[Bug] tvm.cuda().exist return false while torch.cuda.is_available() return true

[vfdff]

As the title mentions, tvm.cuda().exist return unexpected value

Expected behavior

tvm.cuda().exist return true

Actual behavior

tvm.cuda().exist return false

Environment

Any environment details, such as: Operating System, TVM version, etc
a) gpu

(venv) root@d00469708debug6-7c75445547-frh8j:/usr1/project/zhongyunde/source/osdi22_artifact/artifacts/roller# nvidia-smi   
Wed Dec 11 19:48:21 2024       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.129.03             Driver Version: 535.129.03   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA L40S                    Off | 00000000:9A:00.0 Off |                    0 |
| N/A   27C    P8              34W / 350W |     13MiB / 46068MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|

b) tvm: apache-tvm 0.11.1
c) mlc-ai: mlc-ai-nightly-cu122-0.1

Steps to reproduce

pip install apache-tvm
python3 -m pip install --pre mlc-ai-nightly-cu122 -- don't fix

Triage

Please refer to the list of label tags here to find the relevant tags and add them below in a bullet format (example below).

• needs-triage

[vfdff]

I try to build the tvm base on source on gpu machine (83),
(venv3) root@d00469708debug6-7c75445547-frh8j:/usr1/project/zhongyunde/source/tvm/cmake# git diff .

diff --git a/cmake/config.cmake b/cmake/config.cmake
index 791751a..a2e293a 100644
--- a/cmake/config.cmake
+++ b/cmake/config.cmake
@@ -46,7 +46,7 @@
 # - ON: enable CUDA with cmake's auto search
 # - OFF: disable CUDA
 # - /path/to/cuda: use specific path to cuda toolkit
-set(USE_CUDA OFF)
+set(USE_CUDA ON)
 
 # Whether to enable NCCL support:
 # - ON: enable NCCL with cmake's auto search
@@ -158,7 +158,8 @@ set(USE_PROFILER ON)
 # - OFF: disable llvm, note this will disable CPU codegen
 #        which is needed for most cases
 # - /path/to/llvm-config: enable specific LLVM when multiple llvm-dev is available.
-set(USE_LLVM OFF)
+set(USE_LLVM "/usr/bin/llvm-config --link-static")
+set(HIDE_PRIVATE_SYMBOLS ON)
 
 # Whether use MLIR to help analyze, requires USE_LLVM is enabled
 # Possible values: ON/OFF
@@ -307,7 +308,7 @@ set(USE_CLML_GRAPH_EXECUTOR OFF)
 set(USE_ANTLR OFF)
 
 # Whether use Relay debug mode
-set(USE_RELAY_DEBUG OFF)
+set(USE_RELAY_DEBUG ON)

and It is surprised that finally there is only libtvm_runtime.so and no libtvm.so under the build ?
#17562