---
language:
- en
- ja
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
model_type: mistral
---

# Swallow-MS-7b-v0.1

Our Swallow-MS-7b-v0.1 model has undergone continual pre-training from the Mistral-7B-v0.1, primarily with the addition of Japanese language data. 

# Model Release Updates

We are excited to share the release schedule for our latest models:
- **April 26, 2024**: Released the [Swallow-MS-7b-instruct-v0.1](https://huggingface.co/tokyotech-llm/Swallow-MS-7b-instruct-v0.1)
- **March 11, 2024**: Released the [Swallow-MS-7b-v0.1](https://huggingface.co/tokyotech-llm/Swallow-MS-7b-v0.1)
![logo](./logo.png)

This repository provides large language models developed by [TokyoTech-LLM](https://tokyotech-llm.github.io/).

## Model Details

* **Model type**: Please refer to Mistral technical report for details on the model architecture. 
* **Language(s)**: Japanese English
* **Tokenizer**: This model employs a tokenizer that features a broadened vocabulary based on Japanese data. This allows for a more efficient representation of text using fewer tokens, leading to a notably faster inference process.
* **Contact**: swallow[at]nlp.c.titech.ac.jp 

## Instruct Model Performance

### MT-Bench JA

#### Turn-Wise Performance

We report overall (i.e., average over scores of the first and second turns), first, and second turn scores.

##### Overall

|Model|Average|Writing|Roleplay|Reasoning|Math|Coding|Extraction|STEM|Humanities|
|---|---|---|---|---|---|---|---|---|---|
| Swallow-MS-7b-instruct-v0.1 |0.3411|0.3770|0.4290|0.3454|0.1040|0.2400|0.3677|0.3907|0.4750|

##### First Turn

|Model|Average|Writing|Roleplay|Reasoning|Math|Coding|Extraction|STEM|Humanities|
|---|---|---|---|---|---|---|---|---|---|
| Swallow-MS-7b-instruct-v0.1 |0.3699|0.4880|0.4260|0.3900|0.1080|0.2364|0.3780|0.4500|0.4800|

##### Second Turn

|Model|Average|Writing|Roleplay|Reasoning|Math|Coding|Extraction|STEM|Humanities|
|---|---|---|---|---|---|---|---|---|---|
| Swallow-MS-7b-instruct-v0.1 |0.3130|0.2624|0.4320|0.2996|0.1000|0.2430|0.3564|0.3291|0.4700|

#### Comparison to the past model

We only provide the overall score in this section.

|Model|Average|Writing|Roleplay|Reasoning|Math|Coding|Extraction|STEM|Humanities|
|---|---|---|---|---|---|---|---|---|---|
| Swallow-MS-7b-instruct-v0.1 |0.3411|0.3770|0.4290|0.3454|0.1040|0.2400|0.3677|0.3907|0.4750|
| ELYZA-japanese-Llama-2-7b-fast-instruct |0.2827|0.3289|0.3907|0.2424|0.1480|0.1584|0.3511|0.3053|0.3365|
| calm2-7b-chat |0.3204|0.4657|0.4898|0.1837|0.1005|0.1414|0.3927|0.3601|0.4293|
| calm2-7b-chat-dpo-experimental |0.3493|0.5312|0.5237|0.1857|0.1000|0.1813|0.3355|0.4320|0.5051|
| RakutenAI-7B-instruct |0.2994|0.3623|0.3711|0.3333|0.1763|0.1581|0.4215|0.2824|0.2901|
| RakutenAI-7B-chat |0.3667|0.4229|0.4644|0.3990|0.2161|0.2390|0.3416|0.3904|0.4601|


## Evaluation Benchmarks

### MT-Bench JA

We used [Japanese MT-Bench](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_question) to assess the instruction-following capabilities of models.
We utilized the following settings:

- Implemantation: FastChat [Zheng+, 2023] (commit #e86e70d0)
- Question: [Nejumi LLM-Leaderboard NEO, mtbench_ja_question_v3](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_question/v3)
- Reference Answer: [Nejumi LLM-Leaderboard NEO, mtbench_ja_referenceanswer_v1](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_referenceanswer/v1)
- Prompt for Judge: [Nejumi LLM-Lederboard NEO, mtbench_ja_prompt_v1](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_prompt/v1)
- Judge: `gpt-4-1106-preview`
- Scoring: Absolute scale normalized to a 0-1 range, averaged over five runs.


