AWS Bedrock - Add support for cross-region inference new feature

[DEG-7]

Hey team,
First of all, thanks for the effort you are doing for this amazing project.
I would like to ask for the support of a recently and very important addition in AWS Bedrock, that's the cross-region inference support.
You can find all the tech information in AWS doc:
https://aws.amazon.com/blogs/machine-learning/getting-started-with-cross-region-inference-in-amazon-bedrock/

Thanks in advance

[dafriz]

You should be able to use withought any code changes by adding a "us." or "eu." prefix to the modelId.

Some extra enums could be added but wondering if the extra code is useful enough to add - something like this

/**
 * Cross Region Inference mapping
 * https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference-support.html
 */
public enum CrossRegion {

	US("us", Set.of(Region.US_EAST_1, Region.US_WEST_2)),
	EU("eu", Set.of(Region.EU_CENTRAL_1, Region.EU_WEST_1, Region.EU_WEST_3));

	private final String prefix;
	private final Set<Region> regions;

	public String prefix() {
		return prefix;
	}

	public Set<Region> regions() {
		return regions;
	}

	CrossRegion(String prefix, Set<Region> regions) {
		this.prefix = prefix;
		this.regions = regions;
	}

}

and then adding more enums for Llama and Anthropic like below

/**
     * Cross Region Inference Llama models version.
	 * https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference-support.html
     */
	public enum CrossRegionLlamaChatModel implements ChatModelDescription {

		US_LLAMA3_2_1B_INSTRUCT_V1(CrossRegion.US, LlamaChatModel.LLAMA3_2_1B_INSTRUCT_V1),
		US_LLAMA3_2_3B_INSTRUCT_V1(CrossRegion.US, LlamaChatModel.LLAMA3_2_3B_INSTRUCT_V1),
		US_LLAMA3_2_11B_INSTRUCT_V1(CrossRegion.US, LlamaChatModel.LLAMA3_2_11B_INSTRUCT_V1),
		US_LLAMA3_2_90B_INSTRUCT_V1(CrossRegion.US, LlamaChatModel.LLAMA3_2_90B_INSTRUCT_V1),
		EU_LLAMA3_2_1B_INSTRUCT_V1(CrossRegion.EU, LlamaChatModel.LLAMA3_2_1B_INSTRUCT_V1),
		EU_LLAMA3_2_3B_INSTRUCT_V1(CrossRegion.EU, LlamaChatModel.LLAMA3_2_3B_INSTRUCT_V1);

		private final String id;

		CrossRegionLlamaChatModel(CrossRegion crossRegion, LlamaChatModel model) {
			this.id = crossRegion.prefix() + '.' + model.id;
		}

		public String id() {
			return id;
		}

		@Override
		public String getName() {
			return this.id;
		}

	}

Thoughs on what support would be helpful ? Do we want to validate if someone uses a cross region inference modelId but then uses a region that's not supported in that cross-region ? ( That's where the mapping enum might be useful ).

[DEG-7]

I agree, it sounds good to me. As you mentioned, it's not absolutely essential, and we could simply add the prefix 'us.' or 'eu.' to the model ID. However, having a specific enum for cross-region models (for those that support it) would be nicer and a better approach than just adding a string prefix.