Improve pte loading when given manual params_path (#1178)

Author: Jack-Khuu

torchchat/cli/builder.py

@@ -29,7 +29,7 @@
 
 from torchtune.training import set_default_dtype
 
-from torchchat.model import Model, ModelType
+from torchchat.model import Model, ModelArgs, ModelType
 
 from torchchat.model_config.model_config import resolve_model_config
 from torchchat.utils.build_utils import (
@@ -527,18 +527,22 @@ def _initialize_model(
             )
             builder_args.device = "cpu"
 
-        # assert (
-        #     quantize is None or quantize == "{ }"
-        # ), "quantize not valid for exported PTE model. Specify quantization during export."
-
-        with measure_time("Time to load model: {time:.02f} seconds"):
-            model = _load_model(builder_args)
-            device_sync(device=builder_args.device)
+        # Resolve ModelArgs for constructing the PTEModel
+        # If a manual params_path is provided, use that
+        if builder_args.params_path:
+            config: ModelArgs = ModelArgs.from_params(builder_args.params_path)
+        else:
+            # TODO: Instead of loading the whole model, refactor to call a
+            # helper that generate just model.config
+            with measure_time("Time to load model: {time:.02f} seconds"):
+                model = _load_model(builder_args)
+                device_sync(device=builder_args.device)
+                config = model.config
 
         try:
             from torchchat.model import PTEModel
 
-            model = PTEModel(model.config, builder_args.pte_path)
+            model = PTEModel(config, builder_args.pte_path)
         except Exception:
             raise RuntimeError(f"Failed to load ET compiled {builder_args.pte_path}")
     else: