Skip to content

Docker size - #133

Merged
ali6parmak merged 9 commits into
mainfrom
docker-size
May 6, 2026
Merged

Docker size#133
ali6parmak merged 9 commits into
mainfrom
docker-size

Conversation

@ali6parmak

Copy link
Copy Markdown
Collaborator
  • Docker files improved & optimized
  • Image size reduced
  • Support for old and new GPUs added

@ali6parmak ali6parmak self-assigned this May 6, 2026
@ali6parmak
ali6parmak merged commit f361d1e into main May 6, 2026
1 check passed
@ali6parmak
ali6parmak deleted the docker-size branch May 6, 2026 11:27
ali6parmak pushed a commit that referenced this pull request Jul 13, 2026
Extends select_gpu_stack.sh's existing legacy/nextgen mechanism (added
in #133) with an explicit grace profile for NVIDIA Grace-Blackwell
(aarch64) hosts such as the DGX Spark: CUDA 13.0.3 base image, cu130
torch wheels, sm_120/sm_121 arch list.

nextgen's compute-capability check already matches these GPUs and
their CUDA base images already ship arm64 manifests, so nextgen runs
on this hardware today via PTX JIT. grace is an explicit opt-in for
native SASS instead, verified end-to-end (build, GPU-backed inference,
smoke tests) on a DGX Spark (GB10, sm_121, driver 580/CUDA 13.0).

README documents the new profile, the verified build-arg values, and
what was and wasn't observed on this hardware.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant