---
license: other
license_name: seallms
license_link: https://huggingface.co/SeaLLMs/SeaLLM-13B-Chat/blob/main/LICENSE
language:
- en
- zh
- vi
- id
- th
- ms
- km
- lo
- my
- tl
tags:
- multilingual
- sea
---

<p align="center">
  <img src="seal_logo.png" width="200" />
</p>

# *SeaLLM-7B-v2* - Large Language Models for Southeast Asia

<p align="center">
<a href="https://seallm.github.io" target="_blank" rel="noopener">Technical Blog</a>
&nbsp;&nbsp;
<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5" target="_blank" rel="noopener"> 🤗 Tech Memo</a>
&nbsp;&nbsp;
<a href="https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B" target="_blank" rel="noopener"> 🤗 DEMO</a>
&nbsp;&nbsp;
<a href="https://github.com/DAMO-NLP-SG/SeaLLMs" target="_blank" rel="noopener">Github</a>
&nbsp;&nbsp;
<a href="https://arxiv.org/pdf/2312.00738.pdf" target="_blank" rel="noopener">Technical Report</a>
</p>

We introduce [SeaLLM-7B-v2.5](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5), the state-of-the-art multilingual LLM for Southeast Asian (SEA) languages 🇬🇧 🇨🇳 🇻🇳 🇮🇩 🇹🇭 🇲🇾 🇰🇭 🇱🇦 🇲🇲 🇵🇭. It is the most significant upgrade since [SeaLLM-13B](https://huggingface.co/SeaLLMs/SeaLLM-13B-Chat), with half the size, outperforming performance across diverse multilingual tasks, from world knowledge, math reasoning, instruction following, etc.

### Highlights
* [SeaLLM-7B-v2.5](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5) outperforms GPT-3.5 and achieves 7B SOTA on most multilingual knowledge benchmarks for SEA languages (MMLU, M3Exam & VMLU).
* It achieves 79.0 and 34.9 on GSM8K and MATH, surpassing GPT-3.5 in MATH.

### Release and DEMO

- DEMO: [SeaLLMs/SeaLLM-7B](https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B).
- Technical report: [Arxiv: SeaLLMs - Large Language Models for Southeast Asia](https://arxiv.org/pdf/2312.00738.pdf).
- Model weights:
  - [SeaLLM-7B-v2](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2).
  - [SeaLLM-7B-v2-GGUF](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5-GGUF).
- Run locally:
  - [LM-studio](https://lmstudio.ai/): 
    - [SeaLLM-7B-v2.5-q4_0-chatml](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5-GGUF/blob/main/seallm-7b-v2.5-chatml.Q4_K_M.gguf) with ChatML template (`<eos>` token changed to `<|im_end|>`)
    - [SeaLLM-7B-v2.5-q4_0](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5-GGUF/blob/main/seallm-7b-v2.5.Q4_K_M.gguf) - must use SeaLLM-7B-v2.5 chat format.
  - [MLX for Apple Silicon](https://github.com/ml-explore/mlx): [SeaLLMs/SeaLLM-7B-v2.5-mlx-quantized](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5-mlx-quantized)

<blockquote style="color:red">
<p><strong style="color: red">Terms of Use and License</strong>: 
By using our released weights, codes, and demos, you agree to and comply with the terms and conditions specified in our <a href="https://huggingface.co/SeaLLMs/SeaLLM-Chat-13b/edit/main/LICENSE" target="_blank" rel="noopener">SeaLLMs Terms Of Use</a>.
</blockquote>

> **Disclaimer**:
> We must note that even though the weights, codes, and demos are released in an open manner, similar to other pre-trained language models, and despite our best efforts in red teaming and safety fine-tuning and enforcement, our models come with potential risks, including but not limited to inaccurate, misleading or potentially harmful generation.
> Developers and stakeholders should perform their own red teaming and provide related security measures before deployment, and they must abide by and comply with local governance and regulations.
> In no event shall the authors be held liable for any claim, damages, or other liability arising from the use of the released weights, codes, or demos.

