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

[Attention] MLA with chunked prefill (#12639)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Patrick Horn <patrick.horn@gmail.com>
Co-authored-by: simon-mo <xmo@berkeley.edu>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
This commit is contained in:
Lucas Wilkinson
2025-02-21 18:30:12 -05:00
committed by GitHub
parent 900edbfa48
commit 288cc6c234
18 changed files with 1910 additions and 1275 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
csrc/cache.h
View File

@@ -39,3 +39,10 @@ void concat_and_cache_mla(torch::Tensor& kv_c, torch::Tensor& k_pe,
// Just for unittest
void convert_fp8(torch::Tensor& dst_cache, torch::Tensor& src_cache,
const double scale, const std::string& kv_cache_dtype);
void gather_cache(
torch::Tensor const& src_cache, // [NUM_BLOCKS, BLOCK_SIZE, ENTRIES...]
torch::Tensor const& dst, // [TOT_TOKENS, ENTRIES...]
torch::Tensor const& block_table, // [BATCH, BLOCK_INDICES]
torch::Tensor const& cu_seq_lens, // [BATCH+1]
int64_t batch_size, std::optional<torch::Tensor> seq_starts = std::nullopt);