[Docker] Update to Torch 1.10.1 (apache#9781)

* update pytorch to 1.10.1

* fix missing import test only on llvm and cuda

* Revert "[Docker][Onnx] Upgrade ONNX to latest version (apache#9519)"

This reverts commit 3f5dca5.

* skip testing if target is not enabled

docker/install/ubuntu_install_onnx.sh

@@ -27,14 +27,13 @@ set -o pipefail
 # https://github.com/onnx/onnx/pull/2834).  When updating the CI image
 # to onnx>=1.9, onnxoptimizer should also be installed.
 pip3 install \
-    onnx==1.10.2 \
-    onnxruntime==1.9.0 \
-    onnxoptimizer==0.2.6
-
+    onnx==1.8.1 \
+    onnxruntime==1.7.0
+
 # torch depends on a number of other packages, but unhelpfully, does
 # not expose that in the wheel!!!
 pip3 install future
 
 pip3 install \
-    torch==1.7.0 \
-    torchvision==0.8.1
+    torch==1.10.1 \
+    torchvision==0.11.2

python/tvm/relay/frontend/pytorch.py

@@ -23,6 +23,7 @@
 import itertools
 import math
 import sys
+import logging
 
 import numpy as np
 import tvm

tests/python/frontend/onnx/test_forward.py

@@ -4967,38 +4967,11 @@ def verify_eyelike(indata):
 )
 
 unsupported_onnx_tests = [
-    "test_basic_convinteger",
-    "test_batchnorm_epsilon_training_mode",
-    "test_batchnorm_example_training_mode",
-    "test_bernoulli",
-    "test_bernoulli_expanded",
-    "test_bernoulli_double",
-    "test_bernoulli_double_expanded",
-    "test_bernoulli_seed",
-    "test_bernoulli_seed_expanded",
     "test_cast_BFLOAT16_to_FLOAT",
     "test_cast_DOUBLE_to_FLOAT16",
     "test_cast_FLOAT_to_BFLOAT16",
     "test_cast_FLOAT_to_STRING",
     "test_cast_STRING_to_FLOAT",
-    "test_castlike_BFLOAT16_to_FLOAT",
-    "test_castlike_BFLOAT16_to_FLOAT_expanded",
-    "test_castlike_DOUBLE_to_FLOAT",
-    "test_castlike_DOUBLE_to_FLOAT16",
-    "test_castlike_DOUBLE_to_FLOAT16_expanded",
-    "test_castlike_FLOAT16_to_DOUBLE",
-    "test_castlike_FLOAT16_to_FLOAT",
-    "test_castlike_FLOAT_to_BFLOAT16",
-    "test_castlike_FLOAT_to_BFLOAT16_expanded",
-    "test_castlike_FLOAT_to_DOUBLE",
-    "test_castlike_FLOAT_to_FLOAT16",
-    "test_castlike_FLOAT_to_STRING",
-    "test_castlike_FLOAT_to_STRING_expanded",
-    "test_castlike_STRING_to_FLOAT",
-    "test_castlike_STRING_to_FLOAT_expanded",
-    "test_convinteger_with_padding",
-    "test_convinteger_without_padding",
-    "test_convtranspose_autopad_same",
     "test_convtranspose_dilations",
     "test_convtranspose_output_shape",
     "test_cumsum_1d",
@@ -5014,13 +4987,9 @@ def verify_eyelike(indata):
     "test_dropout_default_mask",
     "test_dropout_default_mask_ratio",
     "test_dropout_default_ratio",
-    "test_gru_batchwise",
-    "test_hardswish",
-    "test_identity_sequence",
     "test_if_seq",
     "test_loop11",
     "test_loop13_seq",
-    "test_lstm_batchwise",
     "test_matmulinteger",
     "test_maxpool_2d_same_lower",
     "test_maxpool_2d_same_upper",
@@ -5030,10 +4999,6 @@ def verify_eyelike(indata):
     "test_mvn",
     # This test fails llvm with a lowering error:
     "test_nllloss_NCd1d2d3_none_no_weight_negative_ii_expanded",
-    "test_optional_has_element",
-    "test_optional_get_element",
-    "test_optional_get_element_sequence",
-    "test_optional_has_element_empty",
     "test_qlinearmatmul_3D",
     "test_range_float_type_positive_delta_expanded",
     "test_range_int32_type_negative_delta_expanded",
@@ -5051,13 +5016,6 @@ def verify_eyelike(indata):
     "test_round",
     "test_sequence_insert_at_back",
     "test_sequence_insert_at_front",
-    "test_shape_end_1",
-    "test_shape_end_negative_1",
-    "test_shape_start_1",
-    "test_shape_start_1_end_2",
-    "test_shape_start_1_end_negative_1",
-    "test_shape_start_negative_1",
-    "test_simple_rnn_batchwise",
     "test_simple_rnn_defaults",
     "test_simple_rnn_with_initial_bias",
     "test_split_variable_parts_1d",
@@ -5083,24 +5041,6 @@ def verify_eyelike(indata):
     "test_training_dropout_mask",
     "test_training_dropout_zero_ratio",
     "test_training_dropout_zero_ratio_mask",
-    "test_tril",
-    "test_tril_pos",
-    "test_tril_square",
-    "test_tril_square_neg",
-    "test_tril_neg",
-    "test_tril_one_row_neg",
-    "test_tril_out_neg",
-    "test_tril_out_pos",
-    "test_tril_zero",
-    "test_triu",
-    "test_triu_one_row",
-    "test_triu_out_neg_out",
-    "test_triu_out_pos",
-    "test_triu_neg",
-    "test_triu_pos",
-    "test_triu_square",
-    "test_triu_square_neg",
-    "test_triu_zero",
     # These unsqueeze tests work, but take 2+ hrs to run
     "test_unsqueeze_three_axes",
     "test_unsqueeze_two_axes",

