wpt_results.md

WebNN WPT results

Remote passes means results from https://rustnn.github.io/rustnnpt/, which runs rustnn using pywebnn.

(local) Passes are for results of ./mach test-wpt webnn using https://github.com/gterzian/servo/tree/webnn on Mac OS using CoreML (the only supported backend as of now).

Test	Total	Passes	Remote passes
abs	57	21
arg_min_max	108	54	60
averagePool2d	120	24	38
batch_normalization	69	24	24
batch_normalization_constant	9	6	2
byob_readtensor	12		0
cast	90	24	49
ceil	48	21	14
clamp	144	51	51
concat	141	60	45
constant-reshape-optimization	9	6
conv2d	126	48	40
conv_transpose2d	132	51	42
cos	48	21	14
cumulative_sum	27	15	7
dequantizeLinear	33	6	32
div	66	33	20
elu	66	27	20
equal	72	33	24
erf	48	21	14
exp	48	21	14
expand	138	54	44
floor	48	21	14
gather	120	54	42
gatherElements	33	15	11
gatherND	42	6	16
gelu	45	6	13
gemm	144	72	46
graph_devices	6	3	0
greater	72	30	24
greater_or_equal	72	33	24
gru	78	6	0
gru_cell	30	6	0
hard_sigmoid	96	48	30
hard_swish	48	21	14
identity	102	12	14
inputs-are-not-modified	24	18
inputs-with-special-names	30	9
instance_normalization	48	15	14
is_infinite	51	6	15
is_nan	42	6	12
l2Pool2d	48	6	0
layer_normalization	81	33	25
leaky_relu	66	30	20
lesser	72	30	24
lesser_or_equal	72	30	24
linear	84	39	26
log	48	21	14
logical_and	39	30	12
logical_not	27	24	7
logical_or	39	30	12
logical_xor	39	30	12
lstm	90	6	0
lstm_cell	36	6	0
matmul	72	42	22
max	66	6	20
maxPool2d	84	24	26
min	66	6	20
mlNumber	12	6	10
mul	66	33	20
neg	57	21	19
not_equal	72	30	24
operations-with-special-names	18	18
pad	75	6	25
parallel-dispatch	66	36
pow	60	24	20
prelu	78	9	27
qdq_subgraph	228	12	41
quantizeLinear	30	6	30
reciprocal	48	21	14
reduce_l1	138	54	45
reduce_l2	135	54	43
reduce_log_sum	123	48	39
reduce_log_sum_exp	135	54	43
reduce_max	117	45	37
reduce_mean	135	54	43
reduce_min	117	45	37
reduce_product	117	45	37
reduce_sum	135	54	43
reduce_sum_square	138	54	44
relu	48	6	15
resample2d	45	6	13
reshape	198	93	66
reverse	30	18	8
round_even	36	18	10
scalars	48	6	0
scatterElements	30	6	8
scatterND	18	6	5
shared_arraybuffer_constant	15	6	0
sigmoid	48	21	14
sign	18	15	7
sin	48	21	14
slice	60	24	19
softmax	33	6	9
softplus	48	21	14
softsign	60	24	18
split	3		20
sqrt	48	21	14
sub	72	33	26
subgraph	300	132	48
tan	48	21	14
tanh	42	21	12
tensor	12		20
tile	24	9	7
transpose	60	36	18
triangular	102	39	32
where	108	6	34

Total conformance tests: 7473 \
Total local passes: 2745
Total remote passes: 2213

Validation tests

Test	Total	Passes
argMinMax	54	30
batchNormalization	72	72
build-more-than-once	66	60
cast	12	9
clamp	33	18
concat	42	21
constant-changed-buffer	36	24
constant	276	126
conv2d	3
convTranspose2d	3
createContext	72	60
cumulativeSum	30	24
dequantizeLinear	51	27
destroyContext	78	54
destroyGraph	30	24
elementwise-binary	252	147
elementwise-logical	189	135
elementwise-unary	141	51
elu	24	18
expand	39	24
gather	39	24
gatherElements	33	18
gatherND	30	15
gelu	15	9
gemm	63	30
gru	78	24
gruCell	93	24
hardSigmoid	24	18
hardSwish	15	9
input	66	54
instanceNormalization	60	27
invalid-rank	24	18
layerNormalization	66	36
leakyRelu	24	18
linear	24	18
lstm	78	30
lstmCell	117	30
matmul	48	27
opSupportLimits	18	6
pad	39	15
pooling-and-reduction-keep-dims	30	24
pooling	126	69
prelu	33	12
quantizeLinear	48	27
reduction	186	102
relu	15	6
resample2d	87	36
reshape	36	24
reverse	21	15
scatterElements	48	24
scatterND	33	15
sigmoid	15	9
slice	39	15
softmax	24	9
softplus	15	9
softsign	15	9
split	36	15
tanh	15	9
tile	3
transpose	27	18
triangular	12	9
unprintableNames	18	6
where	42	21

Total validation tests: 3381 \
Total validation passes: 1857