--- thumbnail: https://github.com/rinnakk/japanese-pretrained-models/blob/master/rinna.png license: mit datasets: - Anthropic/hh-rlhf language: - ja - en inference: false --- # bilingual-gpt-neox-4b-instruction-sft ![rinna-icon](./rinna.png) --- # Update - **2023/08/02** We uploaded the newly trained `rinna/bilingual-gpt-neox-4b-instruction-sft` with the MIT license. - Please refrain from using the previous model released on 2023/07/31 for commercial purposes if you have already downloaded it. - The new model released on 2023/08/02 is built from datasets with less strict licenses and has better evaluation performance, so we suggest using the new model. - For reference, we provide the MD5 checksum values for the `pytorch_model.bin` files of the previous and current models. - 2023/07/31 model: `edf190a323c0ae63f71476700fb0b462` - 2023/08/02 model: `de72aa5b66beee7b65783c96f687d186` - **2023/07/31** In the previously released `rinna/bilingual-gpt-neox-4b-instruction-sft`, we found that part of the training data (i.e. Openchat ShareGPT4 and WizardLM) have a non-commercial license, and thus it does not comply with **the MIT license**. We decided to remove the previous version and build a new SFT model from datasets with less strict licenses. The new model will be uploaded in a few days. We sincerely apologize for our careless mistake. --- # Overview This repository provides an English-Japanese bilingual GPT-NeoX model of 3.8 billion parameters. The model is based on [`rinna/bilingual-gpt-neox-4b`](https://huggingface.co/rinna/bilingual-gpt-neox-4b) and has been finetuned to serve as an instruction-following conversational agent. * **Model architecture** A 36-layer, 2816-hidden-size transformer-based language model. * **Fine-tuning** The fine-tuning data is the subset of the following datasets. * [Anthropic HH RLHF data](https://huggingface.co/datasets/Anthropic/hh-rlhf) and its Japanese translation * [FLAN Instruction Tuning data](https://github.com/google-research/FLAN) and its Japanese translation * **Model Series** | Variant | Link | | :-- | :--| | Bilingual 4B MiniGPT4 | https://huggingface.co/rinna/bilingual-gpt-neox-4b-minigpt4 | | Bilingual 4B PPO | https://huggingface.co/rinna/bilingual-gpt-neox-4b-instruction-ppo | | Bilingual 4B SFT | https://huggingface.co/rinna/bilingual-gpt-neox-4b-instruction-sft | | Bilingual 4B 8K | https://huggingface.co/rinna/bilingual-gpt-neox-4b-8k | | Bilingual 4B | https://huggingface.co/rinna/bilingual-gpt-neox-4b | | Japanese 3.6B PPO | https://huggingface.co/rinna/japanese-gpt-neox-3.6b-instruction-ppo | | Japanese 3.6B SFT-v2 | https://huggingface.co/rinna/japanese-gpt-neox-3.6b-instruction-sft-v2 | | Japanese 3.6B SFT | https://huggingface.co/rinna/japanese-gpt-neox-3.6b-instruction-sft | | Japanese 3.6B | https://huggingface.co/rinna/japanese-gpt-neox-3.6b | * **Contributors** [Tianyu Zhao](https://huggingface.co/tianyuz) and [Kei Sawada](https://huggingface.co/keisawada) --- # Benchmarking Our evaluation experiments suggest that the bilingual-gpt-neox-4b-instruction-sft model performs slightly better than the previous [Japanese GPT-NeoX 3.6B PPO](https://huggingface.co/rinna/japanese-gpt-neox-3.6b-instruction-ppo) in Japanese tasks. - *The 4-task average accuracy is based on results of JCommonsenseQA, JNLI, MARC-ja, and JSQuAD.* - *The 6-task average accuracy is based on results of JCommonsenseQA, JNLI, MARC-ja, JSQuAD, XWinograd, and JAQKET-v2.* | Model | 4-task average accuracy | 6-task average accuracy | | :-- | :-- | :-- | | bilingual-gpt-neox-4b-instruction-ppo | 61.01 | 61.16 | | **bilingual-gpt-neox-4b-instruction-sft** | **61.02** | **61.69** | | bilingual-gpt-neox-4b | 56.12 | 51.83 | | japanese-gpt-neox-3.6b-instruction-ppo | 59.86 | 60.07 | | japanese-gpt-neox-3.6b | 55.07 | 50.32 | --- # I/O Format A special format has been adopted to construct inputs. * An input prompt is formatted as a conversation between `ユーザー` and `システム`. * Each input utterance consists of (1) its speaker (`"ユーザー"` or `"システム"`), (2) a colon (`":"`), (3) a whitespace (`" "`), and (4) utterance text (e.g. `"世界で一番高い山は？"