Relay from_tensorflow failed - Couldn't find ANTLR parser

Questions
#1

I tried to use Tensorflow tutorial with the latest TVM built from source with LLVM and tensorflow-1.15

I got the following error

Exception: Couldn't find ANTLR parser. Try building with USE_ANTLR=ON.

Full output:

root@164895bae70c:~# python3 from_tensorflow.py 
File /root/.tvm_test_data/data/elephant-299.jpg exists, skip.
File /root/.tvm_test_data/tf/InceptionV1/classify_image_graph_def-with_shapes.pb exists, skip.
File /root/.tvm_test_data/data/imagenet_2012_challenge_label_map_proto.pbtxt exists, skip.
File /root/.tvm_test_data/data/imagenet_synset_to_human_label_map.txt exists, skip.
WARNING:tensorflow:From from_tensorflow.py:92: FastGFile.__init__ (from tensorflow.python.platform.gfile) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.gfile.GFile.
WARNING:tensorflow:From from_tensorflow.py:93: The name tf.GraphDef is deprecated. Please use tf.compat.v1.GraphDef instead.

2019-10-29 18:58:40.311461: W tensorflow/core/framework/op_def_util.cc:357] Op BatchNormWithGlobalNormalization is deprecated. It will cease to work in GraphDef version 9. Use tf.nn.batch_normalization().
WARNING:tensorflow:From from_tensorflow.py:99: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.

2019-10-29 18:58:40.419274: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory
2019-10-29 18:58:40.419337: E tensorflow/stream_executor/cuda/cuda_driver.cc:318] failed call to cuInit: UNKNOWN ERROR (303)
2019-10-29 18:58:40.419389: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (164895bae70c): /proc/driver/nvidia/version does not exist
2019-10-29 18:58:40.419751: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-10-29 18:58:40.442045: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3192000000 Hz
2019-10-29 18:58:40.443908: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x72c93a0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2019-10-29 18:58:40.443964: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
WARNING:tensorflow:From /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/testing/tf.py:86: convert_variables_to_constants (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.convert_variables_to_constants`
WARNING:tensorflow:From /usr/local/lib/python3.7/dist-packages/tensorflow_core/python/framework/graph_util_impl.py:277: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.extract_sub_graph`
[18:58:40] /root/tvm/src/relay/ir/module.cc:286: Importing: /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/std/prelude.rly
Traceback (most recent call last):

  File "from_tensorflow.py", line 129, in <module>
    shape=shape_dict)

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/frontend/tensorflow.py", line 2484, in from_tensorflow
    g = GraphProto()

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/frontend/tensorflow.py", line 1965, in __init__
    self._prelude = Prelude(self._mod)

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/prelude.py", line 533, in __init__
    self.load_prelude()

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/prelude.py", line 549, in load_prelude
    self.mod.import_from_std("prelude.rly")

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/module.py", line 239, in import_from_std
    return _module.Module_ImportFromStd(self, file_to_import)

  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/_ctypes/function.py", line 207, in __call__
    raise get_last_ffi_error()

tvm._ffi.base.TVMError: Traceback (most recent call last):
  [bt] (5) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(TVMFuncCall+0x63) [0x7f0f40339573]
  [bt] (4) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(+0xc8e425) [0x7f0f4022e425]
  [bt] (3) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(tvm::relay::ModuleNode::ImportFromStd(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)+0x189) [0x7f0f4022e259]
  [bt] (2) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(tvm::relay::ModuleNode::Import(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)+0x3c7) [0x7f0f4022df17]
  [bt] (1) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(tvm::relay::FromText(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&)+0xf7) [0x7f0f40227ae7]
  [bt] (0) /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so(+0x42075d) [0x7f0f3f9c075d]
  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/_parser.py", line 38, in <module>
    from .grammar.py3.RelayVisitor import RelayVisitor
  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/_ctypes/function.py", line 72, in cfun
    rv = local_pyfunc(*pyargs)
  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/parser.py", line 25, in fromtext
    from tvm.relay import _parser
  File "/usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/_parser.py", line 42, in <module>
    raise Exception("Couldn't find ANTLR parser. Try building with USE_ANTLR=ON.")
ModuleNotFoundError: No module named 'tvm.relay.grammar.py3'
During handling of the above exception, another exception occurred:

Exception: Couldn't find ANTLR parser. Try building with USE_ANTLR=ON.

ANTLR needs Java. Do I really need to have Java and build TVM with ANTLR in order to compile tensorflow models?

#2

@soiferj @MarisaKirisame @jroesch

#3

I found the commit which causes the issue

commit 36a96773bc24f65a52404056d9f1c170ebea206b (HEAD)
Author: Wei Chen <ipondering.weic@gmail.com>
Date:   Thu Oct 17 22:41:34 2019 -0700

    [Relay][Frontend][TF] Add tensor array ops (#3798)

@wweic

Now python/tvm/relay/frontend/tensorflow.py has dependency on Prelude

# tensorflow.py
from tvm.relay.prelude import Prelude

# GraphProto __init__ has
self._prelude = Prelude(self._mod)
#4

Thanks for filing the issue. I can confirm after my change, tensorflow converter now has the dependency on ANTLR since it depends on relay text format program, and ANTLR is used to parse the text format program.

#5

I build TVM with ANTLR and LLVM but after the installation tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/grammar/ does not have py3 submodule
As a result Iā€™m still getting

ModuleNotFoundError: No module named 'tvm.relay.grammar.py3'

cmake output: https://gist.github.com/apivovarov/4fc21f96ec6e94d1813147bd483959c1

ANTLR4=/usr/bin/antlr4

install output: https://gist.github.com/apivovarov/7423d2604825530ffdde97a2cc0937e3

copying build/lib/tvm/relay/grammar/__init__.py -> build/bdist.linux-x86_64/egg/tvm/relay/grammar
#6

Can anyone check that after TVM installation you have py3 submodule under tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/grammar on Linux ?

#7
#8

have same issue with u

#9

missing py3 module issue was fixed today.

Another issue is missing tvm/relay/std/prelude.rly file

I copied prelude.rly file manually to /usr/local/lib/python3.7/dist-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/relay/std/prelude.rly (std sub-folder should be created as well)
After that Tensorflow models compilation is working now

Couple additional notes:
To compile TVM with ANTLR on Ubuntu we need to install antlr4 package

sudo apt install antlr4

To run TVM compilation we need to install python package antlr4-python3-runtime

sudo pip3 install antlr4-python3-runtime

apt and pip3 packages versions should match. On Ubuntu 19.10 both have versions 4.7.2

1 Like