> The logo was generated by DALL-E 3.


### What's new since SeaLLM-7B-v2?

* SeaLLM-7B-v2.5 was built on top of Gemma-7b, and underwent large scale SFT and carefully designed alignment.


## Evaluation


### Multilingual World Knowledge


We evaluate models on 3 benchmarks following the recommended default setups: 5-shot MMLU for En, 3-shot [M3Exam](https://arxiv.org/pdf/2306.05179.pdf) (M3e) for En, Zh, Vi, Id, Th, and zero-shot [VMLU](https://vmlu.ai/) for Vi.

| Model | Langs | En<br>MMLU | En<br>M3e | Zh<br>M3e | Vi<br>M3e | Vi<br>VMLU | Id<br>M3e | Th<br>M3e
|-----| -----  | --- |  -- | ----- | ---- | --- | --- | --- |
| GPT-3.5         | Multi | 68.90 | 75.46 | 60.20 | 58.64 | 46.32 | 49.27 | 37.41
| Vistral-7B-chat | Mono  | 56.86 | 67.00 | 44.56 | 54.33 | 50.03 | 36.49 | 25.27
| Qwen1.5-7B-chat | Multi | 61.00 | 52.07 | 81.96 | 43.38 | 45.02 | 24.29 | 20.25
| SailorLM        | Multi | 52.72 | 59.76 | 67.74 | 50.14 | ---   | 39.53 | 37.73
| SeaLLM-7B-v2    | Multi | 61.89 | 70.91 | 55.43 | 51.15 | 45.74 | 42.25 | 35.52
| SeaLLM-7B-v2.5  | Multi | 64.05 | 76.87 | 62.54 | 63.11 | 53.30 | 48.64 | 46.86


### Zero-shot CoT Multilingual Math Reasoning

<!--
[SeaLLM-7B-v2](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2) achieves with **78.5** score on the GSM8K with zero-shot CoT reasoning, making it the **state of the art** in the realm of 7B models. It also outperforms GPT-3.5 in the same GSM8K benchmark as translated into SEA languages (🇨🇳 🇻🇳 🇮🇩 🇹🇭). [SeaLLM-7B-v2](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2) also surpasses GPT-3.5 on the Thai-translated MATH benchmark, with **28.4** vs 18.1 scores.

![fig_sea_math_side_by_side.png](fig_sea_math_side_by_side.png)
-->

| Model | GSM8K<br>en | MATH<br>en | GSM8K<br>zh | MATH<br>zh | GSM8K<br>vi | MATH<br>vi | GSM8K<br>id | MATH<br>id | GSM8K<br>th | MATH<br>th
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| GPT-3.5 | 80.8 | 34.1 | 48.2 | 21.5 | 55 | 26.5 | 64.3 | 26.4 | 35.8 | 18.1
| Qwen-14B-chat | 61.4 | 18.4 | 41.6 | 11.8 | 33.6 | 3.6 | 44.7 | 8.6 | 22 | 6.0
| Vistral-7b-chat | 48.2 | 12.5 |  |  | 48.7 | 3.1 |  |  |  | 
| Qwen1.5-7B-chat | 56.8 | 15.3 | 40.0 | 2.7 | 37.7 | 9 | 36.9 | 7.7 | 21.9 | 4.7
| SeaLLM-7B-v2 | 78.2 | 27.5 | 53.7 | 17.6 | 69.9 | 23.8 | 71.5 | 24.4 | 59.6 | 22.4
| SeaLLM-7B-v2.5 | 78.5 | 34.9 | 51.3 | 22.1 | 72.3 | 30.2 | 71.5 | 30.1 | 62.0 | 28.4



Baselines were evaluated using their respective chat-template and system prompts ([Qwen1.5-7B-chat](https://huggingface.co/Qwen/Qwen1.5-7B-Chat/blob/main/tokenizer_config.json), [Vistral](https://huggingface.co/Viet-Mistral/Vistral-7B-Chat)).

