github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

fix onnx mobilenetv2-7-quantized.onnx (#8574)

Browse Source 
* is 67% considered fixed?

* move test up

* share function

* add qgemm too

* make sure qgemm comes out as int

* actually that note is not right

* remove qgemm (I did it wrong) and add it later lol.
This commit is contained in:
geohotstan
2025-01-14 01:25:06 +08:00
committed by GitHub
parent d19c1c7f03
commit 4abe631b56
3 changed files with 18 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
test/external/external_test_onnx_backend.py vendored
View File

@@ -44,6 +44,12 @@ backend_test = onnx.backend.test.BackendTest(TinygradBackend, __name__)
backend_test.exclude('test_adam_multiple_cpu')
backend_test.exclude('test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_True_cpu')
# BUG: onnxruntime 1.20.1 fails these tests too
backend_test.exclude('test_qlinearmatmul_2D_int8_float16_cpu')
backend_test.exclude('test_qlinearmatmul_3D_int8_float16_cpu')
backend_test.exclude('test_qlinearmatmul_2D_int8_float32_cpu')
backend_test.exclude('test_qlinearmatmul_3D_int8_float32_cpu')
# about different dtypes
if not is_dtype_supported(dtypes.float64):
backend_test.exclude('float64')
@@ -86,9 +92,6 @@ backend_test.exclude('test_dequantizelinear_e4m3fn_cpu')
backend_test.exclude('test_dequantizelinear_e4m3fn_zero_point_cpu')
backend_test.exclude('test_dequantizelinear_e5m2_cpu')
# TODO: unsure. The number is off by 1. and it's not because of float16
backend_test.exclude('test_qlinearmatmul_3D_int8_float16_cpu')
# we don't support indexes
backend_test.exclude('test_nonzero_*')