eniktab

Triton on NVIDIA Blackwell sm_121 (GB10 / DGX Spark): the two env vars that fix it

PyTorch 2.9 advertises max compute capability sm_120. Blackwell consumer / GB10 parts report sm_121. Triton (and anything that lowers through it — torch.compile, FlashAttention, hand-written kernels) fails with "Triton Error [CUDA]: no kernel image is available for execution on the device".

The fix is two env vars plus removing one that often appears in "workarounds":

eniktab

function KeepClicking(){
console.log("Clicking");
document.querySelector("colab-connect-button").click()
}
setInterval(KeepClicking,60000)

eniktab

from google.colab import auth
auth.authenticate_user()

project_id = '[your Cloud Platform project ID]'
!gcloud config set project {project_id}

!gsutil cp /tmp/to_upload.txt gs://{bucket_name}/

eniktab

find . -maxdepth 1 -name "*.jpg" -print0 | xargs -0 rm

eniktab

#get a Token 
# Install colab_ssh
!pip install colab_ssh --upgrade

from colab_ssh import launch_ssh
launch_ssh('Token', 'SOME_PASSWORD')

eniktab

SELECT *
FROM INFORMATION_SCHEMA.COLUMNS
WHERE COLUMN_NAME LIKE '%name%'