biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Chore] Remove unused batched RoPE op & kernel (#24789)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2025-09-13 00:08:20 -07:00
committed by GitHub
parent 99bfef841f
commit 5febdc8750
8 changed files with 16 additions and 348 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
csrc/torch_bindings.cpp
View File

@@ -214,16 +214,6 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
" Tensor cos_sin_cache, bool is_neox) -> ()");
ops.impl("rotary_embedding", torch::kCUDA, &rotary_embedding);
// Apply GPT-NeoX or GPT-J style rotary embedding to query and key
// (supports multiple loras).
ops.def(
"batched_rotary_embedding(Tensor positions, Tensor! query,"
" Tensor!? key, int head_size,"
" Tensor cos_sin_cache, bool is_neox,"
" int rot_dim,"
" Tensor cos_sin_cache_offsets) -> ()");
ops.impl("batched_rotary_embedding", torch::kCUDA, &batched_rotary_embedding);
// Quantization ops
#ifndef USE_ROCM
// Quantized GEMM for AWQ.