`). * The input prompt should be ended with `"システム: "` to acknowledge the model to generate a response. * All the utterances in the input prompt should be separated by a newline `\n`. Following is an example to construct input from a conversation. ~~~python prompt = [ { "speaker": "ユーザー", "text": "Hello, you are an assistant that helps me learn Japanese." }, { "speaker": "システム", "text": "Sure, what can I do for you?" }, { "speaker": "ユーザー", "text": "VRはなんですか。" } ] prompt = [ f"{uttr['speaker']}: {uttr['text']}" for uttr in prompt ] prompt = "\n".join(prompt) prompt = ( prompt + "\n" + "システム: " ) print(prompt) """ ユーザー: Hello, you are an assistant that helps me learn Japanese. システム: Sure, what can I do for you? ユーザー: VRはなんですか。 システム: """ ~~~ --- # How to use the model **Notice:** Since the model is **sensitive to decoding hyper-parameters** (e.g. `temperature`, `top_p`, `top_k`, `repetition_penalty`), it is suggested to explore the best setting for your task. ~~~~python import torch from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("rinna/bilingual-gpt-neox-4b-instruction-sft", use_fast=False) model = AutoModelForCausalLM.from_pretrained("rinna/bilingual-gpt-neox-4b-instruction-sft") if torch.cuda.is_available(): model = model.to("cuda") token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt") with torch.no_grad(): output_ids = model.generate( token_ids.to(model.device), max_new_tokens=512, do_sample=True, temperature=1.0, top_p=0.85, pad_token_id=tokenizer.pad_token_id, bos_token_id=tokenizer.bos_token_id, eos_token_id=tokenizer.eos_token_id ) output = tokenizer.decode(output_ids.tolist()[0][token_ids.size(1):]) print(output) """VRとはVirtual Realityの略で、仮想現実とも呼ばれます。これは、コンピューターを使用して仮想世界を作り出し、仮想世界上でコンピューターのゲームや仮想世界を体験するための技術です。この技術は、コンピューターやモバイ ルデバイスの進歩によって、2015年以降、ますます普及しています。VRは、ゲームや仮想世界、その他のアプリケー ションなどのさまざまな分野で、コンピューターと人間の相互作用の新しい方法を提供しています。""" ~~~~ --- # Tokenization The model uses a [sentencepiece](https://github.com/google/sentencepiece)-based tokenizer. * The tokenizer has a vocabulary size of 65,536. * It uses *byte fallback* to decompose unknown text pieces into UTF-8 byte pieces to avoid producing `` tokens. * It can recognize *consecutive whitespaces*, *newlines*, and *tabs* to handle structured texts better. * We turned off the default behaviour of prepending leading whitespace because it is not beneficial for processing Japanese. * Specifically, single whitespace is always processed as one token so that any English word won't have a preceding whitespace like in many other tokenizers (e.g. `_Hello`). * This decision trades the English processing efficiency for a unified way to treat whitespaces. * It leads to a significantly lower loss of next token prediction on English data because whitespaces are easy to predict. * **Don't forget to set `use_fast=False` to make the above features function correctly.** --- # How to cite ~~~ @misc{rinna-bilingual-gpt-neox-4b-instruction-sft, title = {rinna/bilingual-gpt-neox-4b-instruction-sft}, author = {Zhao, Tianyu and Sawada, Kei}, url = {https://huggingface.co/rinna/bilingual-gpt-neox-4b-instruction-sft}, } @inproceedings{sawada2024release, title = {Release of Pre-Trained Models for the {J}apanese Language}, author = {Sawada, Kei and Zhao, Tianyu and Shing, Makoto and Mitsui, Kentaro and Kaga, Akio and Hono, Yukiya and Wakatsuki, Toshiaki and Mitsuda, Koh}, booktitle = {Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)}, month = {5}, year = {2024}, url = {https://arxiv.org/abs/2404.01657}, } ~~~ --- # Licenese [The MIT license](https://opensource.org/licenses/MIT)