Model Builder: Add an API for listing and testing supported models

[samuel100]

In the create_model function of Model Builder (MB), there is a test to check whether the model architecture is supported:

onnxruntime-genai/src/python/py/models/builder.py

Lines 2654 to 2687 in dac90d5

	if "config_only" not in extra_options:
	# List architecture options in alphabetical order
	if config.architectures[0] == "GemmaForCausalLM":
	onnx_model = GemmaModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Gemma2ForCausalLM":
	onnx_model = Gemma2Model(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "LlamaForCausalLM":
	onnx_model = LlamaModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "MistralForCausalLM":
	onnx_model = MistralModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "PhiForCausalLM":
	onnx_model = PhiModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Phi3ForCausalLM" and config.max_position_embeddings == 4096:
	onnx_model = Phi3Mini4KModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Phi3ForCausalLM" and config.max_position_embeddings == 131072:
	onnx_model = Phi3Mini128KModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "PhiMoEForCausalLM" and config.max_position_embeddings == 131072:
	print("WARNING: This model only works for CUDA currently because `MoE` is only supported for CUDA in ONNX Runtime. Setting `--execution_provider cuda` by default.")
	print("WARNING: This model currently only supports quantized version. Setting `--precision int4` by default.")
	execution_provider = "cuda"
	precision = "int4"
	onnx_model = Phi3MoE128KModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Phi3SmallForCausalLM" and config.max_position_embeddings == 8192:
	onnx_model = Phi3Small8KModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Phi3SmallForCausalLM" and config.max_position_embeddings == 131072:
	onnx_model = Phi3Small128KModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Phi3VForCausalLM":
	print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
	extra_options["exclude_embeds"] = True
	onnx_model = Phi3VModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	elif config.architectures[0] == "Qwen2ForCausalLM":
	onnx_model = QwenModel(config, io_dtype, precision, execution_provider, cache_dir, extra_options)
	else:
	raise NotImplementedError(f"The {hf_name} model is not currently supported.")

Would it be possible to make this more formal by adding a method called is_model_supported(str model_name) where model_name is a HF model path e.g. microsoft/Phi-3-mini-4k-instruct-onnx?

This API would allow the Olive CLI to test whether a model input is supported by MB for graph capture. If it is Olive will use MB for graph capture, if the model is not supported by MB it will fall back to TorchDynamo.