T-LITE-IT-1.0

Introduction

T-lite-it-1.0 is a model derived from the Qwen 2.5 model family, designed for further fine-tuning. It is not intended for direct use as a conversational assistant without additional training and ethical oversight. Users deploying this model in industrial or commercial applications bear responsibility for its outcomes.

Architecture

T-lite-it-1.0 integrates continual pre-training and alignment techniques to enhance performance. It builds upon the robust foundation of the Qwen 2.5 model family.

Training

Pre-training Stage 1: Utilized 100 billion tokens from a variety of Russian data sources, including Common Crawl, books, code, and proprietary datasets. English data was incorporated due to its prevalence in the base model.
Pre-training Stage 2: Employed 40 billion tokens, combining instruction and pre-training data.
Supervised Fine-Tuning (SFT): Consumed 1 billion tokens from diverse instruction data.
Preference Tuning: Involved 1 billion tokens aimed at enhancing the model's helpfulness.

Guide: Running Locally

Basic Steps

Setup: Ensure you have Python installed along with the transformers library.

Load Model:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "t-tech/T-lite-it-1.0"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name, 
    torch_dtype="auto",
    device_map="auto"
)

Generate Text:

prompt = "Напиши стих про машинное обучение"
messages = [
    {"role": "system", "content": "Ты T-lite, виртуальный ассистент в Т-Технологии. Твоя задача - быть полезным диалоговым ассистентом."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(**model_inputs, max_new_tokens=256)
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)

Suggest Cloud GPUs

Consider using cloud services such as AWS, GCP, or Azure for access to GPUs, which can significantly speed up model execution and training processes.

License

Users must adhere to the licensing terms provided by T-TECH and Hugging Face. The use of T-lite-it-1.0 for further fine-tuning and deployment is subject to these terms, emphasizing ethical and responsible application of the model.

More Related APIs

T lite it 1.0