diff --git a/ci/jenkins/docker-images.ini b/ci/jenkins/docker-images.ini
index 6e55160521b3..175917f887b7 100644
--- a/ci/jenkins/docker-images.ini
+++ b/ci/jenkins/docker-images.ini
@@ -17,13 +17,13 @@
 
 # This data file is read during when Jenkins runs job to determine docker images.
 [jenkins]
-ci_arm: tlcpack/ci-arm:20240428-060115-0b09ed018
-ci_cortexm: tlcpack/ci-cortexm:20240428-060115-0b09ed018
-ci_cpu: tlcpack/ci_cpu:20240428-060115-0b09ed018
-ci_gpu: tlcpack/ci-gpu:20240428-060115-0b09ed018
-ci_hexagon: tlcpack/ci-hexagon:20240428-060115-0b09ed018
-ci_i386: tlcpack/ci-i386:20240428-060115-0b09ed018
-ci_lint: tlcpack/ci-lint:20240428-060115-0b09ed018
-ci_minimal: tlcpack/ci-minimal:20240428-060115-0b09ed018
-ci_riscv: tlcpack/ci-riscv:20240428-060115-0b09ed018
-ci_wasm: tlcpack/ci-wasm:20240428-060115-0b09ed018
+ci_arm: tlcpack/ci-arm:20240917-153130-9f281758
+ci_cortexm: tlcpack/ci-cortexm:20240917-153130-9f281758
+ci_cpu: tlcpack/ci_cpu:20240917-153130-9f281758
+ci_gpu: tlcpack/ci-gpu:20240917-153130-9f281758
+ci_hexagon: tlcpack/ci-hexagon:20240917-153130-9f281758
+ci_i386: tlcpack/ci-i386:20240917-153130-9f281758
+ci_lint: tlcpack/ci-lint:20240917-153130-9f281758
+ci_minimal: tlcpack/ci-minimal:20240917-153130-9f281758
+ci_riscv: tlcpack/ci-riscv:20240917-153130-9f281758
+ci_wasm: tlcpack/ci-wasm:20240917-153130-9f281758
diff --git a/tests/python/frontend/pytorch/test_fx_quant.py b/tests/python/frontend/pytorch/test_fx_quant.py
index 7f3083a7dcd0..8ed6e1a74797 100644
--- a/tests/python/frontend/pytorch/test_fx_quant.py
+++ b/tests/python/frontend/pytorch/test_fx_quant.py
@@ -87,6 +87,9 @@ def forward(self, inp):
     quantize_and_build(model, 300)
 
 
+@pytest.mark.skip(
+    reason="Model binary isn't uploaded to S3. See https://github.com/apache/tvm/pull/17397"
+)
 def test_imagenet():
     for model_func in [resnet50, efficientnet_b4]:
         quantize_and_build(model_func(pretrained=True).eval(), 224)
diff --git a/tests/python/relax/test_frontend_onnx.py b/tests/python/relax/test_frontend_onnx.py
index 8f4e9881f497..0e7cfbd7c093 100644
--- a/tests/python/relax/test_frontend_onnx.py
+++ b/tests/python/relax/test_frontend_onnx.py
@@ -76,6 +76,7 @@ def check_correctness(
     inputs: Optional[Dict[str, np.ndarray]] = None,
     ir_version: int = 8,
     opset: int = 14,
+    rtol: float = 1e-7,
     atol: float = 1e-5,
 ) -> None:
     """Run an onnx model in both onnxruntime and TVM through our importer
@@ -154,7 +155,7 @@ def check_correctness(
         # TODO Allow configurable tolerance.
         # Sometimes None is used to indicate an unused output.
         if ort_out is not None:
-            tvm.testing.assert_allclose(tvm_out.numpy(), ort_out, atol=atol)
+            tvm.testing.assert_allclose(tvm_out.numpy(), ort_out, rtol=rtol, atol=atol)
 
