) |
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
|
text_gen_pipe = pipeline( |
|
"text-generation", |
|
model=model, |
|
tokenizer=tokenizer |
|
) |
|
|
|
init_text = 'རིན་' |
|
|
|
outputs = text_gen_pipe(init_text, |
|
do_sample=True, |
|
max_new_tokens=200, |
|
temperature=.9, |
|
top_k=10, |
|
top_p=0.92, |
|
num_return_sequences=10, |
|
truncate=True) |
|
for idx, output in enumerate(outputs, start=1): |
|
print(idx) |
|
print(output['generated_text']) |
|
``` |
|
|
|
# About |
|
|
|
This model is trained and released by Sangjee Dondrub [sangjeedondrub at live dot com], the mere purpose of conducting these experiments is to improve my familiarity with Transformers APIs. |