sdkv2/falcon-h1-1.5b-mlx-4bit

Remove included inference code and benchmark files

Use mlx_lm runtime and remove custom implementation references

Remove custom implementation; use mlx_lm runtime

Compare custom inference with mlx_lm benchmarks

Remove unrelated CUDA benchmark comparison

Add PyTorch speed comparison to benchmarks

Add inference code and M4 benchmarks

Upload Falcon-H1 1.5B MLX 4-bit

initial commit