Add support for vLLM containers

[petermuller]

This commit adds support for vLLM containers with some example documentation to support it. The schema has been updated to allow for the vLLM container. These changes help toward the migration to the v2.x architecture.

Description of changes:

1. Update README with examples of usage
2. Update schema to guarantee no periods in names so that I can split on them later
3. Add Dockerfile and entrypoint script for vLLM-served models
4. Add a short-circuit for non-TGI/TEI containers so that we don't fail on "registering" the model, which will be deprecated as part of an upcoming v2 release.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[estohlmann]

Are these variables always apart of these base images? or is this something that we should make sure we add to the README?

[petermuller]

These are provided from the container build params within the typescript definition, so they should all be defined as part of this block:

LISA/lib/serve/ecs-model/index.ts

Lines 86 to 123 in 646e0fa

	/**
	* Generates environment variables for Docker at runtime based on the configuration. The environment variables
	* include the local model path, S3 bucket for models, model name, and other variables depending on the model type.
	*
	* @param {Config} config - The application configuration.
	* @param {ModelConfig} modelConfig - Configuration for the specific model.
	* @returns {Object} An object containing the environment variables. The object has string keys and values, which
	* represent the environment variables for Docker at runtime.
	*/
	private getEnvironmentVariables(config: Config, modelConfig: ModelConfig): { [key: string]: string } {
	const environment: { [key: string]: string } = {
	LOCAL_MODEL_PATH: `${config.nvmeContainerMountPath}/model`,
	S3_BUCKET_MODELS: config.s3BucketModels,
	MODEL_NAME: modelConfig.modelName,
	LOCAL_CODE_PATH: modelConfig.localModelCode, // Only needed when s5cmd is used, but just keep for now
	AWS_REGION: config.region, // needed for s5cmd
	};
	if (modelConfig.modelType === 'embedding') {
	environment.SAGEMAKER_BASE_DIR = config.nvmeContainerMountPath;
	}
	if (config.mountS3DebUrl) {
	environment.S3_MOUNT_POINT = 's3-models-mount';
	// More threads than files during S3 mount point copy to NVMe is fine; by default use half threads
	environment.THREADS = Math.ceil(Ec2Metadata.get(modelConfig.instanceType).vCpus / 2).toString();
	}
	if (modelConfig.containerConfig.environment) {
	for (const [key, value] of Object.entries(modelConfig.containerConfig.environment)) {
	if (value !== null) {
	environment[key] = String(value);
	}
	}
	}
	return environment;
	}