Project config lives in tccp.yaml at the root of your project.
The only required field is container. Everything else has
reasonable defaults.
container: pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime
That's it. tccp will pull this container, start a shell inside it on a compute node with a GPU, and sync your project files.
container: pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime init: "pip install -r requirements.txt" gpu: a100 gpu-count: 1 partition: gpu cpus: 4 memory: 32G time: 4h output: output/ ports: [6006, 8888] rodata: - data - models/pretrained
| field | default | description |
|---|---|---|
container |
(required) | Docker image to run. Pulled as a Singularity SIF file.
e.g. pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime,
python:3.11-slim, or any Docker Hub / NGC image. |
init |
(none) | Shell command run inside the container after file sync, before the
persistent shell starts. Runs once per tccp start.
e.g. pip install -r requirements.txt |
host |
(global) | Override the DTN/transfer host for this project. |
login |
(global) | Override the login/scheduler node for this project. |
output |
output/ |
Directory synced back to your laptop on tccp sync and
tccp stop. Bind-mounted from NFS into the container, so
files written there persist immediately. |
ports |
(none) | Ports forwarded to localhost during tccp shell. Can be
a single number or a list. e.g. [6006, 8888] |
rodata |
(none) | Read-only data directories bind-mounted from your NFS home directory into the scratch dir. Avoids syncing large datasets every time. |
If no init: is set in the config, but tccp_init.sh
exists in your project root, tccp runs bash tccp_init.sh
automatically.
rodata: - data - models/pretrained
These directories are bind-mounted from
/cluster/home/{user}/{dir} into the scratch dir inside the
container. They're read-only and don't need to be synced.
These control what SLURM allocates. Set them in tccp.yaml
or in the global config for defaults.
| field | default | description |
|---|---|---|
gpu |
(auto) | GPU type. e.g. a100, v100, t4.
If omitted, tccp auto-selects the best available GPU by VRAM on the
partition. See GPU guide. |
gpu-count |
1 | Number of GPUs. |
partition |
gpu |
SLURM partition. e.g. gpu, preempt. |
cpus |
4 | Number of CPUs. |
memory |
32G |
RAM allocation. |
time |
4h |
Walltime limit. Accepts 4h, 30m,
1d, or HH:MM:SS format. |
Created by tccp setup. Stores credentials and default
resource settings. Project tccp.yaml overrides these.
| field | default | description |
|---|---|---|
host | xfer.cluster.tufts.edu | DTN/transfer node hostname |
login | login.pax.tufts.edu | Login/scheduler node (sbatch, squeue) |
user | (required) | Cluster username |
password | (required) | Cluster password |
partition | gpu | Default SLURM partition |
gpu | (none) | Default GPU type |
gpu-count | 1 | Default GPU count |
cpus | 4 | Default CPUs |
memory | 32G | Default RAM |
time | 4h | Default walltime |
cache-containers | false | When true, store container SIF on NFS (persists across nodes). When false, store on compute /tmp (faster but ephemeral). |
Settings merge in this order (later wins):
~/.tccp/config.yaml)tccp.yaml)v