apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Attention] remove cum_offsets from atten, and use cu_seqlens_q (#2870)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
[Attention] remove cum_offsets from atten, and use cu_seqlens_q (#2870)
This commit is contained in:
周周周
2025-07-16 20:10:57 +08:00
committed by GitHub
parent 42b80182e0
commit aa76085d1f
47 changed files with 237 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
custom_ops/gpu_ops/cpp_extensions.cc
View File

@@ -54,7 +54,7 @@ std::vector<paddle::Tensor> AppendAttention(
const paddle::Tensor &value_cache, const paddle::Tensor &seq_lens_encoder,
const paddle::Tensor &seq_lens_decoder,
const paddle::Tensor &seq_lens_this_time,
const paddle::Tensor &padding_offsets, const paddle::Tensor &cum_offsets,
const paddle::Tensor &padding_offsets, const paddle::Tensor &cu_seqlens_q,
const paddle::Tensor &block_tables, const paddle::Tensor &encoder_batch_ids,
const paddle::Tensor &encoder_tile_ids_per_batch,
const paddle::Tensor &encoder_num_blocks,
@@ -94,7 +94,7 @@ std::vector<paddle::Tensor> GQARopeWriteCacheKernel(
const paddle::Tensor &seq_lens_this_time,
const paddle::Tensor &seq_lens_encoder,
const paddle::Tensor &seq_lens_decoder,
const paddle::Tensor &padding_offsets, const paddle::Tensor &cum_offsets,
const paddle::Tensor &padding_offsets,
const paddle::Tensor &block_tables, const paddle::Tensor &kv_batch_ids,
const paddle::Tensor &kv_tile_ids, const paddle::Tensor &kv_num_blocks,
const paddle::Tensor &cache_batch_ids, const paddle::Tensor &cache_tile_ids,
@@ -331,7 +331,7 @@ std::vector<paddle::Tensor> DecodeMLAWriteCacheKernel(
const paddle::Tensor& seq_lens,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& cum_offsets,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const std::string& cache_quant_type_str,
const int max_seq_len,
@@ -344,7 +344,7 @@ std::vector<paddle::Tensor> DecodeMLAWriteCacheKernel(
const paddle::Tensor& seq_lens,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& cum_offsets,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const std::string& cache_quant_type_str,
const int max_seq_len);
@@ -370,7 +370,6 @@ std::vector<paddle::Tensor> MultiHeadLatentAttention(
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& cum_offsets,
const paddle::Tensor& block_tables,
const paddle::Tensor& encoder_batch_ids,
const paddle::Tensor& encoder_tile_ids_per_batch,