Strela

🚀 Strela - A Powerful Language Model

Strela is a powerful language model designed to ensure high - speed operation and quality responses on low - end devices. It is recommended for the following purposes:

• Chatbot for dialogues
• Story writer
• Song writer
• Translation between Russian and English
• When it's inefficient to use heavier models

📚 Documentation

Self - description from Strela

I am a computer program developed for processing and analyzing natural language. I have the ability to understand, analyze, and process natural language, enabling me to communicate with people through various communication channels. My main goal is to help people solve problems and provide information based on their requests. I can be used for various purposes, from automatic text generation, translation from one language to another, or even creating my own poems and songs.

Using the model online

You can try it here.

Using the model for chat in an application

It is recommended to use GTP4ALL. It supports GGUF, so you need to download a special version of the model in GGUF format.

Using the model for chat in Unity

It is recommended to use LLM for Unity. It supports GGUF, so you need to download a special version of the model in GGUF format.

Using the quantized model for chat in Python | Recommended

You should install gpt4all

pip install gpt4all

Then, download the GGUF version of the model and move the file to your script's directory.

# Import libraries
import os
from gpt4all import GPT4All

# Initialize the model from the strela - q4_k_m.gguf file in the current directory
model = GPT4All(model_name='strela-q4_k_m.gguf', model_path=os.getcwd())


# Function that stops generation if Strela generates the '#' symbol, which is the start of a role declaration
def stop_on_token_callback(token_id, token_string):
    if '#' in token_string:
        return False
    else:
        return True


# System prompt
system_template = """### System:
You are an AI assistant who gives a helpfull response to whatever human ask of you.
"""

# Human and AI prompt
prompt_template = """
### Human:
{0}
### Assistant:
"""

# Chat session
with model.chat_session(system_template, prompt_template):
    print("To exit, enter 'Exit'")
    while True:
        print('')
        user_input = input(">>> ")
        if user_input.lower() != "exit":

            # Stream generation
            for token in model.generate(user_input, streaming=True, callback=stop_on_token_callback):
                print(token, end='')
        else:
            break

To exit, enter 'Exit'

>>> Hello
Hello! How can I help you today?
>>> 

Using the full - fledged model for chat in Python

# Import libraries
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model
tokenizer = AutoTokenizer.from_pretrained("gai-labs/strela")
model = AutoModelForCausalLM.from_pretrained("gai-labs/strela")

# System prompt
system_prompt = "You are an AI assistant who gives a helpfull response to whatever human ask of you."

# Your prompt
prompt = "Hello!"

# Chat template
chat = f"""### System:
{system_prompt}
### Human:
{prompt}
### Assistant:
"""

# Generation
model_inputs = tokenizer([chat], return_tensors="pt")
generated_ids = model.generate(**model_inputs, max_new_tokens=64) # Set the maximum number of tokens for generation
output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

# Clean the output from the chat template
output = output.replace(chat, "")

# Print the generation results
print(output)

Hello! How can I help?

Using the model for text generation in Python

# Import libraries
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model
tokenizer = AutoTokenizer.from_pretrained("gai-labs/strela")
model = AutoModelForCausalLM.from_pretrained("gai-labs/strela")

# Prompt
prompt = "AI - "

# Generation
model_inputs = tokenizer([prompt], return_tensors="pt")
generated_ids = model.generate(**model_inputs, max_new_tokens=64) # Set the maximum number of tokens for generation
output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

# Print the generation results
print(output)

AI - is a field of computer science and technology that deals with creating machines capable of "understanding" humans or performing tasks with similar logic as humans.

📄 License

This project is licensed under the CC - BY - SA 4.0 license.