budecosystem/boomer-bitnet-634m

Democratizing access to LLMs for the open-source community.
Let's advance AI, together.

---

Introduction 🎉

We are open-sourcing one of our early experiments of BitNet b1.58 paper. This 634m parameter model is pre-trained from scratch using a custom synthetic dataset of 5B tokens. The model's architecture experiments contain the modification of using higher depth and shallow configuration

Run the model

Please note that, at the moment, trust_remote_code=True is required for running the model.

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("budecosystem/boomer-bitnet-634m",
                                             trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("budecosystem/boomer-bitnet-634m")

input_ids = tokenizer("In the recent Super Bowl LVIII,", return_tensors='pt').to(model.device)["input_ids"]

outputs = model.generate(input_ids, max_new_tokens=216)

print(tokenizer.batch_decode(outputs))

Evaluations

We have evaluated the pre-trained model on few of the benchmarks

Model Name	ARC	MMLU	Winogrande	Hellaswag	MathQA	GSM8K
boomer-bitnet-634m	26.19	25.23	51.07	34.08	23.38	0.91

Final thought on Boomer!

This isn't the end. It's just the beginning of a journey towards creating more advanced, more efficient, and more accessible language models. We invite you to join us on this exciting journey.

Aknowledgements

We'd like to thank the open-source community and the researchers whose foundational work laid the path for BOOMER. Special shoutout to team who published BitNet b1.58 paper.