tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 12:35:16 +00:00
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.h
View File

@ -57,7 +57,7 @@ public:
//
llama_memory_state_ptr init_batch(
const llama_batch & batch,
llama_batch_allocr & balloc,
uint32_t n_ubatch,
bool embd_all) override;
@ -231,7 +231,6 @@ public:
// used to create a decode state from a batch
llama_kv_cache_unified_state(
llama_kv_cache_unified * kv,
llama_sbatch sbatch,
ubatch_heads heads,
std::vector<llama_ubatch> ubatches);
@ -244,8 +243,6 @@ 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;
@ -286,8 +283,6 @@ private:
// batch processing state
//
llama_sbatch sbatch;
// the index of the next ubatch to process
size_t i_next = 0;