tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-29 21:54:07 -04:00
Code Issues Packages Projects Releases Wiki Activity

ubatch : new splitting logic (#14217)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-20 10:14:14 +03:00
committed by GitHub
parent 9eaa51e7f0
commit 4c9fdfbe15
19 changed files with 992 additions and 915 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llama-kv-cache-unified-iswa.h
View File

@@ -32,7 +32,7 @@ public:
//
llama_memory_state_ptr init_batch(
const llama_batch & batch,
llama_batch_allocr & balloc,
uint32_t n_ubatch,
bool embd_all) override;
@@ -90,7 +90,6 @@ public:
// used to create a state from a batch
llama_kv_cache_unified_iswa_state(
llama_kv_cache_unified_iswa * kv,
llama_sbatch sbatch,
std::vector<uint32_t> heads_base,
std::vector<uint32_t> heads_swa,
std::vector<llama_ubatch> ubatches);
@@ -104,8 +103,6 @@ public:
bool next() override;
bool apply() override;
std::vector<int64_t> & out_ids() override;
llama_memory_status get_status() const override;
const llama_ubatch & get_ubatch() const override;
@@ -119,8 +116,6 @@ public:
private:
//llama_kv_cache_unified_iswa * kv;
llama_sbatch sbatch;
// the index of the next ubatch to process
size_t i_next = 0;