mawei/sunnypilot
1
0
Fork 0
mirror of https://github.com/sunnypilot/sunnypilot.git synced 2026-07-23 16:32:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix inconsistent indentation in numpy_inputs initialization

Browse Source 
Corrected the indentation within the loop for numpy_inputs assignment in `modeld.py`. This ensures proper execution of the loop and prevents potential runtime issues caused by misaligned code blocks.

Refactor TINYGRAD usage logic and simplify checks.

Consolidated the TINYGRAD flag with TICI conditions to reduce redundancy. Adjusted tensor initialization flow to handle different device setups more cleanly. This simplifies the code and improves maintainability.

Enable tinygrad integration via environment variable

Added support for using tinygrad on non-TICI devices by introducing the `USE_TINYGRAD` environment variable. Conditional logic was updated to accommodate this change, ensuring compatibility with both tinygrad and ONNX runtimes. This allows more flexibility in choosing the computation framework.
This commit is contained in:
DevTekVE
2024-12-28 19:39:48 +01:00
parent d9c22271d6
commit 341b92176e
1 changed files with 13 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
selfdrive/modeld/modeld.py
+13 -8
View File
@@ -3,7 +3,8 @@ import os
from openpilot.system.hardware import TICI
#
if TICI:
USE_TINYGRAD = os.getenv('USE_TINYGRAD', True) or TICI
if USE_TINYGRAD:
from tinygrad.tensor import Tensor
from tinygrad.dtype import dtypes
from openpilot.selfdrive.modeld.runners.tinygrad_helpers import qcom_tensor_from_opencl_address
@@ -60,7 +61,7 @@ class ModelState:
self.frames = {'input_imgs': DrivingModelFrame(context), 'big_input_imgs': DrivingModelFrame(context)}
self.prev_desire = np.zeros(ModelConstants.DESIRE_LEN, dtype=np.float32)
self.full_features_20Hz = np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN, ModelConstants.FEATURE_LEN), dtype=np.float32)
self.desire_20Hz = np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN + 1, ModelConstants.DESIRE_LEN), dtype=np.float32)
self.desire_20Hz = np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN + 1, ModelConstants.DESIRE_LEN), dtype=np.float32)
# img buffers are managed in openCL transform code
self.numpy_inputs = {}
@@ -78,7 +79,7 @@ class ModelState:
self.output = np.zeros(net_output_size, dtype=np.float32)
self.parser = Parser()
if TICI:
if USE_TINYGRAD:
self.tensor_inputs = {k: Tensor(v, device='NPY').realize() for k,v in self.numpy_inputs.items()}
with open(MODEL_PKL_PATH, "rb") as f:
self.model_run = pickle.load(f)
@@ -112,13 +113,17 @@ class ModelState:
imgs_cl = {'input_imgs': self.frames['input_imgs'].prepare(buf, transform.flatten()),
'big_input_imgs': self.frames['big_input_imgs'].prepare(wbuf, transform_wide.flatten())}
if TICI:
if USE_TINYGRAD:
# The imgs tensors are backed by opencl memory, only need init once
for key in imgs_cl:
if key not in self.tensor_inputs:
if not TICI or key not in self.tensor_inputs:
index = self.model_run.captured.expected_names.index(key)
_, _, dtype, _ = self.model_run.captured.expected_st_vars_dtype_device[index]
self.tensor_inputs[key] = qcom_tensor_from_opencl_address(imgs_cl[key].mem_address, self.input_shapes[key], dtype=dtype)
_, _, dtype, device = self.model_run.captured.expected_st_vars_dtype_device[index]
if TICI:
self.tensor_inputs[key] = qcom_tensor_from_opencl_address(imgs_cl[key].mem_address, self.input_shapes[key], dtype=dtype)
else:
shape = self.frames[key].buffer_from_cl(imgs_cl[key]).reshape(self.input_shapes[key])
self.tensor_inputs[key] = Tensor(shape, device=device, dtype=dtype).realize()
else:
for key in imgs_cl:
dtype = self.onnx_model_metadata[key]
@@ -127,7 +132,7 @@ class ModelState:
if prepare_only:
return None
if TICI:
if USE_TINYGRAD:
self.output = self.model_run(**self.tensor_inputs).numpy().flatten()
else:
self.output = self.onnx_cpu_runner.run(None, self.numpy_inputs)[0].flatten()