Nikosar / ru-gpts

Russian GPT3 models.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ruGPT3Large, ruGPT3Medium and ruGPT2Large

Russian GPT trained with 2048 context length (ruGPT3Large), Russian GPT Medium trained with context 2048 (ruGPT3Medium) and Russian GPT2 large (ruGPT2Large) trained with 1024 context length.

We suggest you use ruGPT2Large because this model is more stable and tested.

Examples here

Note: If you cannot download the checkpoint, try adding it to your google drive following this issue

Table of contents

Setup

Setup ruGPT3Large

Code reused from microsoft implementation of Megatron-LM. Supports only python3.6.

To use this repo please install the latest supported versions of PyTorch with GPU support.

Additionally, part of this codebase leverages tensorflow-cpu to (optionally) perform dataloading of TFRecords for GPT training. We recommend creating a virtual environment (to avoid breaking existing tf installations) and install our requirements.txt.

python -m pip install virtualenv
virtualenv gpt_env
source gpt_env/bin/activate
pip install -r requirements.txt

For using of sparse operations in attention additionally install torch-blocksparse:

source gpt_env/bin/activate
pip install torch-blocksparse

Torch-Blocksparse depends on CUDA 10.1 and the Triton language and compiler, which requires llvm-9.

Setup ruGPT3Medium

For this model you can use code from microsoft implementation of Megatron-LM in our repo or use transformers interface. Therefore, you should follow the instructions for ruGPT2Large or ruGPT3Large for installation.

Setup ruGPT2Large

This model is smaller and was trained with transformers==v2.8.0. For installing use command:

pip install transformers

Details of pretraining

All GPUs are Tesla V100-SXM3 32 Gb.

Details of pretraining ruGPT3Large

Model was trained on 1024 context length with transformers by SberDevices team on 80B tokens around 3 epochs. After that we finetune this on 2048 context. For load transformers checkpoint use --load-openai.

The training process took around two weeks on 8 DGX2 (128 GPUs) for 1024 context and few days on 16 GPUs for 2048 context on Christophari.

Perplexity is 16 on test set.

You can obtain this model here GDrive Yandex.Disk GDrive option-2 or use in transformers with model name sberbank-ai/rugpt3large_based_on_gpt2 (see usage for details).

🤗HuggingFace model card link

Details of pretraining ruGPT3Medium

Model was trained on 1024 context length with transformers by SberDevices team on 80B tokens around 3 epoch. After that model was finetuned on 2048 context.

Total training time took around 16 days on 64 GPUs.

You can obtain this model here GDrive Yandex.Disk GDrive option-2 or use in transformers with model name sberbank-ai/rugpt3medium_based_on_gpt2 (see usage for details).

🤗HuggingFace model card link

Details of pretraining ruGPT2Large

Model was trained on 1024 context length with transformers by SberDevices team on 170Gb data on 64 GPUs 3 weeks.

You can obtain this model here GDrive Yandex.Disk GDrive option-2 or use in transformers with model name sberbank-ai/rugpt2large (see usage for details).

🤗HuggingFace model card link

Usage

Usage ruGPT3Large

We've provided 2 scripts that pretrain and generate with ruGPT3Large. Save and load model checkpoints with --save and --load.

Finetuning

Data preparation

We support three file formats for training, but all require preprocessing. First, place your training data in a loose json format, with one json containing a text sample per line. For example:

{"src": "KISH", "text": "Как же джокер ты хитер", "type": "Ru", "id": "0", "title": "First Part"}
{"src": "The Internet", "text": "Ты удачи приговор", "type": "Ru", "id": "42", "title": "Second Part"}

The name of the text field of the json can be changed by using the --text-key flag. The other metadata are optional and are not used in training.

