#｜ hidesolveit-modal
Grant SolveIt access to a GPU via Modal
SolveIt doesn’t come with GPU access. But the platform is too good to ignore. This library allows you to give SolveIt free GPU access, with the free monthly US$30 credits Modal provides
Contributions are greatly appreciated to make this library even better.
Installation
Install latest from the GitHub repository:
$ pip install git+https://github.com/ForBo7/solveit-modal.gitor from pypi
$ pip install solveit_modalDocumentation
Documentation can be found hosted on this GitHub repository’s pages. Additionally you can find package manager specific guidelines on pypi.
How to use
After installing, if you haven’t already, run !modal setup to authenticate your Modal account to SolveIt.
You will also need a SSH_PUBKEY enivronment variable. Run !ssh-keygen -f ~/.ssh/id_ed25519 -y to check if you already have one available. If you don’t, run !ssh-keygen -t ed25519 -f ~/.ssh/id_ed25519 -N '' -q to generate one. Run !ssh-keygen -f ~/.ssh/id_ed25519 -y once more to copy your newly generated public key and save it as an environment variable in SolveIt called SSH_PUBKEY.
Then, it should be as simple as ABC.
from solveit_modal.core import *sb, ssh = gpu_on()INFO - ∞ creating sandbox; this may take 5-10 minutes if you are creating this sandbox for the first time... | 2026-06-19 06:55:13,138
INFO - ✔ sandbox ready | 2026-06-19 06:55:13,677
INFO - ✔ gotten tunnel: r449.modal.host:34889 | 2026-06-19 06:55:28,772
INFO - ✔ public key injected | 2026-06-19 06:55:29,744
INFO - ✔ started ssh service | 2026-06-19 06:55:31,831
INFO - ∞ starting kernel | 2026-06-19 06:55:33,421System: Linux
Hostname: modal
User: root
Kernel: 4.4.0
Architecture: x86_64
OS Type: GNU/Linux
GPU: Tesla T4INFO - ✔ remote kernel ready: /root/.local/share/jupyter/runtime/kernel-ipyf.json | 2026-06-19 06:55:38,413ipyf_remote_kernel is already a registered kernel
/app/data/.ssh/config file updated.
Successfully created connection file and forwarded ports!INFO - ✔ connected to remote kernel | 2026-06-19 06:55:40,457
WARNING - ! remote kernel environment has a different set of libraries installed | 2026-06-19 06:55:40,458Success: connected to remote kernel via r449.modal.host:34889You are provided with a Modal sandbox object, sb, and also an SSH function, ssh, allowing you to directly interact with your Modal sandbox with bash commands.
ssh('hostname')'modal'Of course, since the goal is run code in the GPU enabled environment, we can skip ssh and directly take advantage of IPython with the %remote line magic or %%remote cell magic!
#%%remote <-- cell magic (remove # when actually using)
!hostnamemodal#%%remote <-- cell magic (remove # when actually using)
!nvidia-smiWed Jun 17 07:03:35 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla T4 On | 00000000:18:00.0 Off | 0 |
| N/A 28C P8 10W / 70W | 0MiB / 15360MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+If you want to avoid having to use magics, do gpu_on(sticky=True). If you have set sticky, you will have to use the %local line magic or %%local cell magic to run code locally, and when manually turning the GPU off. However, setting sticky will prevent SolveIt’s dialoghelper tools from working.
Note that the Python libraries preinstalled in this Modal sandbox are different from those preinstalled in SolveIt.
default_pips['ipykernel',
'fastai',
'transformers',
'diffusers',
'accelerate',
'datasets',
'huggingface_hub',
'peft',
'safetensors',
'sentence-transformers',
'xformers',
'bitsandbytes',
'ninja',
'einops',
'wandb',
'gradio',
'python-fasthtml',
'plotly',
'ipywidgets',
'fsspec',
's3fs',
'gcsfs',
'librosa',
'imageio',
'imageio-ffmpeg']fastai includes PyTorch, Numpy, Matplotib, Pandas, and SciPy, amongst amother libraries. See here for more info.
