ci(fix): NGC build

[ko3n1g]

Description

Please include a brief summary of the changes, relevant motivation and context.

Fixes # (issue)

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[greptile-apps]

Greptile Overview

Greptile Summary

This PR refactors the NGC PyTorch image build workflow to fix several structural issues. The changes include:

• Fixed the NGC images matrix structure by creating a proper matrix object with os, release-version, and container-image fields instead of just passing an array of images
• Added the missing release-version parameter to the NGC build step (line 197)
• Standardized quote style from single to double quotes throughout the workflow
• Added permissions: write-all to both build jobs

Critical Issue Remaining:

The build_wheels_for_ngc job still has a blocking issue: it only passes release-version and base-image to the build action, but the action definition marks python-version, cuda-version, cudnn-version, torch-version, cxx11_abi, and aarch as required: true. GitHub Actions will fail validation when these required inputs are missing, even though they're not used when base-image is provided.

The action logic correctly handles the optional use of these parameters (checking if BASE_IMAGE is empty), but the workflow won't pass GitHub's validation phase.

Confidence Score: 2/5

• This PR will fail during GitHub Actions validation due to missing required action inputs
• While the PR addresses several previous issues (matrix structure, release-version parameter), it leaves a critical validation error: the build action requires 6 parameters that aren't being passed for NGC builds. The workflow cannot execute successfully until this is resolved.
• .github/workflows/attach-wheels-to-release.yml requires attention - specifically the NGC build action call needs required parameters added

Important Files Changed

Filename	Overview
.github/workflows/attach-wheels-to-release.yml	Refactored NGC build matrix structure and fixed release-version parameter, but still missing required action inputs for NGC builds

Sequence Diagram

sequenceDiagram
    participant User
    participant GHA as GitHub Actions
    participant PreFlight as pre-flight job
    participant BuildWheels as build_wheels job
    participant BuildNGC as build_wheels_for_ngc job
    participant Action as build-pytorch-wheel action
    participant Release as GitHub Release

    alt Release Event
        User->>GHA: Publish Release
        GHA->>PreFlight: Trigger with release event
        PreFlight->>PreFlight: Build matrix (release mode)
        PreFlight->>PreFlight: Run check_for_ngc_images.sh
        PreFlight->>PreFlight: Create ngc-images-matrix with os, release-version, container-image
    else Workflow Dispatch
        User->>GHA: Manually trigger with inputs
        GHA->>PreFlight: Trigger with workflow_dispatch
        PreFlight->>PreFlight: Build matrix (inputs mode)
        PreFlight->>PreFlight: Create ngc-images-matrix from inputs
    end

    PreFlight->>BuildWheels: Pass build-wheel-matrix
    PreFlight->>BuildNGC: Pass ngc-images-matrix

    alt NGC Image Not Provided
        BuildWheels->>Action: Call with all params (python-version, cuda-version, etc.)
        Action->>Action: Build custom Docker image from scratch
        Action->>Action: Build wheel in container
        Action->>BuildWheels: Return wheel
        BuildWheels->>Release: Upload wheel as release asset
    end

    alt NGC Image Provided
        BuildNGC->>Action: Call with release-version + base-image only
        Note over BuildNGC,Action: ⚠️ Missing required params:<br/>python-version, cuda-version,<br/>cudnn-version, torch-version,<br/>cxx11_abi, aarch
        Action->>Action: Pull NGC Docker image
        Action->>Action: Build wheel in NGC container
        Action->>BuildNGC: Return wheel
        BuildNGC->>Release: Upload wheel as release asset
    end

Loading

[greptile-apps]

_{1 file reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

In release events, inputs.release-version is undefined. Should use matrix.release-version instead.

Suggested change

	release-version: ${{ inputs.release-version }}
	release-version: ${{ matrix.release-version }}

[greptile-apps]

Missing required parameters: python-version, cuda-version, cudnn-version, torch-version, cxx11_abi, and aarch. The action at .github/actions/build-pytorch-wheel/action.yml:12-32 requires these inputs when base-image is not provided. Since NGC images may not have these pre-configured, ensure they're either in the matrix or passed explicitly.

[greptile-apps]

_{1 file reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

JSON syntax error: container-image should be an array. The current command substitution with double quotes will produce invalid JSON.

Suggested change

	"container-image": "$(cat ngc_images.json | jq -cr)"
	"container-image": $(cat ngc_images.json | jq -cr)

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Missing required action inputs. The action at .github/actions/build-pytorch-wheel/action.yml:12-32 marks python-version, cuda-version, cudnn-version, torch-version, cxx11_abi, and aarch as required: true. Even though they're not used when base-image is provided, GitHub Actions will fail validation without them.

Either:

1. Change the action definition to mark these as required: false, OR
2. Pass placeholder values here (e.g., from the matrix or hardcoded defaults)