This revision is from 2024/10/18 01:35. You can Restore it.

• git clone https://github.com/LumenScopeAI/BrainTransformers-SNN-LLM.git
• https://huggingface.co/LumenscopeAI/BrainTransformers-3B-Chat

Custom transfomer that comes with git clone: BrainTransformers-SNN-LLM/transformers/models/braingpt/configuration_braingpt.py

Add:

from transformers import PretrainedConfig
Copy

git clone https://github.com/LumenScopeAI/BrainTransformers-SNN-LLM.git
cd BrainTransformers-SNN-LLM/
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
sudo apt-get install git-lfs
git lfs install
git clone https://huggingface.co/LumenscopeAI/BrainTransformers-3B-Chat
Copy

BrainTransformers-SNN-LLM/transformers/models/braingpt/modeling_braingpt.py

def load_silu_approximator(device, dtype):
    act_fn = SiLUApproximator().to(device).to(dtype)
    pos_checkpoint = os.path.join(os.path.dirname(__file__), 'model_pos.pth')
    neg_checkpoint = os.path.join(os.path.dirname(__file__), 'model_neg.pth')
    
    if os.path.exists(pos_checkpoint) and os.path.exists(neg_checkpoint):
        act_fn.pos_model.load_state_dict(
            torch.load(pos_checkpoint, map_location=device, weights_only=True)
        )
        act_fn.neg_model.load_state_dict(
            torch.load(neg_checkpoint, map_location=device, weights_only=True)
        )
        """
        act_fn.pos_model.load_state_dict(
            torch.load(pos_checkpoint, map_location=device)
        )
        act_fn.neg_model.load_state_dict(
            torch.load(neg_checkpoint, map_location=device)
        )
        """
    else:
        raise FileNotFoundError(
            f"SiLUApproximator parameters not found at {pos_checkpoint} and {neg_checkpoint}"
        )
    return act_fn
Copy

Open run.py and change the model path to BrainTransformers-3B-Chat

import torch
from transformers import AutoTokenizer, BrainGPTForCausalLM
from tqdm import tqdm
from torch.utils.data import DataLoader

# Define the model and tokenizer path
model_path = "BrainTransformers-3B-Chat"

# Load the model and tokenizer
print("Loading model and tokenizer...")
model = BrainGPTForCausalLM.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Move the model to GPU (if available)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

print(f"Model loaded on {device}")

# Define a function to generate text
def generate_text(messages, max_new_tokens=200, temperature=0.7, top_k=50, top_p=0.95, do_sample=True):
    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    model_inputs = tokenizer([text], return_tensors="pt").to(device)

    with torch.no_grad():
        generated_ids = model.generate(
            **model_inputs,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            top_k=top_k,
            top_p=top_p,
            do_sample=do_sample,
            pad_token_id=tokenizer.eos_token_id
        )

    generated_ids = [
        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
    ]

    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return response

# Test the generation function
def test_generation(test_prompts):
    print("\nTesting text generation:")
    for prompt in test_prompts:
        messages = [
            {"role": "system", "content": "You are a knowledgeable assistant."},
            {"role": "user", "content": prompt}
        ]
        print(f"\nPrompt: {prompt}")
        generated = generate_text(messages)
        print(f"Generated: {generated}")

# Define a function to interact with the model
def interact_with_model():
    while True:
        user_input = input("You: ")
        if user_input.lower() in ["exit", "quit"]:
            break
        messages = [
            {"role": "system", "content": "You are a knowledgeable assistant."},
            {"role": "user", "content": user_input}
        ]
        response = generate_text(messages)
        print(f"Assistant: {response}")

test_prompts = [
    "Please explain the Pythagorean theorem!",
    "What is artificial intelligence?",
    "Write a poem about spring.",
    "Explain the basic principles of quantum computing.",
    "How to make a simple pizza?"
]

print("\nTesting text generation before STDP training:")
test_generation(test_prompts)

# Interact with the model
print("\nYou can now interact with the model. Type 'exit' or 'quit' to stop.")
interact_with_model()
Copy