commaai
/
tinygrad
mirror of https://github.com/commaai/tinygrad.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

run_thneed with test

This commit is contained in:
George Hotz 2022-08-22 09:44:32 -07:00
parent ec5d9b355c
commit 11626053b0
3 changed files with 24 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/test.yml vendored
View File

@ -127,7 +127,7 @@ jobs:
run: |
UNSAFE_FLOAT4=1 DEBUGCL=1 python3 openpilot/compile.py
FLOAT16=1 UNSAFE_FLOAT4=1 DEBUGCL=1 python3 openpilot/compile.py
python3 openpilot/run_thneed.py /tmp/output.thneed
python3 openpilot/run_thneed.py /tmp/output.thneed https://github.com/commaai/openpilot/raw/ea449f1fe0bbff0eff5b12d64f0b5e75b7983998/selfdrive/modeld/models/supercombo.onnx
testmypy:
name: Mypy Tests

4
openpilot/compile.py
View File

@ -30,8 +30,8 @@ OPENPILOT_MODEL = "https://github.com/commaai/openpilot/raw/ea449f1fe0bbff0eff5b
np.random.seed(1337)
def get_random_input_tensors():
np_inputs = {
"input_imgs": np.random.randn(*(1, 12, 128, 256)),
"big_input_imgs": np.random.randn(*(1, 12, 128, 256)),
"input_imgs": np.random.randn(*(1, 12, 128, 256))*256,
"big_input_imgs": np.random.randn(*(1, 12, 128, 256))*256,
"desire": np.zeros((1, 8)),
"traffic_convention": np.array([[1., 0.]]),
"initial_state": np.zeros((1, 512))

49
openpilot/run_thneed.py
View File

@ -56,16 +56,10 @@ def load_thneed_model(fn="model.thneed", float32=False, replace=None):
if 'buffer_id' in o and bufs_loaded[o['buffer_id']]:
arr = np.zeros(bufs[o['buffer_id']].size // 2, dtype=np.float16)
cl.enqueue_copy(q, arr, bufs[o['buffer_id']])
if (o['width'], o['height'], 4) == (108, 8, 4) and replace is not None:
replace = replace.flatten()
arr[:replace.shape[0]] = replace
o['row_pitch'] = 108*4*2
buf = cl.Image(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR,
cl.ImageFormat(cl.channel_order.RGBA, cl.channel_type.HALF_FLOAT),
buf = cl.Image(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR, image_fmt,
shape=(o['width'], o['height']), pitches=(o['row_pitch'],), hostbuf=arr)
elif o['needs_load']:
buf = cl.Image(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR,
cl.ImageFormat(cl.channel_order.RGBA, cl.channel_type.HALF_FLOAT),
buf = cl.Image(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR, image_fmt,
shape=(o['width'], o['height']), pitches=(o['row_pitch'],), hostbuf=o['data'])
else:
buf = cl.Image(ctx, mf.READ_WRITE, image_fmt, shape=(o['width'], o['height']))
@ -119,13 +113,6 @@ def load_thneed_model(fn="model.thneed", float32=False, replace=None):
vision = vision[0:1]
vnum = vnum[0] if len(vnum) >= 1 else None
if float32:
for k in ['image1d_to_buffer_float', 'concatenation_to_buffer_float']:
v = open(BASEDIR+"kernels/%s.cl" % k).read()
prgs[k] = cl.Program(ctx, v).build().__getattr__(k)
for i,k in enumerate(jdat['kernels']):
k['name'] = k['name'].replace("_half", "_float")
def runner(inp=[], policy_only=False, vision_only=False, debug=False):
kernels = []
total_runtime = 0
@ -232,34 +219,40 @@ def load_thneed_model(fn="model.thneed", float32=False, replace=None):
return runner
if __name__ == "__main__":
runner = load_thneed_model("/data/openpilot/selfdrive/modeld/models/supercombo.thneed" if len(sys.argv) == 1 else sys.argv[1])
runner = load_thneed_model("/data/openpilot/selfdrive/modeld/models/supercombo.thneed" if len(sys.argv) == 1 else sys.argv[1], float32=bool(int(os.getenv("FLOAT32", "0"))))
inputs = [
np.zeros((1,512,), dtype=np.float32),
np.zeros((1,2), dtype=np.float32),
np.zeros((1,8), dtype=np.float32),
np.zeros((1,12,128,256), dtype=np.float32),
np.zeros((1,12,128,256), dtype=np.float32)]
np.random.seed(1338)
np_inputs = {
"input_imgs": np.random.randn(*(1, 12, 128, 256))*256,
"big_input_imgs": np.random.randn(*(1, 12, 128, 256))*256,
"desire": np.zeros((1, 8)),
"traffic_convention": np.array([[1., 0.]]),
"initial_state": np.zeros((1, 512))
}
np_inputs = {k:v.astype(np.float32) for k,v in np_inputs.items()}
inputs = list(np_inputs.values())[::-1]
ret = runner(inputs, vision_only=False, debug=True)
print(ret.shape)
if len(sys.argv) > 2:
print("comparing to ONNX")
from xx.torch.lib.reporter import TorchOnnxRunner
mdl = TorchOnnxRunner(sys.argv[2], force_cpu=True)
print(mdl.get_inputs())
out = mdl.run(dict(zip(mdl.get_inputs(), inputs[::-1])))['outputs'].flatten()
print(out.shape)
from test.test_onnx import run_onnx_torch
from extra.utils import fetch
import onnx, io
dat = fetch(sys.argv[2])
onnx_model = onnx.load(io.BytesIO(dat))
out = run_onnx_torch(onnx_model, np_inputs).numpy()[0]
diff = 0
for i in range(ret.shape[0]):
if out[i]-ret[i] > 0.1:
if out[i]-ret[i] > 0.1 and (out[i]-ret[i])/out[i] > 0.01:
print(i, out[i], ret[i], out[i]-ret[i])
diff += 1
if diff == 10:
print("...")
break
assert diff == 0
"""
for i in range(0, len(ret), 0x10):