## Usage

First install additional dependencies in [requirements.txt](./requirements.txt):

```sh
pip install -r requirements.txt
```

### Instruction format Ver0.1
This format must be adhered to strictly, as deviations may result in less optimal outputs from the model.

The template used to construct a prompt for the Instruct model is specified as follows:

```
<s>[INST] <<SYS>>\n{SYSTEM_PROMPT}\n<</SYS>>\n\n{USER_MESSAGE_1} [/INST] {BOT_MESSAGE_1} </s>[INST] {USER_MESSAGE_2}[/INST] 
```


Please be aware that ``<s>`` and ``</s>`` are special tokens used for the beginning of string (BOS) and end of string (EOS), respectively, while [INST] and [/INST] are considered regular strings.

For the "{SYSTEM_PROMPT}" part, We recommend using "あなたは誠実で優秀な日本人のアシスタントです。"

For the "{USER_MESSAGE_1}" part, We recommend using {instruction}\n{input}

In other words, We recommend the following:

```
<s>[INST] <<SYS>>\nあなたは誠実で優秀な日本人のアシスタントです。\n<</SYS>>\n\n{instruction1}\n{input1} [/INST] {BOT_MESSAGE_1}</s>[INST] \n\n{instruction2}\n{input2} [/INST] 
```

### Use the instruct model Ver0.1

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

model_name = "tokyotech-llm/Swallow-MS-7b-instruct-v0.1"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)

device = "cuda"

messages = [
    {"role": "system", "content": "あなたは誠実で優秀な日本人のアシスタントです。"},
    {"role": "user", "content": "東京工業大学の主なキャンパスについて教えてください"}
]

encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt")

model_inputs = encodeds.to(device)
model.to(device)

generated_ids = model.generate(model_inputs, max_new_tokens=128, do_sample=True)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])
```

## Training Datasets

### Instruction Tuning Ver0.1

The following datasets were used for the instruction tuning. 

- [OpenAssistant Conversations Dataset](https://huggingface.co/datasets/llm-jp/oasst1-21k-ja) was used, where human utterances are included but the responses are not used. Instead, the responses were generated using the [Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/datasets/llm-jp/oasst1-21k-jahttps://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) model.
- [OpenAssistant Conversations Dataset 21k Ja](https://huggingface.co/datasets/llm-jp/oasst1-21k-ja)
- [OpenAssistant Conversations Dataset 21k En](https://huggingface.co/datasets/llm-jp/oasst1-21k-en)
- [Databricks Dolly 15k Ja](https://huggingface.co/datasets/llm-jp/databricks-dolly-15k-ja)
- [Databricks Dolly 15k En](https://huggingface.co/datasets/databricks/databricks-dolly-15k)

Please note that some of the data had issues with quality or format, so not all of it was used.

## Risks and Limitations

The models released here are still in the early stages of our research and development and have not been tuned to ensure outputs align with human intent and safety considerations.

## Acknowledgements

We thank Mistral AI for releasing Mistral 7B v0.1 under an open license for others to build on.

Our project is supported by the [ABCI Large-scale Language Model Building Support Program](https://abci.ai/en/link/llm_support_program.html) of the National Institute of Advanced Industrial Science and Technology. 

## License

apache-2.0

## Authors

Here are the team members:
- From [Okazaki Laboratory](https://www.nlp.c.titech.ac.jp/index.en.html), the following members:
  - [Naoaki Okazaki](https://www.chokkan.org/index.ja.html)
  - [Sakae Mizuki](https://s-mizuki-nlp.github.io/)
  - [Hiroki Iida](https://meshidenn.github.io/)
  - [Mengsay Loem](https://loem-ms.github.io/)
  - [Shota Hirai](https://huggingface.co/Kotemo428)
  - [Kakeru Hattori](https://aya-se.vercel.app/)
  - [Masanari Ohi](https://twitter.com/stjohn2007)
- From [YOKOTA Laboratory](https://www.rio.gsic.titech.ac.jp/en/index.html), the following members:
  - [Rio Yokota](https://twitter.com/rioyokota)
  - [Kazuki Fujii](https://twitter.com/okoge_kaz)
  - [Taishi Nakamura](https://twitter.com/Setuna7777_2)
  - [Takumi Okamoto](https://www.linkedin.com/in/takumi-okamoto)
  - [Ishida Shigeki](https://www.wantedly.com/id/reborn27)