By default, the sandbox switches off after 30 minutes if there is no activity. To turn off the sandbox manually, pass the sb object to gpu_off.
gpu_off(sb)INFO - ✔ unlinked from remote kernel | 2026-06-17 07:05:19,170
INFO - ✔ terminated sandbox | 2026-06-17 07:05:19,304Code cells will now run locally.
Code cells will now run locally.
Shutting down remote kernelEvery sandbox is stored under an app on Modal. You will see on your Modal dashboard that the app created to run this sandbox will still be running. AFAIK, you will not be charged for the app being open.
gpu_on allows you to customize the following sandbox parameters out of the box. By default, secrets in your SolveIt instance are also passed along. If you don’t want to, set secrets={}.
?gpu_onSignature: gpu_on( app_name: 'str' = 'solveit-modal', pips: 'list' = ['ipykernel', 'fastai', 'transformers', 'diffusers', 'accelerate', 'datasets', 'huggingface_hub', 'peft', 'safetensors', 'sentence-transformers', 'xformers', 'bitsandbytes', 'ninja', 'einops', 'wandb', 'gradio', 'python-fasthtml', 'plotly', 'ipywidgets', 'fsspec', 's3fs', 'gcsfs', 'librosa', 'imageio', 'imageio-ffmpeg'], apts: 'list' = ['openssh-server'], vol_name: 'str' = 'solveit-volume', timeout: 'int' = 1800, gpu: 'str' = 'T4', secrets: 'dict' = None, sticky: 'bool' = False, ) -> 'tuple' Docstring: Provision a GPU sandbox, wire up SSH, and hijack cells onto a remote kernel. File: Dynamically generated function. No source code available. Type: function
solveit_modal also copies all secrets on SolveIt to the Modal Sandbox by default. If you don’t want to, set gpu_on(..., secrets={}).
?get_secretsdef get_secrets(
)->dict:Secrets from solveit_settings.json.File: ~/库/solveit-modal/solveit_modal/core.py
Type: function
You can also explore GPUs on Modal with solveit_modal. These rates are valid as of 22 May, 2026.
from solveit_modal.gpu import *
gpu = GPU(); gpuT4 · $0.59/hr · 16GB
L4 · $0.80/hr · 24GB
A10 · $1.10/hr · 24GB
L40S · $1.95/hr · 48GB
A100-40GB · $2.10/hr · 40GB
A100-80GB · $2.50/hr · 80GB
RTX-PRO-6000 · $3.03/hr · 96GB
H100 · $3.95/hr · 80GB
H200 · $4.54/hr · 141GB
B200 · $6.25/hr · 180GB
gpu.by_vram(min_gb=36)L40S · $1.95/hr · 48GB
A100-40GB · $2.10/hr · 40GB
A100-80GB · $2.50/hr · 80GB
RTX-PRO-6000 · $3.03/hr · 96GB
H100 · $3.95/hr · 80GB
H200 · $4.54/hr · 141GB
B200 · $6.25/hr · 180GB
gpu.cheapest()T4 · $0.59/hr · 16GB
Note that these are only GPU charges. Modal charges additionally for storage, RAM, and CPU. See Modal’s full pricing breakdown here.
Developer Guide
This library is built using nbdev, a way to create delightful software with Jupyter Notebooks. Learn how to get started with nbdev here.
Install solveit_modal in Development mode
# make sure solveit_modal package is installed in development mode
$ pip install -e .
# make changes under nbs/ directory
# ...
# compile to have changes apply to solveit_modal
$ nbdev_prepareAfter cloning, run nbdev-install-hooks in your terminal to install Jupyter and git hooks that clean, trust, and fix merge conflicts in notebooks.
Anytime you make changes in the repo, run nbdev-prepare.
Credit
This library is built using nbdev on SolveIt, both by Answer.AI. If you are using SolveIt to explore this library, use the provided CONTROLLER dialog to get started.
Also thanks to Dr. Scott Hawley for his super easy to use ipyfernel library, which allows you straightfowardly swap IPython kernels between SolveIt and other machines.
This library also uses the Modal API.