Fix test_llm_config (pytorch#11977)

Summary: Pull Request resolved: pytorch#11977

Test Plan:
Imported from GitHub, without a `Test Plan:` line.

Rollback Plan:

Reviewed By: GregoryComer

Differential Revision: D77321057

Pulled By: jackzhxng

Author: jackzhxng

examples/models/llama/config/test_llm_config.py

@@ -41,7 +41,7 @@ def test_local_global_attention_without_kv(self):
 
     def test_invalid_export_config_context_length(self):
         with self.assertRaises(ValueError):
-            ExportConfig(max_seq_length=128, max_context_length=256)
+            ExportConfig(max_seq_length=256, max_context_length=128)
 
     def test_invalid_qmode(self):
         with self.assertRaises(ValueError):
@@ -84,8 +84,8 @@ def test_valid_llm_config(self):
                 local_global_attention="[16, 32]",
             ),
             export=ExportConfig(
-                max_seq_length=256,
-                max_context_length=128,
+                max_seq_length=128,
+                max_context_length=256,
                 output_dir="/tmp/export",
                 output_name="model.pte",
             ),
@@ -94,7 +94,7 @@ def test_valid_llm_config(self):
             backend=BackendConfig(
                 xnnpack=XNNPackConfig(enabled=False),
                 coreml=CoreMLConfig(
-                    enabled=True, ios=17, compute_units=CoreMLComputeUnit.ALL
+                    enabled=True, ios=17, compute_units=CoreMLComputeUnit.cpu_only
                 ),
             ),
         )

extension/llm/export/test/test_export_llm.py

@@ -67,7 +67,7 @@ def test_with_config(self, mock_export_llama: MagicMock) -> None:
 model:
   dtype_override: fp16
 export:
-  max_seq_length: 256
+  max_seq_length: 128
 quantization:
   pt2e_quantize: xnnpack_dynamic
   use_spin_quant: cuda

pytest.ini

@@ -18,6 +18,7 @@ addopts =
     --ignore=devtools/visualization/visualization_utils_test.py
     # examples
     examples/models/llama/tests
+    examples/models/llama/config
     examples/models/llama3_2_vision/preprocess
     examples/models/llama3_2_vision/vision_encoder/test
     examples/models/llama3_2_vision/text_decoder/test