Merge branch 'main' of github.com:NVIDIA/TileGym into tilegym_ci_init

Author: arjkesh

README.md

@@ -16,6 +16,7 @@ TileGym is a CUDA Tile kernel library that provides a rich collection of kernel
 ## Overview
 
 This repository aims to provide helpful kernel tutorials and examples for tile-based GPU programming. TileGym is a playground for experimenting with CUDA Tile, where you can learn how to build efficient GPU kernels and explore their integration into real-world large language models such as Llama 3.1 and DeepSeek V2. Whether you're learning tile-based GPU programming or looking to optimize your LLM implementations, TileGym offers practical examples and comprehensive guidance.
+<img width="90%" alt="TileGym_repo" src="https://github.com/user-attachments/assets/1d8741f0-f15c-49ff-ad5c-32d1ae6ec71e" />
 
 ## Features
 

src/tilegym/ops/cutile/silu_and_mul.py

@@ -36,7 +36,6 @@ def silu_and_mul_kernel_row_wise(
     input,
     output,
     TILE_SIZE: ConstInt,
-    n_elements: ConstInt,
     hidden_size: ConstInt,
 ):
     bid = ct.bid(0)  # this gives us our row
@@ -47,7 +46,6 @@ def silu_and_mul_kernel_row_wise(
     row_idx = bid
     a_col_idx = offsets  # First half: [0, hidden_size)
     b_col_idx = offsets + hidden_size  # Second half: [hidden_size, 2*hidden_size)
-    out_offsets = bid * hidden_size + offsets
 
     # Load tiles using gather with 2D indices
     # gather broadcasts (scalar, tile) to (tile,)
@@ -95,7 +93,6 @@ def silu_and_mul(
     # Flatten input to 2D: (batch_size, 2 * hidden_size)
     input_flat = input.view(-1, original_shape[-1])
     batch_size = input_flat.shape[0]
-    n_elements = batch_size * hidden_size  # Total elements to process in output
 
     # Get final output shape
     output_shape = list(original_shape)
@@ -123,6 +120,11 @@ def silu_and_mul(
         torch.cuda.current_stream(),
         grid,
         silu_and_mul_kernel_row_wise,
-        (input_flat, output, TILE_SIZE, n_elements, hidden_size),
+        (
+            input_flat,
+            output,
+            TILE_SIZE,
+            hidden_size
+        ),
     )
     return output.reshape(*output_shape)