build-blaze-images

name	build-blaze-images
description	Trigger Tenstorrent Blaze media server and tt-metal upstream Docker image builds with gh. Use when the user asks to build Blaze and Metal Docker images, run the tt-shield media server image workflow, or build tt-metal images from the tt-llm-engine submodule reference.
disable-model-invocation	true

Build Blaze Images

Purpose

Use this skill to trigger both image builds:

• Blaze media server Docker image through tenstorrent/tt-shield workflow on-dispatch-build-media-server.yml.
• tt-metal upstream Docker images through tenstorrent/tt-metal workflow upstream-tests.yaml, using the tt-metal commit referenced by tt-media-server/cpp_server/tt-llm-engine.

Workflow

1. Work from the tt-inference-server repository root.

2. Resolve the currently checked out tt-inference-server ref for the Blaze build:

   inference_server_ref="$(git branch --show-current)"
   if [ -z "$inference_server_ref" ]; then
     inference_server_ref="$(git rev-parse HEAD)"
   fi
   printf '%s\n' "$inference_server_ref"
   

   Use this value for the tt-shield workflow's inference-server-git-ref input. If the current branch has not been pushed to tenstorrent/tt-inference-server, stop and ask the user whether to push it or use a commit/ref that exists on GitHub.

3. Inspect the tt-llm-engine submodule and the nested tt-metal submodule:

   git submodule status
   cd tt-media-server/cpp_server/tt-llm-engine
   git submodule status
   cd tt-metal
   git show -s --format='%H%n%D%n%s%n%ci' HEAD
   git branch -r --contains HEAD
   

4. Pick the tt-metal workflow ref:

   • Prefer a remote branch whose HEAD exactly matches the nested tt-metal submodule commit.
   • If no branch points at the commit, use the exact commit SHA if GitHub Actions accepts it.
   • If multiple branches contain the commit, prefer the branch shown in git show -s --format=%D HEAD if it names an origin/<branch> HEAD; otherwise ask the user which branch to use.

5. Check workflow inputs before dispatching:

   gh workflow view on-dispatch-build-media-server.yml --repo tenstorrent/tt-shield --yaml
   gh workflow view upstream-tests.yaml --repo tenstorrent/tt-metal --yaml
   

6. Trigger the Blaze image build from the currently checked out tt-inference-server ref. The tt-shield workflow input option is blaze-media-inference-server; if the user says blaze-media-server, use this workflow option.

   gh workflow run on-dispatch-build-media-server.yml \
     --repo tenstorrent/tt-shield \
     -f inference-server-type=blaze-media-inference-server \
     -f inference-server-git-ref="$inference_server_ref"
   

7. Trigger the tt-metal image build with default workflow inputs and the selected tt-metal ref:

   gh workflow run upstream-tests.yaml \
     --repo tenstorrent/tt-metal \
     --ref '<tt-metal-branch-or-sha>'
   

8. Verify both runs and report URLs:

   gh run list --repo tenstorrent/tt-shield \
     --workflow on-dispatch-build-media-server.yml \
     --event workflow_dispatch \
     --limit 5 \
     --json databaseId,displayTitle,headBranch,headSha,status,conclusion,url,createdAt,workflowName
   
   gh run list --repo tenstorrent/tt-metal \
     --workflow upstream-tests.yaml \
     --branch '<tt-metal-branch>' \
     --event workflow_dispatch \
     --limit 5 \
     --json databaseId,displayTitle,headBranch,headSha,status,conclusion,url,createdAt,workflowName
   

Reporting

In the final response, include:

• The Blaze workflow run URL, status, and server type.

• The Metal workflow run URL, status, branch or SHA, and the submodule commit used.

• Monitoring commands:

  gh run watch <shield-run-id> --repo tenstorrent/tt-shield
  gh run watch <metal-run-id> --repo tenstorrent/tt-metal
  gh run view <metal-run-id> --repo tenstorrent/tt-metal --log-failed
  

Failure Investigation

If a run fails, collect more information with:

gh run view <run-id> --repo <owner/repo> --json status,conclusion,jobs,url
gh run view <run-id> --repo <owner/repo> --log-failed

For image publishing uncertainty, inspect the workflow job outputs and the calculate-to-publish job logs in tt-metal, because upstream-tests.yaml may publish only images whose tests passed unless request-force-publish is changed from its default.