README.md

---
license: llama2
datasets:
- HuggingFaceH4/ultrachat_200k
- HuggingFaceH4/ultrafeedback_binarized
- HuggingFaceH4/cai-conversation-harmless
language:
- sl
- en
---



# SambaLingo-Slovenian-Chat

<img src="SambaLingo_Logo.png" width="340" style="margin-left:'auto' margin-right:'auto' display:'block'"/>

<!-- Provide a quick summary of what the model is/does. -->
SambaLingo-Slovenian-Chat is a human aligned chat model trained in Slovenian and English. It is trained using direct preference optimization on top the base model [SambaLingo-Slovenian-Base](https://huggingface.co/sambanovasystems/SambaLingo-Slovenian-Base). The base model adapts [Llama-2-7b](https://huggingface.co/meta-llama/Llama-2-7b-hf) to Slovenian by training on 17 billion tokens from the Slovenian split of the [Cultura-X](https://huggingface.co/datasets/uonlp/CulturaX) dataset. Try this model at [SambaLingo-chat-space](https://huggingface.co/spaces/sambanovasystems/SambaLingo-chat-space).

## Model Description
<!-- Provide a longer summary of what this model is. -->

- **Developed by:** [SambaNova Systems](https://sambanova.ai/)
- **Model type:** Language Model
- **Language(s):** Slovenian, English
- **Finetuned from model:** [Llama-2-7b](https://huggingface.co/meta-llama/Llama-2-7b-hf)
- **Try this model:** [SambaLingo-chat-space](https://huggingface.co/spaces/sambanovasystems/SambaLingo-chat-space)
- **Paper:** [SambaLingo: Teaching Large Language Models New Languages](https://arxiv.org/abs/2404.05829)
- **Blog Post**: [sambalingo-open-source-language-experts](https://sambanova.ai/blog/sambalingo-open-source-language-experts)

## Getting Started

### Loading Model With Hugging Face
Please make sure to set use_fast=False when loading the tokenizer.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("sambanovasystems/SambaLingo-Slovenian-Chat", use_fast=False)
model = AutoModelForCausalLM.from_pretrained("sambanovasystems/SambaLingo-Slovenian-Chat", device_map="auto", torch_dtype="auto")
```

### Interacting With Model Pipeline
Please make sure to set use_fast=False when loading the tokenizer.
```python
from transformers import pipeline
pipe = pipeline("text-generation", model="sambanovasystems/SambaLingo-Slovenian-Chat", device_map="auto", use_fast=False)
messages = [
                {"role": "user", "content": {YOUR_QUESTION}},
]
prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt)[0]
outputs = outputs["generated_text"]
```

### Suggested Inference Parameters
- Temperature: 0.8
- Repetition penalty: 1.0
- Top-p: 0.9

### Prompting Guidelines
To prompt this model, please use the following chat template:
```
<|user|>\n{question}</s>\n<|assistant|>\n
```

### Example Prompts and Generations
```
<|user|>
Kdaj je bil ustvarjen google? Zakaj je prevladoval nad drugimi iskalniki?</s>
<|assistant|>
Google je bil ustanovljen septembra 1998 v garaži Larryja Pagea in Sergeya Brina v Mountain Viewu v Kaliforniji.

Google je prevladoval nad drugimi iskalniki zaradi več dejavnikov:

1. Učinkovitost: Google je bil veliko hitrejši in učinkovitejši od drugih iskalnikov, saj je zagotavljal rezultate iskanja v delčku sekunde.

2. Uporabniku prijazen vmesnik: Googlov vmesnik je bil uporabniku prijazen in enostaven za krmarjenje, kar je uporabnikom olajšalo iskanje, kar iščejo.

3. Personalizacija: Google je uporabnikom omogočil prilagajanje rezultatov iskanja z uporabo njihovih iskalnih navad, zgodovine brskanja in drugih podatkov.

