OpenToken — Request Compute

Create Your OpenToken Account

Set up your account to submit your project and track its progress through our pipeline.

Email address *

Password *

Confirm password *

or

Already have an account? Sign in here

About You & Your Institution

1.1 Your name *

1.2 Your role / title *

1.3 Institution name *

1.4 Country *

1.5 Institution type *

University / Research Institute Government Agency Startup / SME NGO / Social Enterprise International Organisation Other

1.6 Department / Lab / Team name — optional

1.7 Your email address *

Current Compute Access

Understanding what you have today helps us understand the gap OpenToken needs to fill.

2.1 What compute resources does your institution currently have for AI/ML work?

Personal laptops / desktops (CPU only) Personal machines with consumer GPUs (e.g. RTX 3060, 4070) Departmental GPU server(s) Institutional HPC / compute cluster Cloud provider account (AWS, GCP, Azure, etc.) Third-party GPU rental (Lambda, Paperspace, Vast.ai, etc.) Google Colab / Kaggle (free tiers) None — no dedicated AI compute

2.2 If you have local GPU hardware, please describe it briefly

2.3 What is your approximate annual budget for AI/ML compute? Include cloud spend, hardware purchases, and any compute-specific grants

$0 — no dedicated compute budget Under $500 $500 – $2,000 $2,000 – $5,000 $5,000 – $10,000 $10,000 – $50,000 Over $50,000

2.4 Where does your compute budget come from?

Personal / out-of-pocket Institutional operating budget Research grant Government programme Industry partnership / sponsorship Cloud provider credit programme (e.g. Google Research Credits, AWS Educate) No dedicated funding — compute is ad hoc

2.5 How would you rate the reliability and availability of your current compute?

Sufficient — compute is rarely a bottleneck Somewhat constrained — I sometimes wait or scale down work Significantly constrained — compute regularly limits what I can do Severely constrained — I cannot run many of the experiments I need to No access — I have effectively no usable AI compute

What You Need Compute For

This section helps us understand the types of workloads our platform must support, and at what scale.

3.2 What model sizes do you typically work with (or need to work with)? This helps us understand GPU memory and multi-GPU requirements

Small models (<1B parameters) Medium models (1–8B parameters) Large models (8–70B parameters) Very large models (>70B parameters) Not sure / varies widely

3.3 What GPU memory (VRAM) do you typically need per job?

3.4 How long do your typical compute jobs run?

Minutes (quick experiments, inference) Hours (model fine-tuning, medium training runs) Days (large training runs) Weeks or longer (foundation model training) Mix of the above

3.5 How many GPU-hours per month would you need if cost were not a constraint? 1 GPU-hour = 1 GPU running for 1 hour. A 24-hour fine-tuning job on 4 GPUs = 96 GPU-hours.

Under 100 GPU-hours/month 100 – 500 GPU-hours/month 500 – 2,000 GPU-hours/month 2,000 – 10,000 GPU-hours/month Over 10,000 GPU-hours/month Not sure — I lack the experience to estimate

3.6 What frameworks and tools do you use?

PyTorch TensorFlow / Keras Hugging Face Transformers / Datasets JAX scikit-learn Custom / in-house code Jupyter Notebooks

Barriers, Pain Points & Workarounds

4.1 What are the biggest barriers to accessing AI compute for your work? Select your top 3

Cost of cloud GPU services No institutional budget for compute Slow or restrictive grant/procurement processes Lack of local GPU hardware Unreliable electricity or internet Export controls / cannot purchase advanced GPUs Lack of technical skills to use cloud/HPC systems Data sovereignty concerns (data must stay in-country) No institutional IT support for GPU infrastructure

4.2 What workarounds do you currently use when compute is insufficient?

Use smaller models / datasets than ideal Run fewer experiments / hyperparameter searches Wait in queues for shared institutional resources Use free tiers (Colab, Kaggle) with limitations Pay out of pocket Collaborate with better-resourced institutions abroad Change research direction to less compute-intensive topics

4.3 How urgent is your compute need?

Critical — I have active projects stalled for lack of compute now High — I will need compute within the next 1–3 months Moderate — I will need compute within 3–6 months Planning — I am designing projects that will need compute in 6–12 months

Platform Preferences & Data Sovereignty

Help us design the right platform experience for you.

5.1 How important is data sovereignty to your work? Data sovereignty means your data is stored and processed within a specific legal jurisdiction

Essential — my data must remain within my country's jurisdiction Important — I prefer in-country but could accept EU-hosted (GDPR-compliant) infrastructure Nice to have — I'd prefer it but cost/availability matters more Not a concern for my current work

5.2 Do any of your projects involve sensitive data?

Health / medical records Government / citizen data Financial data Personally identifiable information (PII) No — my data is publicly available or non-sensitive

5.3 What would make you trust a new compute platform enough to use it?

Endorsement by my institution or a trusted academic network GDPR or equivalent privacy compliance certification Open-source platform code Clear data handling and deletion policies In-country data residency Backing by a known university or research organisation Track record of serving similar institutions

5.4 What price per GPU-hour would make cloud compute accessible for your work? Current hyperscale pricing is $2.50–$4.00/GPU-hour for high-end GPUs

Free — I have no compute budget Under $0.25/GPU-hour $0.25 – $0.50/GPU-hour $0.50 – $1.00/GPU-hour $1.00 – $2.00/GPU-hour Current cloud pricing is acceptable for my budget

5.5 What additional platform features would be valuable to you?

Pre-installed ML frameworks and model libraries Dataset hosting and management Collaboration tools (shared projects, team access) Model hosting / inference endpoints after training Training and tutorials on using GPU compute effectively Community forum to connect with other researchers Usage analytics and cost reporting

Local Infrastructure & Connectivity

Understanding your local conditions helps us design for real-world constraints.