#### Zero-shot MGSM

[SeaLLM-7B-v2.5](https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5) also outperforms GPT-3.5 and Qwen-14B on the multilingual MGSM for Thai.

| Model | MGSM-Zh | MGSM-Th
|-----| -----  | --- 
| ChatGPT (reported) | 61.2 | 47.2
| Qwen-14B-chat | 59.6 | 28
| SeaLLM-7B-v2 | **64.8** | 62.4
| SeaLLM-7B-v2.5 | 58.0 | **64.8**



### Sea-Bench

![fig_sea_bench_side_by_side.png](fig_sea_bench_side_by_side.png)


### Usage

#### Instruction format

```python
prompt = """<|im_start|>system
You are a helpful assistant.<eos>
<|im_start|>user
Hello world<eos>
<|im_start|>assistant
Hi there, how can I help?<eos>"""

# <|im_start|> is not a special token.
# Transformers chat_template should be consistent with vLLM format below.

# ! ENSURE 1 and only 1 bos `<s>` at the beginning of sequence
print(tokenizer.convert_ids_to_tokens(tokenizer.encode(prompt)))

"""
```

#### Using transformers's chat_template

Install the latest transformers (>4.40)

```python

from transformers import AutoModelForCausalLM, AutoTokenizer

device = "cuda" # the device to load the model onto

# use bfloat16 to ensure the best performance.
model = AutoModelForCausalLM.from_pretrained("SeaLLMs/SeaLLM-7B-v2.5", torch_dtype=torch.bfloat16, device_map=device)
tokenizer = AutoTokenizer.from_pretrained("SeaLLMs/SeaLLM-7B-v2.5")

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Hello world"},
    {"role": "assistant", "content": "Hi there, how can I help you today?"},
    {"role": "user", "content": "Explain general relativity in details."}
]

encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
print(tokenizer.convert_ids_to_tokens(encodeds[0]))

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

generated_ids = model.generate(model_inputs, max_new_tokens=1000, do_sample=True, pad_token_id=tokenizer.pad_token_id)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])

```

#### Using vLLM

```python
from vllm import LLM, SamplingParams
TURN_TEMPLATE = "<|im_start|>{role}\n{content}<eos>\n"
TURN_PREFIX = "<|im_start|>{role}\n"

def seallm_chat_convo_format(conversations, add_assistant_prefix: bool, system_prompt=None):
    # conversations: list of dict with key `role` and `content` (openai format)
    if conversations[0]['role'] != 'system' and system_prompt is not None:
        conversations = [{"role": "system", "content": system_prompt}] + conversations
    text = ''
    for turn_id, turn in enumerate(conversations):
        prompt = TURN_TEMPLATE.format(role=turn['role'], content=turn['content'])
        text += prompt
    if add_assistant_prefix:
        prompt = TURN_PREFIX.format(role='assistant')
        text += prompt    
    return text

sparams = SamplingParams(temperature=0.1, max_tokens=1024, stop=['<eos>', '<|im_start|>'])
llm = LLM("SeaLLMs/SeaLLM-7B-v2.5", dtype="bfloat16")

message = "Explain general relativity in details."
prompt = seallm_chat_convo_format(message, True)
gen = llm.generate(prompt, sampling_params)

print(gen[0].outputs[0].text)
```

