follow comments

thunder/benchmarks/benchmark_litgpt.py

@@ -241,7 +241,6 @@ def __init__(
         use_torchao_fp8_allgather: bool = False,
         use_torchao_fp8_precompute_scale_for_fsdp: bool = False,
         fp8_shard_intermediate_activation: bool = False,
-        save_dynamo_repro: str | None = None,
     ):
         seed = 1337
         torch.manual_seed(seed)
@@ -276,11 +275,6 @@ def __init__(
         self.dump_thunder_traces = dump_thunder_traces
         self.dump_memory_snapshot = dump_memory_snapshot
         self.fp8_shard_intermediate_activation = fp8_shard_intermediate_activation
-        if save_dynamo_repro is not None:
-            assert (
-                "dynamo" in self.compile and "thunder" in self.compile
-            ), "save_dynamo_repro can only be used if --compile=thunder+dynamo"
-        self.save_dynamo_repro = save_dynamo_repro
 
         if use_torchao_fp8_linear:
 
@@ -898,9 +892,6 @@ def benchmark_main(return_metrics_as_json=False, json_path="", **kwargs) -> None
                     print(f"##########\n#{i}-th ThunderModule\n##########")
                     print(b_traces[-1])
 
-        if benchmark.save_dynamo_repro:
-            benchmark.backend.save_reproducer_to_folder(benchmark.save_dynamo_repro)
-
     if global_rank in [0, None]:
         if return_metrics_as_json:
             benchmark.add_model_info_to_metrics()

thunder/dynamo/utils.py

@@ -80,7 +80,7 @@ class SplitReason:
     exception: Exception | None = None
 
 
-@dataclasses.dataclass
+@dataclasses.dataclass(frozen=True)
 class ExampleInputMetaData:
     """
     Describes the metadata of a tensor, used to generate a random tensor with matching properties
@@ -464,6 +464,12 @@ def _get_storage_shape(t: torch.Tensor):
 
 
 def _get_example_input_tensor_metadata(t: torch.Tensor) -> ExampleInputMetaData:
+    min_val = None
+    max_val = None
+    if not isinstance(t, FakeTensor) and t.numel() != 0:
+        minmax: tuple[torch.Tensor, torch.Tensor] = torch.aminmax(t)
+        min_val = minmax[0].cpu().item()
+        max_val = minmax[1].cpu().item()
     meta_ev = ExampleInputMetaData(
         t.requires_grad,
         t.layout,
@@ -472,11 +478,9 @@ def _get_example_input_tensor_metadata(t: torch.Tensor) -> ExampleInputMetaData:
         _concrete_value(t.shape),
         _get_storage_shape(t),
         _concrete_value(t.stride()),
+        min_val,
+        max_val,
     )
-    if not isinstance(t, FakeTensor) and t.numel() != 0:
-        minmax: tuple[torch.Tensor, torch.Tensor] = torch.aminmax(t)
-        meta_ev.min_val = minmax[0].cpu().item()
-        meta_ev.max_val = minmax[1].cpu().item()
     return meta_ev
 
 
@@ -768,8 +772,8 @@ def reproducer(
         code_str += f"{_addindent(input_str, 4)}\n]\n"
 
         if not use_pytest_benchmark:
-            code_str += f"fqn = thunder.jit(DynamoModule(), {thunder_options_str})\n"
-            code_str += "fqn(*inputs)"
+            code_str += f"compiled = thunder.jit(DynamoModule(), {thunder_options_str})\n"
+            code_str += "compiled(*inputs)"
         else:
             code_str += "from thunder.dynamo.compiler_graph_benchmark import ThunderCompilerGraphBenchmarking\n"
             code_str = f"""{code_str}