EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

server : add SWA checkpoints (#15293)

Browse Source 
* server : add SWA checkpoints

ggml-ci

* cont : server clean-up

* server : handle state restore fails

* llama : add extended llama_state_seq_ API

* server : do not make checkpoints if --swa-full

ggml-ci

* llama : remove flags value for NONE

* server : configure number of SWA checkpoints with CLI arg

ggml-ci

* args : fix scope of new argument
This commit is contained in:
Georgi Gerganov
2025-08-14 14:59:50 +03:00
committed by GitHub
parent 3973163bff
commit d32e03f449
15 changed files with 206 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/llama-kv-cache-unified-iswa.cpp
View File

@@ -194,14 +194,20 @@ bool llama_kv_cache_unified_iswa::get_can_shift() const {
return kv_base->get_size() == kv_swa->get_size();
}
void llama_kv_cache_unified_iswa::state_write(llama_io_write_i & io, llama_seq_id seq_id) const {
kv_base->state_write(io, seq_id);
kv_swa ->state_write(io, seq_id);
void llama_kv_cache_unified_iswa::state_write(llama_io_write_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) const {
if ((flags & LLAMA_STATE_SEQ_FLAGS_SWA_ONLY) == 0) {
kv_base->state_write(io, seq_id, flags);
}
kv_swa->state_write(io, seq_id, flags);
}
void llama_kv_cache_unified_iswa::state_read(llama_io_read_i & io, llama_seq_id seq_id) {
kv_base->state_read(io, seq_id);
kv_swa ->state_read(io, seq_id);
void llama_kv_cache_unified_iswa::state_read(llama_io_read_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) {
if ((flags & LLAMA_STATE_SEQ_FLAGS_SWA_ONLY) == 0) {
kv_base->state_read(io, seq_id, flags);
}
kv_swa->state_read(io, seq_id, flags);
}
llama_kv_cache_unified * llama_kv_cache_unified_iswa::get_base() const {