#### Fine-tuning SeaLLM-7B-v2.5

Should follow the chat format and accurately mask out source tokens. Here is an example.

```python
conversations = [
    {"role": "system", "content": "You are helful assistant."},
    {"role": "user", "content": "Hello world."},
    {"role": "assistant", "content": "Hi there, how can I help?"},
    {"role": "user", "content": "Tell me a joke."},
    {"role": "assistant", "content": "Why don't scientists trust atoms? Because they make up everything."},
]
def seallm_7b_v25_tokenize_multi_turns(tokenizer, conversations, add_assistant_prefix=False):
    """
    Inputs:
        conversations: list of dict following openai format, eg
            conversations = [
                {"role": "system", "content": "You are helful assistant."},
                {"role": "user", "content": "Hello world."},
                {"role": "assistant", "content": "Hi there, how can I help?"},
                {"role": "user", "content": "Tell me a joke."},
                {"role": "assistant", "content": "Why don't scientists trust atoms? Because they make up everything."},
            ]
        add_assistant_prefix: whether to add assistant_prefix, only for inference decoding
    Outputs:
        tokenize_output_sample, {
            "input_ids": ...
            "token_type_ids": 1 if train and 0 if masked out (not train)
        }
    During training, need to create a labels, with masked-out tokens = -100 to avoid loss computations.
        labels = sample['input_ids'].clone()
        labels[sample['token_type_ids'] == 0] = -100
    """
    TURN_TEMPLATE = "<|im_start|>{role}\n{content}<eos>\n"
    TURN_PREFIX = "<|im_start|>{role}\n"
    sample = None
    assistant_prefix_len = None
    for turn_id, turn in enumerate(conversations):
        prompt = TURN_TEMPLATE.format(role=turn['role'], content=turn['content'])
        turn_sample = tokenizer(
            prompt, padding=False, truncation=False, verbose=False, add_special_tokens=False,
            return_token_type_ids=True, 
        )
        if turn['role'] == 'assistant':
            if assistant_prefix_len is None:
                assistant_prefix_len = len(tokenizer.encode(TURN_PREFIX.format(role=turn['role']), add_special_tokens=False))
            turn_sample['token_type_ids'][assistant_prefix_len:] = [1] * (len(turn_sample['input_ids']) - assistant_prefix_len)
        if sample is None:
            sample = turn_sample
        else:
            for k in turn_sample.keys():
                sample[k].extend(turn_sample[k])
    if add_assistant_prefix:
        assistant_prefix_sample = tokenizer(
            TURN_PREFIX.format(role="assistant"), padding=False, truncation=False, verbose=False, add_special_tokens=False,
            return_token_type_ids=True, 
        )
        for k in sample.keys():
            sample[k].extend(assistant_prefix_sample[k])
    if tokenizer.add_bos_token:
        sample['input_ids'] = [tokenizer.bos_token_id] + sample['input_ids']
        sample['attention_mask'] = [1] + sample['attention_mask']
        sample['token_type_ids'] = [sample['token_type_ids'][0]] + sample['token_type_ids']
    return sample

# ! testing
sample = seallm_7b_v25_tokenize_multi_turns(tokenizer, conversations)
print(tokenizer.convert_ids_to_tokens(sample['input_ids']))
print(sample['token_type_ids'])


```


## Acknowledgement to Our Linguists

We would like to express our special thanks to our professional and native linguists, Tantong Champaiboon, Nguyen Ngoc Yen Nhi and Tara Devina Putri, who helped build, evaluate, and fact-check our sampled pretraining and SFT dataset as well as evaluating our models across different aspects, especially safety.

## Citation

If you find our project useful, we hope you would kindly star our repo and cite our work as follows: Corresponding Author: [l.bing@alibaba-inc.com](mailto:l.bing@alibaba-inc.com)

**Author list and order will change!**

* `*` and `^` are equal contributions.

```
@article{damonlpsg2023seallm,
  author = {Xuan-Phi Nguyen*, Wenxuan Zhang*, Xin Li*, Mahani Aljunied*, Weiwen Xu, Hou Pong Chan,
            Zhiqiang Hu, Chenhui Shen^, Yew Ken Chia^, Xingxuan Li, Jianyu Wang,
            Qingyu Tan, Liying Cheng, Guanzheng Chen, Yue Deng, Sen Yang,
            Chaoqun Liu, Hang Zhang, Lidong Bing},
  title = {SeaLLMs - Large Language Models for Southeast Asia},
  year = 2023,
  Eprint = {arXiv:2312.00738},
}
```