ctransformers: add mlock and no-mmap options (#3649)

modules/ctransformers_model.py

@@ -19,7 +19,9 @@ class CtransformersModel:
             gpu_layers=shared.args.n_gpu_layers,
             batch_size=shared.args.n_batch,
             context_length=shared.args.n_ctx,
-            stream=True
+            stream=True,
+            mmap=not shared.args.no_mmap,
+            mlock=shared.args.mlock
         )
 
         self.model = AutoModelForCausalLM.from_pretrained(

modules/loaders.py

@@ -101,7 +101,9 @@ loaders_and_params = OrderedDict({
         'n_gpu_layers',
         'n_batch',
         'threads',
-        'model_type'
+        'model_type',
+        'no_mmap',
+        'mlock'
     ]
 })