 
 @pytest.mark.parametrize(
@@ -1010,7 +1011,7 @@ def verify_reduce_func(func, data, axis, keepdims):
 
         inputs_dict = {"x": data}
         # Reduction ops accumulate arithmetic errors, so we use a higher tolerance.
-        check_correctness(model, inputs_dict, opset=11, atol=1e-4)
+        check_correctness(model, inputs_dict, opset=11, rtol=1e-4, atol=1e-4)
 
     for keepdims in [True, False]:
         verify_reduce_func(
diff --git a/tests/python/tir-transform/test_tir_transform_simplify.py b/tests/python/tir-transform/test_tir_transform_simplify.py
index f7887bc61137..0b2d5f16d833 100644
--- a/tests/python/tir-transform/test_tir_transform_simplify.py
+++ b/tests/python/tir-transform/test_tir_transform_simplify.py
@@ -1021,18 +1021,40 @@ class TestMostRestrictiveConditional(BaseBeforeAfter):
     then `a >= b` cannot be proven, but can be reduced to `a == b`.
     """
 
+    class TupleWrapper(tuple):
+        """
+        A custom wrapper for `tuple` to handle element-wise equality comparison
+        to avoid comparison errors when dealing with objects like `ExprOp`.
+        See also: https://github.com/apache/tvm/pull/17397
+        """
+
+        def __new__(self, *args):
+            return super().__new__(self, args)
+
+        def __eq__(self, other):
+            from tvm.tir.expr import ExprOp
+
+            for a, b in zip(self, other):
+                if isinstance(a, ExprOp) and isinstance(a, ExprOp):
+                    if not tvm.ir.structural_equal(a, b):
+                        return False
+                else:
+                    if not a.__eq__(b):
+                        return False
+            return True
+
     i, j, k = [tvm.tir.Var(name, "int32") for name in "ijk"]
     tir_int = tvm.tir.IntImm("int32", 0)
 
     test_case = tvm.testing.parameter(
-        (i <= tir_int, tir_int <= i, i == tir_int),
-        (i <= tir_int, i != tir_int, i < tir_int),
-        (i != tir_int, i <= tir_int, i < tir_int),
-        (i != tir_int, tir_int <= i, tir_int < i),
-        (i <= j, j <= i, j == i),
-        (i <= j, i != j, i < j),
-        (i != j, i <= j, i < j),
-        (i != j, j <= i, j < i),
+        TupleWrapper(i <= tir_int, tir_int <= i, i == tir_int),
+        TupleWrapper(i <= tir_int, i != tir_int, i < tir_int),
+        TupleWrapper(i != tir_int, i <= tir_int, i < tir_int),
+        TupleWrapper(i != tir_int, tir_int <= i, tir_int < i),
+        TupleWrapper(i <= j, j <= i, j == i),
+        TupleWrapper(i <= j, i != j, i < j),
+        TupleWrapper(i != j, i <= j, i < j),
+        TupleWrapper(i != j, j <= i, j < i),
     )
 
     @tvm.testing.fixture
diff --git a/tests/scripts/task_build_hexagon_api.sh b/tests/scripts/task_build_hexagon_api.sh
index 5f811e4e2749..cff6d7a6ba59 100755
--- a/tests/scripts/task_build_hexagon_api.sh
+++ b/tests/scripts/task_build_hexagon_api.sh
@@ -41,10 +41,7 @@ fi
 mkdir -p build
 cd build
 
-cmake -DANDROID_ABI=arm64-v8a \
-    -DANDROID_PLATFORM=android-28 \
-    -DUSE_ANDROID_TOOLCHAIN="${ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake" \
-    -DUSE_HEXAGON_ARCH=v68 \
+cmake -DUSE_HEXAGON_ARCH=v68 \
     -DUSE_HEXAGON_SDK="${HEXAGON_SDK_ROOT}" \
     -DUSE_HEXAGON_TOOLCHAIN="${HEXAGON_TOOLCHAIN}" \
     -DUSE_OUTPUT_BINARY_DIR="${output_directory}" \