4. Kakovost rezultatov iskanja: Google je zagotavljal visokokakovostne rezultate iskanja, ki so bili točni, ustrezni in brez neželene vsebine.

5. Oglaševalski model: Google je uvedel uspešen oglaševalski model, ki je podjetjem omogočil, da ciljajo na uporabnike na podlagi njihovih iskalnih poizvedb, s čimer je povečal učinkovitost njihovih oglaševalskih kampanj.

6. Strateška partnerstva: Google je sklenil strateška partnerstva z drugimi podjetji, kot so Apple, Android in YouTube, kar mu je pomagalo razširiti doseg in povečati bazo uporabnikov.

7. Inovacije: Google je bil vedno v ospredju inovacij, saj je predstavil nove funkcije in tehnologije, kot so Google Maps, Google Now in Google Assistant, ki so izboljšale uporabniško izkušnjo in povečale njegovo priljubljenost.

```

## Training Details
The alignment phase follows the recipe for [Zephyr-7B](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta), and comprises two stages: supervised fine-tuning (SFT) and Direct Performance Optimization (DPO).

The SFT phase was done on the [ultrachat_200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k) dataset mixed with the Google translated version of the ultrachat_200k dataset. It was trained for one epoch with global batch size 512 and max sequence length 2048 tokens. We used a linear decay learning rate of 2e-5 and 10% warmup.

The DPO phase was done on the [ultrafeedback](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized) dataset and [cai-conversation-harmless](https://huggingface.co/datasets/HuggingFaceH4/cai-conversation-harmless) dataset, mixed with 10% of the data Google translated. It was trained with global batch size 32 and for three epochs. We used a linear decay learning rate of 5e-7, 10% warmup and β=0.1 as the regularization factor for DPO. 


## Tokenizer Details
We extended the vocabulary of the base llama model from 32,000 tokens to 57,000 tokens by adding up to 25,000 non-overlapping tokens from the new language.

## Evaluation 
For evaluation results see our paper: [SambaLingo: Teaching Large Language Models New Languages](https://arxiv.org/abs/2404.05829)

## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

### Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
Use of this model is governed by the Meta’s [Llama 2 Community License Agreement](https://ai.meta.com/llama/license/). Please review and accept the license before downloading the model weights.


### Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
SambaLingo should NOT be used for:

- Mission-critical applications
- Applications that involve the safety of others
- Making highly important decisions

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

Like all LLMs, SambaLingo has certain limitations:
- Hallucination: Model may sometimes generate responses that contain plausible-sounding but factually incorrect or irrelevant information.
- Code Switching: The model might unintentionally switch between languages or dialects within a single response, affecting the coherence and understandability of the output.
- Repetition: The Model may produce repetitive phrases or sentences, leading to less engaging and informative responses.
- Coding and Math: The model's performance in generating accurate code or solving complex mathematical problems may be limited.
- Toxicity: The model could inadvertently generate responses containing inappropriate or harmful content.

## Acknowledgments
We extend our heartfelt gratitude to the open-source AI community; this endeavor would not have been possible without open source. SambaNova embraces the open-source community and aspires to actively contribute to this initiative.

We would like to give a special thanks to the following groups:
- Meta for open sourcing LLama 2 and open sourcing FLORES-200 dataset
- Nguyen et al for open sourcing CulturaX dataset
- CohereAI for releasing AYA-101 and open sourcing a multilingual instruction tuning dataset
- EleutherAI for their open source evaluation framework
- Hugging Face-H4 team for open source the zephyr training recipe and alignment handbook repo


## Cite SambaLingo
```
@misc{csaki2024sambalingo,
      title={SambaLingo: Teaching Large Language Models New Languages}, 
      author={Zoltan Csaki and Bo Li and Jonathan Li and Qiantong Xu and Pian Pawakapan and Leon Zhang and Yun Du and Hengyu Zhao and Changran Hu and Urmish Thakker},
      year={2024},
      eprint={2404.05829},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```