Allow sending directly torch/tf/np dtype for generating dummy inputs #38

Job	Run time
build (3.10, macos-15, test_modeling_gemma2.py)	59m 31s
build (3.10, macos-15, test_modeling_gemma.py)	52m 35s
build (3.10, macos-15, test_modeling_llama.py)	19m 27s
build (3.10, macos-15, test_modeling_olmo.py)	10m 9s
build (3.10, macos-15, test_modeling.py)	8m 10s
build (3.10, macos-15, test_modeling_qwen2.py)	4m 36s
build (3.11, macos-15, test_modeling_gemma2.py)	1h 0m 38s
build (3.11, macos-15, test_modeling_gemma.py)	48m 33s
build (3.11, macos-15, test_modeling_llama.py)	15m 19s
build (3.11, macos-15, test_modeling_olmo.py)	10m 16s
build (3.11, macos-15, test_modeling.py)	9m 23s
build (3.11, macos-15, test_modeling_qwen2.py)	3m 34s
build (3.12, macos-15, test_modeling_gemma2.py)	1h 3m 34s
build (3.12, macos-15, test_modeling_gemma.py)	51m 43s
build (3.12, macos-15, test_modeling_llama.py)	15m 17s
build (3.12, macos-15, test_modeling_olmo.py)	7m 51s
build (3.12, macos-15, test_modeling.py)	10m 16s
build (3.12, macos-15, test_modeling_qwen2.py)	4m 14s
	7h 35m 6s

Provide feedback