interneuronai / az-longchat

Model Details

Original Model: lmsys/longchat-7b-v1.5-32k
Fine-Tuned For: Azerbaijani language understanding and generation
Dataset Used: Azerbaijani translation of the Stanford Alpaca dataset
Fine-Tuning Method: Self-instruct method

This model, is part of the "project/Barbarossa" initiative, aimed at enhancing natural language processing capabilities for the Azerbaijani language. By fine-tuning this model on the Azerbaijani translation of the Stanford Alpaca dataset using the self-instruct method, we've made significant strides in improving AI's understanding and generation of Azerbaijani text.

Our primary objective with this model is to offer insights into the feasibility and outcomes of fine-tuning large language models (LLMs) for the Azerbaijani language. The fine-tuning process was undertaken with limited resources, providing valuable learnings rather than creating a model ready for production use. Therefore, we recommend treating this model as a reference or a guide to understanding the potential and challenges involved in fine-tuning LLMs for specific languages. It serves as a foundational step towards further research and development rather than a direct solution for production environments.

This project is a proud product of the Alas Development Center (ADC) . We are thrilled to offer these finely-tuned large language models to the public, free of charge.

How to use?

from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, pipeline

model_path = "alasdevcenter/az-longchat"

model = AutoModelForCausalLM.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)

pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=200)

instruction = "Təbiətin qorunması  "
formatted_prompt = f"""Aşağıda daha çox kontekst təmin edən təlimat var. Sorğunu adekvat şəkildə tamamlayan cavab yazın.
                ### Təlimat:
                {instruction}
                ### Cavab:
                """

result = pipe(formatted_prompt)
print(result[0]['generated_text'])