Try again

EricLBuehler · Mar 12, 2024 · 4460683 · 4460683
1 parent f0644c1
commit 4460683
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/candle-nn/src/layer_norm.rs b/candle-nn/src/layer_norm.rs
@@ -151,7 +151,7 @@ impl crate::Module for LayerNorm {
 #[cfg(feature = "cuda")]
 impl crate::Module for LayerNorm {
     fn forward(&self, x: &Tensor) -> Result<Tensor> {
-        /*const K_CUDABLOCK_REDUCE_NUM_THREADS: u32 = 512;
+        const K_CUDABLOCK_REDUCE_NUM_THREADS: u32 = 512;
         let cuda_dev = match x.device() {
             Device::Cpu | Device::Metal(_) => return self.forward_slow(x),
             Device::Cuda(dev) => dev
@@ -189,7 +189,7 @@ impl crate::Module for LayerNorm {
 
         let layernorm = cuda_dev.get_or_load_func(&format!("layernorm_{}", x.dtype().as_str()), kernels::LAYERNORM)?;
         todo!()
-         */
+         /*
         use candle::{CpuStorage, CustomOp1, Layout, Result, Shape, Tensor};
 
         struct InnerLayerNorm {
@@ -297,7 +297,7 @@ impl crate::Module for LayerNorm {
             .expect("Time travel has occurred!")
             .as_micros();
         println!("{}us", end - start);
-        res
+        res*/
     }
 }