lonas-llama-7b-commonsense-adapter

LoNAS Adapter Card: lonas-llama-7b-commonsense-adapter

The super-adapter-network fine-tuned on LLaMA-7B with some commonsense reasoning datasets using LoNAS.

Paper Abstract

Large Language Models (LLMs) continue to grow, reaching hundreds of billions of parameters and making it challenging for Deep Learning practitioners with resource-constrained systems to use them, e.g., fine-tuning these models for a downstream task of their interest. Adapters, such as low-rank adapters (LoRA), have been proposed to reduce the number of trainable parameters in a model, reducing memory requirements and enabling smaller systems to fine-tune these models. Orthogonal to this work, Neural Architecture Search (NAS) has been used to discover compressed and more efficient architectures without sacrificing performance compared to similar base models. This paper introduces a novel approach, LoNAS, to use NAS on language models by exploring a search space of elastic low-rank adapters while reducing memory and compute requirements of full-scale NAS, resulting in high-performing compressed models obtained from weight-sharing super-networks. Compared to models fine-tuned with LoRA, these models contain fewer total parameters, reducing the inference time with only minor decreases in accuracy and, in some cases, even improving accuracy. We discuss the limitations of LoNAS and share observations for the research community regarding its generalization capabilities, which have motivated our follow-up work.

Model Details

Note

Please note, we only provide the model adapter and do not provide a copy of the base yahma/llama-7b-hf model. Any use of this adapter requires a separate download of the base model.

Information

• Adapter name: lonas-llama-7b-commonsense-adapter
• Base model: LLaMA-7b
• Domain: Commonsense
• Subnetwork version: Super-network
• NNCF Configuration: nncf_lonas_llama_7b.json

Adapter Configuration

• LoRA rank: 32
• LoRA alpha: 64
• LoRA target modules: q_proj, k_proj, v_proj, up_proj, gate_proj, down_proj

Training Hyperparameters

• Batch size: 16
• Learning rate: 3e-4
• Epoch: 6

Training Data

Unified commonsense reasoning dataset: commonsense_15k.json.

Evaluation Data

BoolQ, PIQA, SIQA, HellaSwag, WinoGrande, ARC-e, ARC-c, OBQA.

How to use

Refer to https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/LoNAS#evaluation:

CUDA_VISIBLE_DEVICES=${DEVICES} python run_commonsense.py \
    --dataset_path None \
    --model_name_or_path yahma/llama-7b-hf \
    --lora \
    --lora_weights lonas-llama-7b-commonsense \
    --nncf_config nncf_config/unified_commonsense/nncf_lonas_llama_7b.json \
    --do_test \
    --output_dir lonas-llama-7b-commonsense/results

Evaluation Results

Results of the heuristic sub-network discoverd from the super-network:

Model Sources

Repository: https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/LoNAS

Paper:

• LoNAS: Elastic Low-Rank Adapters for Efficient Large Language Models
• Low-Rank Adapters Meet Neural Architecture Search for LLM Compression

Ethical Considerations

Intel is committed to respecting human rights and avoiding causing or contributing to adverse impacts on human rights. See Intel’s Global Human Rights Principles. Intel’s products and software are intended only to be used in applications that do not cause or contribute to adverse impacts on human rights.

Citation

@inproceedings{munoz-etal-2024-lonas,
    title = "{L}o{NAS}: Elastic Low-Rank Adapters for Efficient Large Language Models",
    author = "Munoz, Juan Pablo  and
      Yuan, Jinjie  and
      Zheng, Yi  and
      Jain, Nilesh",
    editor = "Calzolari, Nicoletta  and
      Kan, Min-Yen  and
      Hoste, Veronique  and
      Lenci, Alessandro  and
      Sakti, Sakriani  and
      Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.940",
    pages = "10760--10776",
}

License

Apache-2.0