6.1 How reliable is your internet connection?

Very reliable — consistent broadband Generally reliable — occasional outages Unreliable — frequent interruptions or very slow speeds Very limited — connectivity is a major constraint

6.2 What is your typical internet bandwidth?

6.3 How reliable is your electricity supply?

Very reliable — rare outages Generally reliable — occasional outages (backup available) Unreliable — frequent outages affect work Major constraint — outages are daily / no backup

6.4 How large are the datasets you typically work with? This helps us understand data transfer requirements

Small (under 1 GB) Medium (1–50 GB) Large (50–500 GB) Very large (over 500 GB) Varies widely

Your Voice

7.1 If you could describe the single biggest compute challenge you face in one sentence, what would it be?

7.2 Is there a specific project or research question you have been unable to pursue due to lack of compute? Please describe it briefly.

7.3 What else should we know about the compute needs of your institution or region?

Part B: Project Intake Form

Complete this section if you have a live or near-ready AI/ML project that could benefit from donated or subsidised compute. These projects may be selected for our first pilot cohort. If you do not have a project to submit, you may skip Part B entirely.

What makes a strong pilot project?

The project has a clear AI/ML compute need
The team is ready to use compute if provided (code exists or is near-ready)
The work addresses a real problem relevant to the Global South
The project can produce visible results within 3–6 months

Project Identity

B1.1 Project title

B1.2 Lead researcher / project lead name and email

B1.3 Institution and department

B1.4 Project website or repository (if available)

B1.5 Describe the project in 2–3 sentences. What problem does it address and why does it matter?

B1.6 What is the intended real-world impact? Who benefits?

B1.7 Project domain

Healthcare / Medical AI Agriculture / Food Security Natural Language Processing / Local Languages Climate / Environment / Remote Sensing Education Financial Inclusion Governance / Digital Public Infrastructure

Technical Requirements

Be as specific as you can. This helps us match the right resources to your project.

B2.1 What type of compute work does this project require?

Model training from scratch Fine-tuning a pre-trained model Inference / evaluation at scale Data processing / feature engineering

B2.2 What model(s) or architecture(s) are you working with?

B2.3 What is the model size (parameters)?

Under 1B parameters 1–8B parameters 8–70B parameters Over 70B parameters Not applicable / custom architecture

B2.4 Minimum GPU memory (VRAM) required per GPU

B2.5 Number of GPUs needed simultaneously

1 GPU 2–4 GPUs 4–8 GPUs 8+ GPUs Flexible — can work with what is available

B2.6 Estimated total GPU-hours needed for this project If unsure, describe the workload in the next question and we can help estimate

Under 500 GPU-hours 500 – 2,000 GPU-hours 2,000 – 10,000 GPU-hours 10,000 – 50,000 GPU-hours Over 50,000 GPU-hours Not sure

B2.6a If unsure about GPU-hours, describe your workload here and we will help estimate

B2.7 What ML frameworks and key dependencies does the project use?

B2.8 Estimated dataset size

B2.9 Does the project involve sensitive data?

No — data is public or non-sensitive Yes — health/medical data Yes — personally identifiable information Yes — government/citizen data

Project Readiness & Timeline

B3.1 Current project stage

Concept / planning — no code yet Early development — code in progress, initial experiments Ready to scale — code works, need compute for full experiments Scaling up — proven approach, need more compute for larger runs Production — deploying model, need inference compute

B3.2 If compute were available tomorrow, could you use it within:

1 week 2–4 weeks 1–3 months Longer — other dependencies must be resolved first

B3.3 What would you deliver within 3 months and within 6 months of receiving compute?

B3.4 Are there any dependencies beyond compute that could affect project delivery?

Impact, Sustainability & Openness

B4.1 What will happen to the outputs of this project?

Published as open-source model / code Published as academic paper Deployed as a service for end users Shared with government / public sector partners Licensed commercially

B4.2 Would you be willing to share your project publicly on the OpenToken platform? Visible to volunteer donors who may contribute compute to your project

Yes — happy to share project details and progress updates Partially — willing to share a description but not all details No — the project must remain confidential

B4.3 How will this project sustain itself after the initial compute allocation? OpenToken wants to support projects that have a path forward

Applying for research grants Institutional funding Revenue from the product/service being built Will need continued donated compute Project is time-bounded (training completes, no ongoing need)

B4.4 Has this project received any prior compute support or grants? If yes, please describe briefly.

B4.5 Anything else you want us to know about this project?

Submit Your Project for Compute

Get Started in 3 Steps