Running script

bash ./scripts/pretrain_ruGPT3Large.sh

This script runs single gpu ruGPT3Large pretraining. This script contains command for running on Christophari:

MP_SIZE=1
NUM_GPUS_PER_WORKER=1

mpirun --np ${NUM_GPUS_PER_WORKER} python pretrain_megatron.py \
       --train-data /home/jovyan/data/train.jsonl \
       --valid-data /home/jovyan/data/valid.jsonl \
       --test-data /home/jovyan/data/valid.jsonl \
       --save /home/jovyan/ruGPT3Large/checkpoints_${now}_${host} \
       --load /home/jovyan/ruGPT3Large \
       --tensorboard-dir /home/jovyan/ruGPT3Large/runs_${now}_${host} \
       --save-interval 500 \
       --eval-interval 500 \
       --log-interval 100 \
       --model-parallel-size ${MP_SIZE} \
       --num-layers 24 \
       --hidden-size 1536 \
       --num-attention-heads 16 \
       --seq-length 2048 \
       --max-position-embeddings 2048 \
       --vocab-size 50257 \
       --batch-size 1 \
       --train-iters 200000 \
       --distributed-backend nccl \
       --lr 0.00015 \
       --lr-decay-style cosine \
       --weight-decay 1e-2 \
       --clip-grad 1.0 \
       --warmup .01 \
       --fp16 \
       --lazy-loader \
       --checkpoint-activations \
       --loose-json \
       --text-key \
       --tokenizer-path /home/jovyan/ruGPT3Large \
       --tokenizer-type GPT2BPETokenizer \
       --finetune \

Or you can use use transformers interface:

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("sberbank-ai/rugpt3large_based_on_gpt2")

model = AutoModel.from_pretrained("sberbank-ai/rugpt3large_based_on_gpt2")

Text Generation

bash ./scripts/generate_ruGPT3Large.sh

Starts an interactive terminal session that generates text either conditionally or unconditionally depending on what the user enters into the prompt.

The script is capable of top-k, or top-p sampling as specified by the appropriate variables within the script.

Example of generation:

Context: на словах ты лев толстой
ruGPT3Large: а в сущности, - ты тоже не дурак, просто так же, как и твой человек, то есть твоя "жизнь", а также как и ты думаешь по-настоящему "ты" и есть твои "жизнь" или "выбор" в отношении твоего положения.

Context: как же джокер ты хитер
ruGPT3Large: или автор книги по бизнесу!

Example of generation in colab Open In Colab

Usage ruGPT3Medium

You can run megatron script with option --load-openai or use transformers interface:

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("sberbank-ai/rugpt3medium_based_on_gpt2")

model = AutoModel.from_pretrained("sberbank-ai/rugpt3medium_based_on_gpt2")

Text Generation

bash ./scripts/generate_ruGPT3Medium.sh

Starts an interactive terminal session that generates text either conditionally or unconditionally depending on what the user enters into the prompt.

The script is capable of top-k, or top-p sampling as specified by the appropriate variables within the script.

Example of generation:

Context >>> На словах ты Лев Толстой, а на деле
ruGPT: На словах ты Лев Толстой, а на деле я — Лев Давидович Троцкий, — сказал я. — Так что мы еще посмотрим

Context: как же джокер ты хитер
ruGPT: как же джокер ты хитер, в этой игре
 - Я не злодей, просто хотел узнать, можно ли узнать о чём?

Usage ruGPT2Large

We've provided 2 scripts that pretrain and generate with ruGPT2Large from transformers original code.

Finetuning

Data preparation

We can pass to model raw text files.

Running script

bash ./scripts/pretrain_ruGPT2Large.sh

This script runs single gpu ruGPT3Large pretraining. This script contains command for running on Christophari:

python pretrain_transformers.py \
    --output_dir=/home/jovyan/rugpt2large/checkpoints_"${now}"_"${host}" \
    --model_type=gpt2 \
    --model_name_or_path=/home/jovyan/gpt2_large_bbpe_v50 \
    --do_train \
    --train_data_file=/home/jovyan/data/train.txt \
    --do_eval \
    --eval_data_file=/home/jovyan/data/valid.txt \
    --fp16

Or use transformers interface:

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("sberbank-ai/rugpt2large")

model = AutoModel.from_pretrained("sberbank-ai/rugpt2large")

Text Generation

bash ./scripts/generate_ruGPT2Large.sh

Starts an interactive terminal session that generates text either conditionally or unconditionally depending on what the user enters into the prompt.

The script is capable of top-k, or top-p sampling as specified by the appropriate variables within the script.

Example of generation:

Context: На словах ты Лев Толстой, а на деле
ruGPT: На словах ты Лев Толстой, а на деле – козел!» – так я про себя подумал, но решил не отвечать. Я встал, поклонился

About

Russian GPT3 models.

License:MIT License


Languages

Language:Python 99.6%Language:Shell 0.4%