ODSC 68320: Modify the AQUA API to Accept GPU Count as an Optional Input

[elizjo]

• Updated the GET aqua/deployments/<model_ocid>/params endpoint to return vLLM Parameters specifically for a set number of GPUs
• accepts gpu_count as an optional parameter
• used the new AQUA Service Managed Model Configs with 'multi_model_deployment' key

Request
GET aqua/deployments/<model_ocid>/params?instance_shape=<shape_name>&gpu_count=<gpu_count_value>

Output

{
    "data": [
        "--max-model-len 4096"
    ]
} 

Modified the test_get_deployment_default_params() unit test
we test cases such as:

• gpu_count = 1, we return the vLLM params as specified by the gpu_count : 1 field in the AQUA Service Managed Model Configs with 'multi_model_deployment' key
• gpu_count = 2, we return the vLLM params as specified by the gpu_count : 2 field in the AQUA Service Managed Model Configs with 'multi_model_deployment' key
• gpu_count is None, we always return the single model vLLM params
• TGI is attached as the inference container for the model and gpu_count is specified (always returns data: [] or single model TGI params since TGI is not supported by multi model)

[darenr]

Also really nice code.

[VipulMascarenhas]

lgtm 👍