Add GitHub Actions workflow for Llama3.1 8B training

[kylasa]

New workflow to trigger pre training for llama3.1 8B model using Primus/Megatron framework.

[wenxie-amd]

LGTM

[Copilot]

Pull Request Overview

This PR introduces a new GitHub Actions workflow to automate pre-training of the Llama3.1 8B model using the Primus/Megatron framework. The workflow includes Docker image building, environment setup, training execution, and artifact upload capabilities.

Key Changes

• Added a workflow with build and train jobs that support multiple GPU runner configurations
• Integrated Docker-based training environment with AWS S3 for log storage
• Implemented validation checks for required data and model directories

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Invalid syntax for default value in workflow run-name. The pipe operator | is not valid YAML syntax for providing a default value. Use || instead: ${{ inputs.runner_label || 'm13-21' }}

Suggested change

	run-name: Primus Llama3.1 8B | ${{ inputs.runner_label | "m13-21" }}
	run-name: Primus Llama3.1 8B | ${{ inputs.runner_label || 'm13-21' }}

[Copilot]

The workflow references inputs.image_tag but this input is not defined in the workflow_dispatch inputs section (lines 5-15). Either add this input definition or remove this condition.

[Copilot]

The variable DOCKER_CREDENTIALS is referenced but never defined. This line should either be removed or the variable should be properly defined.

Suggested change

echo "$DOCKER_CREDENTIALS"

[Copilot]

The workflow references inputs.skip_build but this input is not defined in the workflow_dispatch inputs section. Either add this input definition or remove it from the condition.

[Copilot]

The UUID environment variable is being set incorrectly. The correct format for setting environment variables in GitHub Actions is echo \"UUID=$UUID\" >> $GITHUB_ENV (note the equals sign, not a colon).

Suggested change

	echo "UUID: $UUID" >> $GITHUB_ENV
	echo "UUID=$UUID" >> $GITHUB_ENV

[Copilot]

The workflow attempts to change to /workspace directory, but this is running on a GitHub Actions runner, not inside a container. This will likely fail as /workspace may not exist on the runner. Consider using $GITHUB_WORKSPACE or a relative path.

Suggested change

	cd /workspace
	cd $GITHUB_WORKSPACE

[Copilot]

Using docker exec -it with interactive terminal flags in a CI environment will cause the workflow to hang indefinitely. Remove the -it flags and provide the command to execute directly, or restructure to run commands within the container non-interactively.

Suggested change

	docker exec -it dev_primus bash
	cd Primus && pip install -r requirements.txt
	# Run Training
	EXP=examples/megatron/configs/llama3.1_8B-pretrain-mlperf.yaml bash ./examples/run_pretrain_mlperf.sh
	docker exec dev_primus bash -c "cd Primus && pip install -r requirements.txt && EXP=examples/megatron/configs/llama3.1_8B-pretrain-mlperf.yaml bash ./examples/run_pretrain_mlperf.sh"

[Copilot]

The workflow references inputs.skip_batch_training but this input is not defined in the workflow_dispatch inputs section. Either add this input definition or remove this condition.

[Copilot]

Trailing whitespace detected at the end of line 59. Remove the extra spaces for consistency.

Suggested change

	custom-image-tag: ${{ env.BASE_IMAGE_TAG }}
	custom-image-tag: ${{ env.BASE_IMAGE_TAG }}

[Copilot]

Trailing whitespace detected at the end of line 88. Remove the extra spaces for consistency.

Suggested change