README.md

---
license: cc-by-nc-4.0
---

# tsubaki-10b-instruction-sft

# Overview
This repository provides a Japanese-centric multilingual GPT-NeoX model of 10 billion parameters.

* **Library**
    
    The model was trained using code based on [EleutherAI/gpt-neox](https://github.com/EleutherAI/gpt-neox).

* **Model architecture**

    A 36-layer, 4864-hidden-size transformer-based language model.

* **Pre-training**

    The model was trained on around **600B** tokens from a mixture of the following corpora

    - [Japanese C4](https://huggingface.co/datasets/mc4)
    - [The Pile](https://huggingface.co/datasets/EleutherAI/pile)

* **Instruction-supervised-finetuning**

    The model was finetuned on a subset records from a mixture of the following dataset

    - [Alpaca (English)](https://github.com/gururise/AlpacaDataCleaned/blob/main/alpaca_data_cleaned.json)
    - [Alpaca (Japanese translation)](https://github.com/shi3z/alpaca_ja/blob/main/alpaca_cleaned_ja.json)
    - [Flan 2021](https://huggingface.co/datasets/conceptofmind/flan2021_submix_original)
    - [Flan CoT](https://huggingface.co/datasets/conceptofmind/cot_submix_original)
    - [Flan Dialog](https://huggingface.co/datasets/conceptofmind/dialog_submix_original)

* **Model Series**

    | Variant | Link |
    | :-- | :--|
    | tsubaki-10b-instruction-sft | https://huggingface.co/Kojima777/tsubaki-10b-instruction-sft |
    | tsubaki-10b | https://huggingface.co/Kojima777/tsubaki-10b |

* **Authors**
    
    Takeshi Kojima

---

# Benchmarking

* **Japanese benchmark**

    - *The 4-task average accuracy is based on results of JCommonsenseQA, JNLI, MARC-ja, and JSQuAD.*
   
    | Model | Average | JCommonsenseQA | JNLI | MARC-ja | JSQuAD |
    | :-- | :-- | :-- | :-- | :-- | :-- |
    | tsubaki-10b-instruction-sft | 79.04 | 74.35 | 65.65 | 96.06 | 80.09 |
    | tsubaki-10b | 67.27 | 65.86 | 54.19 | 84.49 | 64.54 |

---

# How to use the model

~~~~python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Kojima777/tsubaki-10b-instruction-sft", use_fast=False)
model = AutoModelForCausalLM.from_pretrained("Kojima777/tsubaki-10b-instruction-sft")

if torch.cuda.is_available():
    model = model.to("cuda")

text = "大規模言語モデルについて説明してください。"
text = f'以下は、タスクを説明する指示です。要求を適切に満たす応答を書きなさい。\n\n### 指示:\n{text}\n\n### 応答:'
token_ids = tokenizer.encode(text, add_special_tokens=False, return_tensors="pt")

with torch.no_grad():
    output_ids = model.generate(
        token_ids.to(model.device),
        max_new_tokens=100,
        do_sample=True,
        temperature=0.6,
        top_p=0.9,
        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])
print(output)

~~~~
---

# Licenese
[cc-by-nc-4.0](https://creativecommons.org/licenses/by-nc/4.0/)