WeberChen opened a new issue #7941:
URL: https://github.com/apache/tvm/issues/7941
Hi all,
I have run the test example in test_pass_partition_graph.py.
It's OK to export the graph_jason, lib.so and the parameters.
However, when I try to insert the nn.maxpool2d operator, the information of
pool_size and pool_stride is missing in graph_json file.
Could someone tell me how to get the above information in graph_json file?
What files should I modify?
Thank you.
---
The following code is the inserted nn.maxpool2d in test_dnnl_fuse() in
test_pass_partition_graph.py.
def get_net(include_bn=True, include_sigmoid=False):
data = relay.var("data", relay.TensorType((1, 3, 224, 224), "float32"))
block1 = get_blocks("block1_", data, 3, 8, include_bn, include_sigmoid)
block2 = get_blocks("block2_", block1, 8, 8, False, include_sigmoid)
block3 = relay.nn.avg_pool2d(block2, pool_size=(2, 2), strides=(2, 2),
padding=(0, 0), layout='NCHW', ceil_mode=False, count_include_pad=False)
return relay.Function(relay.analysis.free_vars(block3), block3)
---
The following code is the part of graph_json about the maxpool2d:
{
"op": "tvm_op",
"name": "fused_nn_avg_pool2d_1",
"attrs": {
"num_outputs": "1",
"num_inputs": "1",
"func_name": "fused_nn_avg_pool2d_1",
"flatten_data": "0"
},
"inputs": [
[
2,
0,
0
]
]
}
---
AND the picture is the graph_json and it can get the correct result.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
