Openelm 450M Instruct

🚀 OpenELM

OpenELM is a family of open, efficient language models. It uses a layer - wise scaling strategy to enhance accuracy and provides models with different parameter sizes, along with a complete research framework.

🚀 Quick Start

OpenELM is a family of Open Efficient Language Models. It employs a layer - wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, thereby enhancing accuracy. The OpenELM models are pretrained using the CoreNet library. We release both pretrained and instruction - tuned models with 270M, 450M, 1.1B, and 3B parameters, along with a complete framework that includes data preparation, training, fine - tuning, and evaluation procedures, multiple pre - trained checkpoints, and training logs to support open research.

The pre - training dataset contains RefinedWeb, deduplicated PILE, a subset of RedPajama, and a subset of Dolma v1.6, totaling approximately 1.8 trillion tokens. Please review the license agreements and terms of these datasets before use.

✨ Features

• Efficient Parameter Allocation: Utilizes a layer - wise scaling strategy for better accuracy.
• Multiple Model Sizes: Offers models with 270M, 450M, 1.1B, and 3B parameters.
• Complete Framework: Provides a full - fledged framework for data preparation, training, fine - tuning, and evaluation.

💻 Usage Examples

Basic Usage

We have provided an example function to generate output from OpenELM models loaded via HuggingFace Hub in generate_openelm.py.

You can try the model by running the following command:

python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2

Please refer to this link to obtain your hugging face access token.

Advanced Usage

Additional arguments to the hugging face generate function can be passed via generate_kwargs. As an example, to speed up the inference, you can try lookup token speculative generation by passing the prompt_lookup_num_tokens argument as follows:

python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 prompt_lookup_num_tokens=10

Alternatively, try model - wise speculative generation with an assistive model by passing a smaller model through the assistant_model argument, for example:

python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 --assistant_model [SMALLER_MODEL]

📚 Documentation

Main Results

Zero - Shot

LLM360

OpenLLM Leaderboard

See the technical report for more results and comparison.

Evaluation

Setup

Install the following dependencies:

# install public lm - eval - harness
harness_repo="public-lm-eval-harness"
git clone https://github.com/EleutherAI/lm-evaluation-harness ${harness_repo}
cd ${harness_repo}
# use main branch on 03 - 15 - 2024, SHA is dc90fec
git checkout dc90fec
pip install -e .
cd ..

# 66d6242 is the main branch on 2024 - 04 - 01 
pip install datasets@git+https://github.com/huggingface/datasets.git@66d6242
pip install tokenizers>=0.15.2 transformers>=4.38.2 sentencepiece>=0.2.0

Evaluate OpenELM

# OpenELM - 450M - Instruct
hf_model=apple/OpenELM-450M-Instruct

# this flag is needed because lm - eval - harness set add_bos_token to False by default, but OpenELM uses LLaMA tokenizer which requires add_bos_token to be True
tokenizer=meta-llama/Llama-2-7b-hf
add_bos_token=True
batch_size=1

mkdir lm_eval_output

shot=0
task=arc_challenge,arc_easy,boolq,hellaswag,piqa,race,winogrande,sciq,truthfulqa_mc2
lm_eval --model hf \
        --model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
        --tasks ${task} \
        --device cuda:0 \
        --num_fewshot ${shot} \
        --output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
        --batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log

shot=5
task=mmlu,winogrande
lm_eval --model hf \
        --model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
        --tasks ${task} \
        --device cuda:0 \
        --num_fewshot ${shot} \
        --output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
        --batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log

shot=25
task=arc_challenge,crows_pairs_english
lm_eval --model hf \
        --model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
        --tasks ${task} \
        --device cuda:0 \
        --num_fewshot ${shot} \
        --output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
        --batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log

shot=10
task=hellaswag
lm_eval --model hf \
        --model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
        --tasks ${task} \
        --device cuda:0 \
        --num_fewshot ${shot} \
        --output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
        --batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log

🔧 Technical Details

The OpenELM models use a layer - wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, which helps in enhancing the accuracy of the models. The pre - training dataset is a combination of multiple publicly available datasets, including RefinedWeb, deduplicated PILE, a subset of RedPajama, and a subset of Dolma v1.6, with a total of approximately 1.8 trillion tokens.

📄 License

This project is licensed under the apple-amlr license.

📚 Citation

If you find our work useful, please cite:

@article{mehtaOpenELMEfficientLanguage2024,
	title = {{OpenELM}: {An} {Efficient} {Language} {Model} {Family} with {Open} {Training} and {Inference} {Framework}},
	shorttitle = {{OpenELM}},
	url = {https://arxiv.org/abs/2404.14619v1},
	language = {en},
	urldate = {2024-04-24},
	journal = {arXiv.org},
	author = {Mehta, Sachin and Sekhavat, Mohammad Hossein and Cao, Qingqing and Horton, Maxwell and Jin, Yanzi and Sun, Chenfan and Mirzadeh, Iman and Najibi, Mahyar and Belenko, Dmitry and Zatloukal, Peter and Rastegari, Mohammad},
	month = apr,
	year = {2024},
}

@inproceedings{mehta2022cvnets, 
     author = {Mehta, Sachin and Abdolhosseini, Farzad and Rastegari, Mohammad}, 
     title = {CVNets: High Performance Library for Computer Vision}, 
     year = {2022}, 
     booktitle = {Proceedings of the 30th ACM International Conference on Multimedia}, 
     series = {MM '22} 
}

⚠️ Important Note

The release of OpenELM models aims to empower and enrich the open research community by providing access to state - of - the - art language models. Trained on publicly available datasets, these models are made available without any safety guarantees. Consequently, there exists the possibility of these models producing outputs that are inaccurate, harmful, biased, or objectionable in response to user prompts. Thus, it is imperative for users and developers to undertake thorough safety testing and implement appropriate filtering mechanisms tailored to their specific requirements.