Model: Add Tensor Parallel support

Use the tensor parallel loader when the flag is enabled. The new loader
has its own autosplit implementation, so gpu_split_auto isn't valid
here.

Also make it easier to determine which cache type to use rather than
multiple if/else statements.

Signed-off-by: kingbri <bdashore3@proton.me>

common/args.py

@@ -107,6 +107,11 @@ def add_model_args(parser: argparse.ArgumentParser):
         type=str_to_bool,
         help="Overrides base model context length",
     )
+    model_group.add_argument(
+        "--tensor-parallel",
+        type=str_to_bool,
+        help="Use tensor parallelism to load models",
+    )
     model_group.add_argument(
         "--gpu-split-auto",
         type=str_to_bool,