Llama Impact Grants // education, environment & open innovation
Llama Impact Grants FAQ This document is intended to inform the submission of your proposal for the first phase of the Llama Impact Grants program. It will be updated with additional questions the week following each webinar. If you have additional questions not currently addressed in this document:
● You can submit them to LlamaImpact@meta.com. All general questions (without attribution) will be added to this document.
● We will host two live Q&A zoom webinars on October 24th at 4:00pm (UTC+0) and November 7th at 12:00pm (UTC+0). Please register in advance. ○ October 24th webinar registration ○ November 7th webinar registration Please review the program’s Terms & Conditions prior to submitting questions. Additional Resources:
● Getting started with Llama: This guide provides information and resources to help you set up Llama including how to access the model, hosting, how-to and integration guides. Additionally, you will find supplemental materials to further assist you while building with Llama.
● Llama GitHub Repository
● Hugging Face
● Llama 2 Research Paper What type of applicants are you seeking for this program? We are open to any non-profit, academic institutions, and research institutions with an idea for how to apply the Llama 2 large-language models to one of our designated problem tracks. For specific questions regarding country eligibility and definitions of what qualify as eligible organization type, please refer to the program’s Terms & Conditions. I am not a non-profit organization, academic institution, or research institution. Is my organization still eligible to participate? If you do not qualify as a non-profit organization, academic institution, or a research institution, you may advise or consult eligible organizations and offer your expertise, however only eligible Organizations are permitted to apply for and eligible to receive Grant Funds. When are applications due? Applications are due by 11:59 pm ET on November 15, 2023. Can my application be submitted in my local language? All applications must be submitted in English. For additional information regarding submission requirements please refer to the Terms & Conditions. We have an idea, but our organization does not have AI expertise. Can we still apply? Yes! Phase 1 proposals will be evaluated using an equal weighted average between the four categories listed on the program website. If selected to proceed to Phase 2, proposals that demonstrate strong potential in non-technical categories will be granted an opportunity during the Refinement Workshop to connect with technical experts to better incorporate the application of the Llama 2 model to their proposed solution. There will also be an opportunity to seek additional collaborators before the Phase 2 submission deadline. I am an AI researcher with a compelling technical application, but less familiar with the social impact problem areas. Can I still apply? Yes! As long as you qualify as an eligible applicant as defined by the Terms & Conditions you are welcome to submit a proposal. Phase 1 proposals will be evaluated using an equal weighted average between the four categories listed on the program website. If selected to proceed to Phase 2, proposals that demonstrate strong potential in technical categories will be granted an opportunity during the Refinement Workshop to connect with non-profit organizations and subject matter experts to incorporate their ideas into an implementation plan. There will also be an opportunity to seek additional collaborators before the Phase 2 submission deadline. What qualifies as a strong proposal under the Open Innovation track? Any proposal that articulates how the application of Llama 2 contributes to societal benefit where the impact vision does not exclusively focus on education or environment outcomes. This might include intersectional ideas that address multiple issue areas, other sectors completely (e.g. health, equal opportunity, economic mobility, etc). If submitting a proposal in this track, please be sure to clearly articulate how the application of Llama 2 will lead to measurable progress within the designated topic area. We welcome creativity! Could you provide some examples of potential applications of large-language models within the different tracks? These examples are intended as illustrative and not comprehensive. Your proposal does not need to specifically address these topics. Education:
● Generate language exercises, provide pronunciation assistance, and offer real-time translations, helping students develop their language skills. This could enhance communication abilities, broaden opportunities, and improve access to global knowledge and resources. This work could also potentially build off recent research published by our Fundamental AI Research group.
● Generate interactive exercises, provide explanations for concepts, and offer immediate feedback, helping students practice and reinforce their learning. By addressing gaps in basic skills, language models could empower students to build a strong educational foundation. Environment:
● Aid in the discovery of novel materials for carbon capture applications. By leveraging language processing capabilities and knowledge of materials science, models could analyze existing databases, research papers, and computational simulations to identify materials with high CO2 adsorption capacity, selectivity, and stability. This could lead to the development of more efficient and robust carbon capture materials.
