Update on "Use llm_config instead of args in export_llama functions"

Differential Revision: [D75484927](https://our.internmc.facebook.com/intern/diff/D75484927)

[ghstack-poisoned]

Author: jackzhxng

examples/models/llama/config/llm_config.py

@@ -162,11 +162,11 @@ class QuantizationConfig:
     pt2e_quantize: Optional[Pt2eQuantize] = None
     group_size: Optional[int] = None
     use_spin_quant: Optional[SpinQuant] = None
-    use_qat: Optional[bool] = None
+    use_qat: bool = False
     calibration_tasks: Optional[List[str]] = None
     calibration_limit: Optional[int] = None
     calibration_seq_length: Optional[int] = None
-    calibration_data: Optional[str] = None
+    calibration_data: str = "Once upon a time"
 
     def __post_init__(self):
         if self.qmode:
@@ -243,7 +243,7 @@ class QNNConfig:
 
 @dataclass
 class MPSConfig:
-    enabled: Optional[bool] = False
+    enabled: bool = False
 
 
 @dataclass

examples/models/llama/tests/test_export_llama_lib.py

@@ -8,10 +8,7 @@
 
 from executorch.devtools.backend_debug import get_delegation_info
 from executorch.examples.models.llama.config.llm_config import LlmConfig
-from executorch.examples.models.llama.export_llama_lib import (
-    _export_llama,
-    build_args_parser,
-)
+from executorch.examples.models.llama.export_llama_lib import _export_llama
 
 UNWANTED_OPS = [
     "aten_permute_copy_default",
@@ -41,13 +38,6 @@ def test_has_expected_ops_and_op_counts(self):
         llm_config.model.use_kv_cache = True
         llm_config.debug.verbose = True
 
-        # We still need args for backward compatibility during transition
-        parser = build_args_parser()
-        args = parser.parse_args([])
-        args.use_sdpa_with_kv_cache = True
-        args.use_kv_cache = True
-        args.verbose = True
-
         builder = _export_llama(llm_config)
         graph_module = builder.edge_manager.exported_program().graph_module
         delegation_info = get_delegation_info(graph_module)