Move profiler -> prototype (#1370)

Author: jainapurva

test/profiler/test_device_spec.py renamed to test/prototype/test_device_spec.py

@@ -8,7 +8,7 @@
 import torch
 from utils import patch_device
 
-from torchao.profiler.device_spec import (
+from torchao.prototype.profiler.device_spec import (
     _AVAILABLE_GPU_SPECS,
     CUDADeviceSpec,
     get_chip_name,

test/profiler/test_performance_counter.py renamed to test/prototype/test_performance_counter.py

@@ -30,8 +30,8 @@
     qkv_proj_io_check,
 )
 
-from torchao.profiler.device_spec import CUDADeviceSpec, DeviceSpec
-from torchao.profiler.performance_counter import (
+from torchao.prototype.profiler.device_spec import CUDADeviceSpec, DeviceSpec
+from torchao.prototype.profiler.performance_counter import (
     CUDAPerformanceTimer,
     PerformanceCounterMode,
     PerformanceStats,

test/profiler/utils.py renamed to test/prototype/utils.py

@@ -5,7 +5,7 @@
 
 import torch
 
-from torchao.profiler import PerformanceTimer
+from torchao.prototype.profiler import PerformanceTimer
 
 
 @contextmanager

torchao/_models/llama/perf_profile.py

@@ -2,9 +2,9 @@
 
 ## Performance Profiling Example
 
-An minimal version of `gpt-fast generate.py` that demonstrates usage of `torchao.profiler.TransformerPerformanceCounter`.
+An minimal version of `gpt-fast generate.py` that demonstrates usage of `torchao.prototype.profiler.TransformerPerformanceCounter`.
 - Outputs from gpt-fast are prefixed with GPT-Fast
-- Outputs from `torchao.profiler.TransformerPerformanceCounter` are prefixed with `TransformerPerfCounter`.
+- Outputs from `torchao.prototype.profiler.TransformerPerformanceCounter` are prefixed with `TransformerPerfCounter`.
 
 ## Usage
 ```python
@@ -118,7 +118,7 @@
 
 from torchao._models.llama.model import Transformer
 from torchao._models.llama.tokenizer import get_tokenizer
-from torchao.profiler import (
+from torchao.prototype.profiler import (
     CUDADeviceSpec,
     TransformerPerformanceCounter,
     total_model_params,