update eager runner to use same options for loading the model (pytorch#6257)

Summary:
Pull Request resolved: pytorch#6257

imported-using-ghimport

Test Plan:
Imported from OSS

Run the following command and make sure it generate the right result:
```
python -m examples.models.llama2.runner.eager \
          -c /home/lunwenh/models/1B_Instruct/consolidated.00.pth \
          -p /home/lunwenh/models/1B_Instruct/params.json \
          -t /home/lunwenh/models/1B_Instruct/tokenizer.model \
          --max_seq_length 128 \
          -kv \
          --prompt "<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are a good assistant<|eot_id|><|start_header_id|>user<|end_header_id|>

What is the capital of France?<|eot_id|><|start_header_id|>assistant<|end_header_id|>"
```
```
Response:
The capital of France is Paris.<|eot_id|>
 Tokens:
 [791, 6864, 315, 9822, 374, 12366, 16134, 91, 68, 354, 851, 91, 29]
```

Reviewed By: mergennachin

Differential Revision: D64442224

Pulled By: helunwencser

fbshipit-source-id: bb8b11de6325ae76423b086491094a4444249553

Author: helunwencser

examples/models/llama2/runner/eager.py

@@ -11,9 +11,12 @@
 import torch
 
 from examples.models.llama2.llama_transformer import ModelArgs
-from executorch.examples.models.model_factory import EagerModelFactory
-
-from .generation import LlamaRunner
+from executorch.examples.models.llama2.export_llama_lib import (
+    _prepare_for_llama_export,
+    build_args_parser as _build_args_parser,
+)
+from executorch.examples.models.llama2.runner.generation import LlamaRunner
+from executorch.extension.llm.export import LLMEdgeManager
 
 
 class EagerLlamaRunner(LlamaRunner):
@@ -25,21 +28,17 @@ def __init__(self, args):
         with open(args.params, "r") as f:
             params = json.loads(f.read())
         model_args: ModelArgs = ModelArgs(
-            max_seq_len=args.max_len,
+            max_seq_len=args.max_seq_length,
             max_batch_size=1,
-            use_kv_cache=True,
+            use_kv_cache=args.use_kv_cache,
             **params,
         )
-        super().__init__(tokenizer_path=args.tokenizer, model_args=model_args)
-        self.model, _, _, _ = EagerModelFactory.create_model(
-            "llama2",
-            "Llama2Model",
-            checkpoint=args.checkpoint,
-            params=args.params,
-            use_kv_cache=True,
-            fairseq2=False,
-            max_seq_len=args.max_len,
-            enable_dynamic_shape=True,
+        super().__init__(tokenizer_path=args.tokenizer_path, model_args=model_args)
+        manager: LLMEdgeManager = _prepare_for_llama_export("llama", args)
+        self.model = (
+            manager.model.eval().to(device="cuda")
+            if torch.cuda.is_available()
+            else manager.model.eval().to(device="cpu")
         )
 
     def forward(
@@ -51,34 +50,7 @@ def forward(
 
 
 def build_args_parser() -> argparse.ArgumentParser:
-    parser = argparse.ArgumentParser()
-
-    parser.add_argument(
-        "--checkpoint",
-        type=str,
-        default=None,
-        help="path to model checkpoint file",
-    )
-
-    parser.add_argument(
-        "--params",
-        type=str,
-        default=None,
-        help="model params file",
-    )
-
-    parser.add_argument(
-        "--max_len",
-        type=int,
-        default=128,
-        help="Maximum length of the generated response sequence.",
-    )
-
-    parser.add_argument(
-        "--tokenizer",
-        type=str,
-        default=None,
-    )
+    parser = _build_args_parser()
 
     parser.add_argument(
         "--prompt",