tests/python/frontend/pytorch/test_forward.py

@@ -212,7 +212,10 @@ def verify_model(
     compiled_input = dict(zip(input_names, [inp.clone().cpu().numpy() for inp in baseline_input]))
 
     with tvm.transform.PassContext(opt_level=3):
-        for target, dev in tvm.testing.enabled_targets():
+        for target in ["llvm", "cuda"]:
+            if not tvm.runtime.enabled(target):
+                continue
+            dev = tvm.device(target, 0)
             relay_graph, relay_lib, relay_params = relay.build(mod, target=target, params=params)
             relay_model = graph_executor.create(relay_graph, relay_lib, dev)
             relay_model.set_input(**relay_params)
@@ -2196,7 +2199,7 @@ def test_3d_models():
 
 
 def _get_default_vm_targets():
-    return [tgt for (tgt, _) in tvm.testing.enabled_targets()]
+    return ["llvm", "cuda"]
 
 
 def verify_script_model(pt_model, ishapes, targets, idtype=None):
@@ -2269,7 +2272,10 @@ def verify_model_vm(input_model, ishapes, idtype=None, idata=None, targets=["llv
     mod, params = relay.frontend.from_pytorch(input_model, input_shapes)
 
     for tgt in targets:
+        if not tvm.runtime.enabled(tgt):
+            continue
         print("Running on target", tgt)
+
         dev = tvm.device(tgt, 0)
 
         evaluator = relay.create_executor("vm", mod=mod, device=dev, target=tgt).evaluate()
@@ -3897,15 +3903,15 @@ def test_fn(x, mask):
     for shape in [(10,), (3, 4), (16, 32, 64)]:
         x = torch.randn(*shape)
         mask = x.ge(0.5)
-        verify_trace_model(test_fn, [x, mask], ["llvm", "cuda", "nvptx"])
+        verify_trace_model(test_fn, [x, mask], ["llvm", "cuda"])
 
 
 def test_unique():
     def test_fn(is_sorted, return_inverse, return_counts):
         return lambda x: torch.unique(x, is_sorted, return_inverse, return_counts)
 
     in_data = torch.randint(0, 20, (10,), dtype=torch.int32)
-    targets = ["llvm", "cuda", "nvptx"]
+    targets = ["llvm", "cuda"]
     verify_trace_model(test_fn(True, True, True), [in_data], targets)
     verify_trace_model(test_fn(True, False, True), [in_data], targets)
     verify_trace_model(test_fn(True, True, False), [in_data], targets)