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

8
src/llama-kv-cache-unified.cpp
View File

@@ -1828,7 +1828,9 @@ bool llama_kv_cache_unified::is_masked_swa(llama_pos p0, llama_pos p1) const {
return false;
}
void llama_kv_cache_unified::state_write(llama_io_write_i & io, llama_seq_id seq_id) const {
void llama_kv_cache_unified::state_write(llama_io_write_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) const {
GGML_UNUSED(flags);
io.write(&n_stream, sizeof(n_stream));
for (uint32_t s = 0; s < n_stream; ++s) {
@@ -1879,7 +1881,9 @@ void llama_kv_cache_unified::state_write(llama_io_write_i & io, llama_seq_id seq
}
}
void llama_kv_cache_unified::state_read(llama_io_read_i & io, llama_seq_id seq_id) {
void llama_kv_cache_unified::state_read(llama_io_read_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) {
GGML_UNUSED(flags);
GGML_ASSERT(seq_id == -1 || (seq_id >= 0 && (size_t) seq_id < seq_to_stream.size()));
uint32_t n_stream_cur;