RTX 4090 error when using demo1

Traceback (most recent call last):
File "/usr/local/bin/lmcache_vllm", line 5, in
from lmcache_vllm.script import main
File "/usr/local/lib/python3.12/dist-packages/lmcache_vllm/__init__.py", line 4, in
from lmcache_vllm.vllm_injection import InitLMCacheEnvironment
File "/usr/local/lib/python3.12/dist-packages/lmcache_vllm/vllm_injection.py", line 15, in
from lmcache_vllm.vllm_adapter import (lmcache_get_config,
File "/usr/local/lib/python3.12/dist-packages/lmcache_vllm/vllm_adapter.py", line 32, in
LMCACHE_CUDA_STREAM = torch.cuda.Stream()
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/torch/cuda/streams.py", line 35, in __new__
return super().__new__(cls, priority=priority, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: CUDA error: forward compatibility was attempted on non supported HW
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Device-side assertions were explicitly omitted for this error check; the error probably arose while initializing the DSA handlers.

Start with the demo1 invocation and the traceback entry point in lmcache_vllm/vllm_adapter.py, where torch.cuda.Stream() fails. Reproduce the error while checking the RTX 4090, PyTorch, and CUDA environment versions. Done means identifying a supported configuration or documenting that the demo cannot run with the reported setup.

Written by the indexing model from the issue text.