LLaMA-MoE-v1-3.5B (4/16)

[💻 Code] | [📜 Technical Report]

👋 Very nice to meet you here~

❤️ This repo contains the model LLaMA-MoE-v1-3.5B (4/16), which activates 4 out of 16 experts (3.5B parameters). This model is NOT fine-tuned by instruction pairs, so it may not be good enough to act like a chatbot.

📢 LLaMA-MoE is a series of Mixture-of-Expert (MoE) models based on LLaMA-2. You can find the code for training this model at this repo.

💎 This series of models are obtained by partitioning original LLaMA FFNs into experts and further continual pre-training. The total model size is only 6.7B parameters, which is very convenient for deployment and research usage. More details could be found at our technical report.

🚀 QuickStart

# python>=3.10

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_dir = "llama-moe/LLaMA-MoE-v1-3_5B-4_16"
tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_dir, torch_dtype=torch.bfloat16, trust_remote_code=True)
model.eval()
model.to("cuda:0")

input_text = "Suzhou is famous of"
inputs = tokenizer(input_text, return_tensors="pt")
inputs = inputs.to("cuda:0")

pred = model.generate(**inputs, max_length=50, temperature=0.0)
print(tokenizer.decode(pred.cpu()[0], skip_special_tokens=True))
# Suzhou is famous of its beautiful gardens. The most famous one is the Humble Administrator's Garden. It is a classical Chinese garden with a history of more than 600 years. The garden is divided into three

📊 Performance

📖 Details

Training Data: 200B tokens from SlimPajama with the same data sampling weights as Sheared LLaMA.

📃 Citation

@article{llama-moe,
  title={LLaMA-MoE: Building Mixture-of-Experts from LLaMA with Continual Pre-training},
  author={Tong Zhu and Xiaoye Qu and Daize Dong and Jiacheng Ruan and Jingqi Tong and Conghui He and Yu Cheng},
  journal={arXiv preprint arXiv:2406.16554},
  year={2024},
  url={https://arxiv.org/abs/2406.16554},
}