[CI] Update ci-gpu image (#15836)

* [CI] Update ci-gpu image

The ci-gpu is updated to make changes in #15819 reflect in CI.

* update image that contains oneflow==0.8.0

* Force old coreml api to fix failing test

Between the previous ci_gpu image and the one in use now, the version of
coreml has changed since it is unpinned (6.3.0 -> 7.0.0). This caused a
test failure in the coreml frontend tests. The problem is that coreml
now uses a new default format for models. This commit forces coreml to
use the old model format so the test passes.

More information can be found here: https://apple.github.io/coremltools/docs-guides/source/convert-learning-models.html

Author: lhutton1

ci/jenkins/docker-images.ini

@@ -20,7 +20,7 @@
 ci_arm: tlcpack/ci-arm:20230615-060132-62a5e7acf
 ci_cortexm: tlcpack/ci-cortexm:20230710-060128-a60cd0fec
 ci_cpu: tlcpack/ci-cpu:20230604-060130-0af9ff90e
-ci_gpu: tlcpack/ci-gpu:20230504-142417-4d37a0a0
+ci_gpu: tlcpack/ci-gpu:20231004-111300-7a1f7d0b
 ci_hexagon: tlcpack/ci-hexagon:20230504-142417-4d37a0a0
 ci_i386: tlcpack/ci-i386:20230504-142417-4d37a0a0
 ci_lint: tlcpack/ci-lint:20230504-142417-4d37a0a0

tests/python/frontend/coreml/test_forward.py

@@ -823,7 +823,9 @@ def test_can_build_keras_to_coreml_to_relay():
         kmodel_fn = path.join(tmpdir, "c1mdl.h5")
         model.save(kmodel_fn)
 
-        mdl = cm.convert(kmodel_fn)
+        mdl = cm.convert(
+            kmodel_fn, convert_to="neuralnetwork", minimum_deployment_target=cm.target.macOS11
+        )
         model_file = path.join(tmpdir, "c1.mlmodel")
         mdl.save(model_file)