Cyclone 0.5 Open-Source Large Language Model - Supports Russian and English, Optimizes JSON Handling and Multi-round Conversations!

It 5.3 Fp16 32k

Developed by Vikhrmodels

Cyclone 0.5* is an open-source large language model supporting Russian and English, extending the RoPE context window to 32k and optimizing JSON processing and multi-turn dialogue capabilities

Large Language Model

Transformers

Supports Multiple LanguagesOpen Source License:Apache-2.0 #Russian optimization #32k long context #Multi-turn dialogue enhancement

Downloads 74

Release Time : 6/4/2024

Model Overview

This model has undergone SFT phase data augmentation training, demonstrating stable performance in long-context processing and complex prompt scenarios, particularly suitable for Russian and English text generation tasks

Model Features

Extended context window

RoPE context window extended to 32k, supporting longer text processing

Multilingual support

Specially optimized bilingual processing capabilities for Russian and English

Structured data processing

Enhanced JSON format processing capability with more stable output

Multi-turn dialogue optimization

Improved dialogue coherence and context retention

Model Capabilities

Russian text generation

English text generation

Multi-turn dialogue processing

JSON format processing

Long text understanding

Use Cases

Natural language processing

Multilingual customer service bot

Building intelligent customer service systems supporting Russian and English

Capable of handling complex multi-turn dialogues and structured queries

Technical documentation generation

Automatically generating technical documentation and API descriptions

Supports structured output formats like JSON

Educational applications

Language learning assistant

Helping Russian or English learners practice the language

Provides natural and fluent bilingual interaction experience

🚀 Transformers Library Release 0.5*

This release of the Transformers library brings significant improvements. A large amount of data has been added to the supervised fine-tuning (SFT). As a result, JSON and multiturn functionality now work more stably, especially on long contexts and difficult prompts. Additionally, the parameters of the pre-trained model have been slightly adjusted, and RoPE with a 32k context has been added.

[Google Colab] - Coming later
GGUF
EXL2

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model = AutoModelForCausalLM.from_pretrained("Vikhrmodels/it-5.3-fp16-32k",
                                             device_map="auto",
                                             attn_implementation="sdpa",
                                             torch_dtype=torch.bfloat16)

tokenizer = AutoTokenizer.from_pretrained("Vikhrmodels/it-5.3-fp16-32k")
from transformers import  AutoTokenizer, pipeline
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
prompts = [
    "В чем разница между фруктом и овощем?",
    "Годы жизни колмагорова?"]

def test_inference(prompt):
    prompt = pipe.tokenizer.apply_chat_template([{"role": "user", "content": prompt}], tokenize=False, add_generation_prompt=True)
    print(prompt)
    outputs = pipe(prompt, max_new_tokens=512, do_sample=True, num_beams=1, temperature=0.25, top_k=50, top_p=0.98, eos_token_id=79097)
    return outputs[0]['generated_text'][len(prompt):].strip()


for prompt in prompts:
    print(f"    prompt:\n{prompt}")
    print(f"    response:\n{test_inference(prompt)}")
    print("-"*50)

📄 License

This project is licensed under the Apache-2.0 license.

📚 Citation

@article{nikolich2024vikhr,
  title={Vikhr: The Family of Open-Source Instruction-Tuned Large Language Models for Russian},
  author={Aleksandr Nikolich and Konstantin Korolev and Artem Shelmanov},
  journal={arXiv preprint arXiv:2405.13929},
  year={2024},
  url={https://arxiv.org/pdf/2405.13929}
}

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご