This runs https://ollama.ai/ in a gVisor container and loads two models:
an English-Chinese translation model, and a code assistant model.
It asks the first one to translate "Hello World" to Chinese, and then asks
the second one to generate a test case to verify that the translation is
correct.
This change includes a server and client library for spawning ollama in a
container and interacting through its HTTP API. This will be useful to turn
it into a benchmark that measures its throughput in tokens/second.
PiperOrigin-RevId: 590295278
- The app attempts to allocate a driver object of class NV_CONFIDENTIAL_COMPUTE
in drivers too old to support said class; said allocation must fail with
status NV_ERR_INVALID_CLASS rather than errno EINVAL for the app to proceed.
- UVM_VALIDATE_VA_RANGE checks that a given address range is known to
nvidia-uvm. For mmaps of /dev/nvidia-uvm, this requires that the application
mmap (handled by nvproxy) immediately result in a host mmap (handled by the
driver). Ensure that mappings of nvproxy's nvidia-uvm have this property.
- Pass through UVM ioctl UVM_DISABLE_READ_DUPLICATION.
Fixes#9593
PiperOrigin-RevId: 577966817
Zach KoopmansandgVisor bot
Andrei VaginandgVisor bot
Etienne PerotandgVisor bot
Jamie LiuandgVisor bot