You are viewing a plain text version of this content. The canonical link for it is here.
Posted to discuss-archive@tvm.apache.org by Ryan Chang via TVM Discuss <no...@discuss.tvm.ai> on 2020/08/25 09:20:42 UTC
[TVM Discuss] [Questions] Why can't I save the file
Tvm version
commit f6e00a0d8b384e8cd81b90668c06f68cd52161b9
Author: Iswariya Manivannan <36...@users.noreply.github.com>
Date: Mon Aug 17 05:53:38 2020 +0200
This is my code
{
...
mod, params = relay.frontend.from_tflite(tflite_model,
shape_dict={input_tensor: input_shape},
dtype_dict={input_tensor: input_dtype})
print(mod)
with relay.quantize.qconfig(nbit_activation=32,
dtype_input="int8",
dtype_weight="int8",
dtype_activation="int32",
skip_conv_layers=[0]
):
print(" ", relay.quantize.current_qconfig())
mod = relay.quantize.quantize(mod, params)
...
# compile the model on Relay
with relay.build_config(opt_level = 3):
graph, lib, params = relay.build_module.build(mod, target, params=params)
# model save
lib.save("test.ll",fmt="ll")
with open('test.json', 'w') as _f:
_f.write(graph)
with open('test.params', 'wb') as _f:
_f.write(relay.save_param_dict(params))
}
This code is ok to save (test.ll, test.json, test. params) but the following will be wrong.
{
...
mod, params = relay.frontend.from_tflite(tflite_model,
shape_dict={input_tensor: input_shape},
dtype_dict={input_tensor: input_dtype})
print(mod)
with relay.quantize.qconfig(nbit_activation=32,
dtype_input="int8",
dtype_weight="int8",
dtype_activation="int32",
skip_conv_layers=[0]
):
print(" ", relay.quantize.current_qconfig())
mod = relay.quantize.quantize(mod, params)
pattern_table = [
("dnnl.conv2d_add_right_shift_clip_cast", make_pattern_())
]
mod["main"] = run_opt_pass(mod["main"], relay.transform.MergeComposite(pattern_table))
mod = composite_partition(mod)
mod = transform.AnnotateTarget(["dnnl"])(mod)
mod = transform.MergeCompilerRegions()(mod)
...
# compile the model on Relay
with relay.build_config(opt_level = 3):
graph, lib, params = relay.build_module.build(mod, target, params=params)
# model save
lib.save("test.ll",fmt="ll")
with open('test.json', 'w') as _f:
_f.write(graph)
with open('test.params', 'wb') as _f:
_f.write(relay.save_param_dict(params))
}
error message
test_pass_partition_graph.py:1469: DeprecationWarning: legacy graph runtime behaviour of producing json / lib / params will be removed in the next release
graph, lib, params = relay.build_module.build(mod, target, params=params)
Traceback (most recent call last):
File "test_pass_partition_graph.py", line 1630, in <module>
test_extern_dnnl_LeNet()
File "test_pass_partition_graph.py", line 1475, in test_extern_dnnl_LeNet
lib.save("test.ll",fmt="ll")
File "/home/ymchang/tvm-master/python/tvm/runtime/module.py", line 164, in save
_ffi_api.ModuleSaveToFile(self, file_name, fmt)
File "/home/ymchang/tvm-master/python/tvm/_ffi/_ctypes/packed_func.py", line 225, in __call__
raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
[bt] (3) /home/ymchang/tvm-master/build/libtvm.so(TVMFuncCall+0x61) [0x7f7c64852631]
[bt] (2) /home/ymchang/tvm-master/build/libtvm.so(+0x14ed260) [0x7f7c64871260]
[bt] (1) /home/ymchang/tvm-master/build/libtvm.so(tvm::runtime::ModuleNode::SaveToFile(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)+0x8d) [0x7f7c6486d31d]
[bt] (0) /home/ymchang/tvm-master/build/libtvm.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x61) [0x7f7c63c6d7c1]
File "/home/ymchang/tvm-master/src/runtime/module.cc", line 90
TVMError: Module[metadata] does not support SaveToFile
---
[Visit Topic](https://discuss.tvm.ai/t/why-cant-i-save-the-file/7718/1) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/527662b00273d3469ee2027558ab5a83f6a89161744bec295de06ca67d30ca01).
[TVM Discuss] [Questions] Why can't I save the file
Posted by Zhi via TVM Discuss <no...@discuss.tvm.ai>.
Yeah, this is due to missing SaveToFile in MetadataModule. It is inconvenient to save them to a file though because there would be multiple modules (i.e. LLVMModule and some subgraph modules).
---
[Visit Topic](https://discuss.tvm.ai/t/why-cant-i-save-the-file/7718/3) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/e448151f7bc7c9cdb6c69b8ebe8e5c89670b06eafe8f78df3c94082a3d2065a9).
[TVM Discuss] [Questions] Why can't I save the file
Posted by "Cody H. Yu via TVM Discuss" <no...@discuss.tvm.ai>.
@zhiics it looks like a problem of missing SavetoFile implementation.
---
[Visit Topic](https://discuss.tvm.ai/t/why-cant-i-save-the-file/7718/2) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/751725e1b89c69205571886ad1f5983c0bb14811b083d08d53dc49e76c25b8e1).