Sambalingo Turkish Chat

🚀 SambaLingo-Turkish-Chat

SambaLingo-Turkish-Chat is a human - aligned chat model trained in Turkish and English, offering high - quality multilingual interaction.

🚀 Quick Start

Loading Model With Hugging Face

Ensure to set use_fast=False when loading the tokenizer.

from transformers import AutoModelForCausalLM, AutoTokenizer

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

Interacting With Model Pipeline

Again, set use_fast=False when loading the tokenizer.

from transformers import pipeline
pipe = pipeline("text-generation", model="sambanovasystems/SambaLingo-Turkish-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, use the following chat template:

<|user|>\n{question}</s>\n<|assistant|>\n

✨ Features

SambaLingo-Turkish-Chat is a human - aligned chat model trained in Turkish and English. It is built on the base model SambaLingo-Turkish-Base using direct preference optimization. The base model adapts [Llama - 2 - 7b](https://huggingface.co/meta - llama/Llama - 2 - 7b - hf) to Turkish by training on 42 billion tokens from the Turkish split of the Cultura - X dataset. Try it at [SambaLingo - chat - space](https://huggingface.co/spaces/sambanovasystems/SambaLingo - chat - space).

📚 Documentation

Model Description

• Developed by: SambaNova Systems
• Model type: Language Model
• Language(s): Turkish, 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
• Blog Post: [sambalingo - open - source - language - experts](https://sambanova.ai/blog/sambalingo - open - source - language - experts)

Training Details

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

The SFT phase was conducted on the ultrachat_200k dataset mixed with the Google - translated version of the ultrachat_200k dataset. It was trained for one epoch with a global batch size of 512 and a max sequence length of 2048 tokens. A linear decay learning rate of 2e - 5 and 10% warmup were used.

The DPO phase was carried out on the ultrafeedback 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 a global batch size of 32 for three epochs. A linear decay learning rate of 5e - 7, 10% warmup, and β = 0.1 as the regularization factor for DPO were used.

Tokenizer Details

The vocabulary of the base llama model was extended 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, refer to our paper: SambaLingo: Teaching Large Language Models New Languages

Uses

Direct Use

Use of this model is governed by the Meta’s Llama 2 Community License Agreement. Review and accept the license before downloading the model weights.

Out - of - Scope Use

SambaLingo should NOT be used for:

• Mission - critical applications
• Applications involving the safety of others
• Making highly important decisions

Bias, Risks, and Limitations

Like all LLMs, SambaLingo has certain limitations:

• Hallucination: The model may sometimes generate responses with plausible - sounding but factually incorrect or irrelevant information.
• Code Switching: The model might unintentionally switch between languages or dialects within a single response, affecting output coherence and understandability.
• Repetition: The model may produce repetitive phrases or sentences, resulting in 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 with inappropriate or harmful content.

📄 License

This model uses the llama2 license.

Acknowledgments

We are deeply grateful to the open - source AI community; this project would not have been possible without open source. SambaNova supports the open - source community and aims to actively contribute to this initiative.

Special thanks to the following groups:

• Meta for open - sourcing LLama 2 and the FLORES - 200 dataset
• Nguyen et al for open - sourcing the 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 - sourcing 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}
}