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>

config_sample.yml

@@ -109,6 +109,12 @@ model:
   # Only use this if the model's base sequence length in config.json is incorrect (ex. Mistral 7B)
   #override_base_seq_len:
 
+  # Load model with tensor parallelism
+  # If a GPU split isn't provided, the TP loader will fallback to autosplit
+  # Enabling ignores the gpu_split_auto and autosplit_reserve values
+  # NOTE: Requires a development build of exllamav2
+  #tensor_parallel: False
+
   # Automatically allocate resources to GPUs (default: True)
   # NOTE: Not parsed for single GPU users
   #gpu_split_auto: True
@@ -118,6 +124,7 @@ model:
   #autosplit_reserve: [96]
 
   # An integer array of GBs of vram to split between GPUs (default: [])
+  # Used with tensor parallelism
   # NOTE: Not parsed for single GPU users
   #gpu_split: [20.6, 24]