huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 10:38:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Dilated convolutions (#657)

Browse Source 
* Add the dilation parameter.

* Restore the basic optimizer example.

* Dilation support in cudnn.

* Use the dilation parameter in the cpu backend.

* More dilation support.

* No support for dilation in transposed convolutions.

* Add dilation to a test.

* Remove a print.

* Helper function.
This commit is contained in:
Laurent Mazare
2023-08-29 16:12:11 +01:00
committed by GitHub
parent ee8bb1bde1
commit a044907ffc
20 changed files with 231 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
candle-core/examples/basics.rs
View File

@ -11,7 +11,7 @@ fn main() -> Result<()> {
let inp = Tensor::randn(0f32, 1., (2, 320, 96, 96), &Device::Cpu)?;
let w = Tensor::randn(0f32, 1., (320, 320, 3, 3), &Device::Cpu)?;
let start = std::time::Instant::now();
let res = inp.conv2d(&w, 0, 1, 1)?;
let res = inp.conv2d(&w, 0, 1, 1, 1)?;
println!("{:?}", start.elapsed());
println!("{res:?}");
Ok(())

4
candle-core/examples/cpu_benchmarks.rs
View File

@ -40,7 +40,7 @@ impl Benchmark for Conv1d {
}
fn run_one(d: &Self::PreProcessData) -> Result<Self::RunResult> {
d.0.conv1d(&d.1, 0, 1, 1)
d.0.conv1d(&d.1, 0, 1, 1, 1)
}
const ITERS: usize = 5;
@ -59,7 +59,7 @@ impl Benchmark for Conv2d {
}
fn run_one(d: &Self::PreProcessData) -> Result<Self::RunResult> {
d.0.conv2d(&d.1, 0, 1, 1)
d.0.conv2d(&d.1, 0, 1, 1, 1)
}
const ITERS: usize = 1;

6
candle-core/examples/cuda_basics.rs
View File

@ -11,11 +11,11 @@ fn main() -> Result<()> {
let device = Device::new_cuda(0)?;
let in_t = Tensor::rand(-1f32, 1f32, (1, 3, 12, 7), &device)?;
let k_t = Tensor::rand(-1f32, 1f32, (6, 3, 1, 1), &device)?;
let out_t = in_t.conv2d(&k_t, 0, 1, 1)?;
let out_t = in_t.conv2d(&k_t, 0, 1, 1, 1)?;
println!("{out_t}");
let in_t = in_t.to_device(&Device::Cpu)?;
let k_t = k_t.to_device(&Device::Cpu)?;
let out_t2 = in_t.conv2d(&k_t, 0, 1, 1)?;
let out_t2 = in_t.conv2d(&k_t, 0, 1, 1, 1)?;
let diff = (out_t.to_device(&Device::Cpu)? - out_t2)?
.sqr()?
.sum_all()?;
@ -23,7 +23,7 @@ fn main() -> Result<()> {
let t = Tensor::randn(0f32, 1f32, (2, 4, 96, 96), &device)?;
let w = Tensor::randn(0f32, 1f32, (320, 4, 3, 3), &device)?;
let res = t.conv2d(&w, 1, 1, 1)?;
let res = t.conv2d(&w, 1, 1, 1, 1)?;
println!("{res:?}");
Ok(())
}