minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[AMD] Dsv4/pr2 compressor opt (#26208)

Browse Source 
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: Thomas Wang <1am9trash@gmail.com>
Co-authored-by: Xinyi Song <86638975+RolaoDenthu@users.noreply.github.com>
Co-authored-by: HaiShaw <hixiao@gmail.com>
Co-authored-by: amd-danli103 <danli103@amd.com>
Co-authored-by: Lin, Soga <soga.lin@amd.com>
Co-authored-by: Raiden-Makoto <Raiden-Makoto@users.noreply.github.com>
Co-authored-by: Hubert Lu <55214931+hubertlu-tw@users.noreply.github.com>
Co-authored-by: yichiche@amd.com <jacky.cheng>
Co-authored-by: yctseng0211 <yctseng@amd.com>
Co-authored-by: Bingxu Chen <bingxche@amd.com>
This commit is contained in:
kk
2026-05-25 23:54:40 -07:00
committed by GitHub
co-authored by wunhuang Thomas Wang Xinyi Song HaiShaw amd-danli103 Lin, Soga Raiden-Makoto Hubert Lu yichiche@amd.com yctseng0211 Bingxu Chen
parent 7c0fbc8c2e
commit 3f5e2c7688
31 changed files with 8829 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-kernel/csrc/common_extension.cc
+15
View File
@@ -214,6 +214,21 @@ TORCH_LIBRARY_FRAGMENT(sgl_kernel, m) {
m.def("apply_shuffle_mul_sum(Tensor input, Tensor output, Tensor permutation, Tensor? factors) -> ()");
m.impl("apply_shuffle_mul_sum", torch::kCUDA, &apply_shuffle_mul_sum);
// DeepSeek-V4 fused norm + rope
m.def(
"dsv4_fused_q_norm_rope(Tensor q_input, Tensor! q_output, Tensor freqs_cis, Tensor positions, float eps) -> ()");
m.impl("dsv4_fused_q_norm_rope", torch::kCUDA, &dsv4_fused_q_norm_rope);
m.def(
"dsv4_fused_k_norm_rope_flashmla(Tensor kv, Tensor kv_weight, Tensor freqs_cis, Tensor positions, "
"Tensor out_loc, Tensor! kvcache, float eps, int page_size) -> ()");
m.impl("dsv4_fused_k_norm_rope_flashmla", torch::kCUDA, &dsv4_fused_k_norm_rope_flashmla);
m.def(
"dsv4_fused_q_indexer_rope_hadamard_quant(Tensor q_input, Tensor! q_fp8, Tensor weight, "
"Tensor! weights_out, float weight_scale, Tensor freqs_cis, Tensor positions) -> ()");
m.impl("dsv4_fused_q_indexer_rope_hadamard_quant", torch::kCUDA, &dsv4_fused_q_indexer_rope_hadamard_quant);
m.def(
"fused_qk_norm_rope(Tensor! qkv, int num_heads_q, "
"int num_heads_k, int num_heads_v, int head_dim, float eps, "