blue-candle
blue-candle copied to clipboard
xnorpx
Object detection service for Blue Iris
fn new(ordinal: usize) -> Result { let device = cudarc::driver::CudaDevice::new(ordinal).w()?; let blas = cudarc::cublas::CudaBlas::new(device.clone()).w()?; let curand = cudarc::curand::CudaRng::new(299792458, device.clone()).w()?; Ok(Self { id: DeviceId::new(), device, blas: Arc::new(blas), curand: Arc::new(Mutex::new(CudaRng(curand))), }) }...
Readd different Cuda versions with Cuda versions in CI with set NVCC_APPEND_FLAGS=-allow-unsupported-compiler as environment variable.
I am just wondering if there will be support for .pt (Pytorch) format for models? I would like to give Blue Candle a serious try but the default yolov8 models...
Implement missing endpoints: / /v1/vision/custom/list curl -i http://localhost:32168/ HTTP/1.1 200 OK Content-Length: 28745 Content-Type: text/html Date: Wed, 24 Jan 2024 15:21:03 GMT Server: Kestrel Accept-Ranges: bytes ETag: "1da4e4175f9a949" Last-Modified: Tue,...
After some hour runing, there is an out of memory issue : > thread 'tokio-runtime-worker' panicked at src\bin\blue_candle.rs:186:52: called `Result::unwrap()` on an `Err` value: DriverError(CUDA_ERROR_OUT_OF_MEMORY, "out of memory")
Implement an actor for each GPU in the system, then use one mpsc channel to schedule work on this actor. If the mpsc channel has to much work for the...
I figured I would create a new issue for this. At current you can plug in one model in the launch options, but that creates potential issues if you have...
