Fix lm_eval import path bugs (#10251)

Summary:
Pull Request resolved: #10251

There were bugs in `executorch`, which did not update `InputRecorder` import from `torchao` given changes in the path, which also required adding dependency to import `lm_eval`.

Reviewed By: metascroy, jackzhxng

Differential Revision: D73166222

Author: Chia-Jung Chang

examples/models/llama/source_transformation/quantize.py

@@ -164,7 +164,7 @@ def quantize(  # noqa C901
 
         try:
             # torchao 0.3+
-            from torchao._eval import InputRecorder  # pyre-fixme[21]
+            from torchao._models._eval import InputRecorder
         except ImportError:
             from torchao.quantization.GPTQ import InputRecorder  # pyre-ignore