● Direct Air Capture (DAC) is a technology that aims to remove carbon dioxide (CO2) directly from the atmosphere and is an essential tool for offsetting carbon impacts. Large language models could aid in developing innovative carbon capture techniques specifically tailored for DAC.
● Faster synthesis and answers from large corpuses of research/policy papers. LLMs can help researchers and sustainability professionals parse and understand scientific information. .
● Extracting structured data from complex and messy text sources. LLMs can help us extract and structure information from messy sources. For example, extracting tabular data from narrative sustainability statements or processing complex disclosure data to see and understand trends.
● Improving the speed of coding new software applications that deliver environmental insights. LLMs are already transforming how we do software development and data science to drive climate action. These tools can help us deliver quantitative research and code new applications that help us understand and act in a changing environment Open Innovation:
● Faster synthesis and answers from large corpuses of research, policy papers, or data sets, including deriving novel insights from unstructured text
● Improving the speed of coding new software applications that help us understand and act quickly in a changing world
● Summarizing known side effects and anecdotal outcomes of medications to mitigate negative interactions and propose solutions to increase medication compliance
● Analysis of court record and documents to identify biases and evidentiary challenges that could reopen cases and bring about more fair judicial outcomes Where can I try out llama2 models?
● You can try it out by setting it up locally from the GitHub repo: ○ Visit the AI at Meta website to accept our License and submit the form to gain access to the model. ○ Clone the repo from Llama 2 GitHub repository and follow the steps to get access to the model weights, and run examples of inference locally in your own machine. ○ Getting started with Llama: This guide provides information and resources to help you set up Llama including how to access the model, hosting, how-to and integration guides. Additionally, you will find supplemental materials to further assist you while building with Llama.
● On HuggingFace Spaces ○ Huggingface Spaces has hosted various Llama2 checkpoints fine-tuned for chat instructions on an interactive playground:
■ Llama2–7B-Chat
■ Llama2–13B-Chat
■ Llama2–70B-Chat
○ Typically, larger models are more powerful and have more capabilities, but they are also more expensive to run. What are the expectations regarding data sharing and model weights sharing from model trained/fine-tuned on non-public data? You are not obliged to disclose the trained model weights, or the data employed in its training. For the eligible organizations, does a research institution have to be the entire organization, or can it include a sub-unit within an organization dedicated to conducting research? We define a research organization as: an establishment outside of higher education founded for doing research. It can be public, private, or non-profit. As long as the legal entity applying for the program is eligible, then the sub-unit within the organization will be eligible for participation in the program. Can an eligible institution partner with a for-profit if the eligible institution is the entity applying? And tangentially related, may the for-profit be considered as a sub-awardee? Organizations that are not classified as a non-profit organization, an academic institution, or a research institution may advise/consult eligible organizations and offer expertise. Eligible organizations may list these consultants/partners in the Phase 1 application form. However, only eligible organizations are permitted to apply and be eligible to receive the grant award. Additionally, only eligible organizations are permitted to be designated as sub-organizations and sub-awardees and receive funding. Is the grant a flat $500K, or should we be budgeting to a specific amount under that ceiling figure? The grant will be a single flat award of $500,000 per track issued at the end of Phase 2. Should I outline how the entire budget will be used in my proposal? If so, how much detail would be required? You do not need to outline how the entire budget will be used in your Phase 1 proposal. It is suggested that you include a high level description of how you anticipate the funds will be used, however, a detailed budget will only be required for finalists who submit proposals in Phase 2 of the program. How many grants will be given per proposal track? One $500,000 award will be granted for each of the three proposal tracks. Three awards total. What kind of files should I upload to the application? We will not accept files uploaded to the application for Phase 1. You can input text or url links to external materials. There will be an opportunity to upload files for the Phase 2 finalist submissions.
