Vikhr Qwen 2.5 1.5 B Instruct

Introduction

The Vikhr-Qwen-2.5-1.5B-Instruct is a bilingual instructive model designed for high-efficiency text processing in Russian and English. It is trained on the GrandMaster-PRO-MAX dataset, making it adept at delivering precise responses and fast task execution for various applications, including professional environments and user-facing applications.

Architecture

The model is based on the Qwen-2.5-1.5B-Instruct architecture, with a focus on bilingual support in Russian and English. It utilizes methodologies such as Supervised Fine-Tuning (SFT) and Chain-Of-Thought (CoT) to enhance its performance in instruction generation, contextual responses, and text analysis.

Training

The Vikhr-Qwen-2.5-1.5B-Instruct model was developed using the SFT method on a synthetic dataset, GrandMaster-PRO-MAX, consisting of 150,000 instructions. The training process incorporated CoT methodology and GPT-4-turbo prompts to achieve high accuracy and coherence in responses.

Guide: Running Locally

To run the model locally, follow these steps:

1. Install Transformers Library: Ensure you have the transformers library installed.

   pip install transformers
   

2. Load the Model and Tokenizer:

   from transformers import AutoModelForCausalLM, AutoTokenizer
   
   model_name = "Vikhrmodels/Vikhr-Qwen-2.5-1.5B-Instruct"
   model = AutoModelForCausalLM.from_pretrained(model_name)
   tokenizer = AutoTokenizer.from_pretrained(model_name)
   

3. Prepare Input and Generate Output:

   input_text = "Напиши краткое описание книги Гарри Поттер."
   messages = [
       {"role": "system", "content": "Вы — Vikhr, ИИ помощник, созданный компанией Vikhr models для предоставления полезной, честной и безопасной информации."},
       {"role": "user", "content": input_text},
   ]
   input_ids = tokenizer.apply_chat_template(messages, truncation=True, add_generation_prompt=True, return_tensors="pt")
   output = model.generate(input_ids, max_length=1512, temperature=0.3, num_return_sequences=1, no_repeat_ngram_size=2, top_k=50, top_p=0.95)
   generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
   print(generated_text)
   

4. Consider Cloud GPUs: For optimal performance, especially for large models like this, consider using cloud GPU services such as AWS, Google Cloud, or Azure.

License

The Vikhr-Qwen-2.5-1.5B-Instruct model is distributed under the Apache-2.0 License. This allows for open-source use with minimal restrictions, promoting collaboration and modification.