Bitnet-Llama3-from8BM-now2B

🚀 BitNet-Llama3 (from 8B to 2B) Transformation & Training

This project transforms a Llama3 model from 8B parameters to a BitNet architecture with 2B parameters, applying BitLinear layers. Additionally, the model is trained with a predefined dataset and uploaded to Hugging Face for future use.

---

Model Description

<!-- Provide a longer summary of what this model is. -->

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Developed by: [email protected]
• Funded by [optional]: ITCL
• Shared by [optional]: [More Information Needed]
• Model type: LLama3 8B Tramsformed to Bitnet
• Language(s) (NLP): Bitnet
• License: [More Information Needed]
• Finetuned from model [optional]: [More Information Needed]

Model Sources [optional]

<!-- Provide the basic links for the model. -->

• Repository: ejbejaranos/Bitnet-Llama3-from8BM-now2B

📄 Description

This repository includes scripts to:

1. 🎯 Transform a Llama3 model to a BitNet architecture.
2. 💻 Train the model using Hugging Face and Weights & Biases.
3. 🚀 Upload the transformed and trained model to Hugging Face for inference and future use.

---

⚙️ Requirements

• Python 3.8+
• Pytorch 1.10+
• Transformers 4.0+
• Hugging Face Hub API
• Weights & Biases

---

🧰 Installation

Make sure you have all required dependencies installed:

pip install torch transformers datasets wandb huggingface_hub

💥 How to Use

1. Using the trained model for inference

from transformers import AutoModelForCausalLM, AutoTokenizer
from utils.bitnet_transformation import replace_linears_in_hf

# Load the BitNet model
model = "ejbejaranos/Bitnet-Llama3-from8BM-now2B"
model = AutoModelForCausalLM.from_pretrained(
    model,
    use_auth_token="YOUR_HF_TOKEN"
)

# Replace BitNet layers for inference
replace_linears_in_hf(model)
tokenizer = AutoTokenizer.from_pretrained("ejbejaranos/Bitnet-Llama3-from8BM-now2B")

# Set up for inference
model.to(device="cuda:0")
prompt = "What is Machine Learning?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
generate_ids = model.generate(inputs.input_ids, max_length=50)
output = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]

print(output)

---

🧑‍🔬 Metrics

During training, the following metrics will be logged to Weights & Biases:

• final_loss: 1.4.
• final_perplexity: 4.2.

---

🎯 Future Goals

• Implement additional quantization layers for inference.
• Test the model on different datasets and contexts.

---

📢 Contact

If you have questions, suggestions, or improvements, feel free to open an Issue or contact us through Hugging Face.

---

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

💡 Acknowledgments

Thanks to Hugging Face and Weights & Biases for providing support and tools.