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

llama: print memory breakdown on exit (#15860)

Browse Source 
* llama: print memory breakdown on exit
This commit is contained in:
Johannes Gäßler
2025-09-24 16:53:48 +02:00
committed by GitHub
parent f2a789e334
commit e789095502
18 changed files with 243 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/llama-memory-hybrid.cpp
View File

@@ -166,6 +166,14 @@ llama_pos llama_memory_hybrid::seq_pos_max(llama_seq_id seq_id) const {
return std::min(mem_attn->seq_pos_max(seq_id), mem_recr->seq_pos_max(seq_id));
}
std::map<ggml_backend_buffer_type_t, size_t> llama_memory_hybrid::memory_breakdown() const {
std::map<ggml_backend_buffer_type_t, size_t> mb = mem_attn->memory_breakdown();
for (const auto & buft_size : mem_recr->memory_breakdown()) {
mb[buft_size.first] += buft_size.second;
}
return mb;
}
void llama_memory_hybrid::state_write(llama_io_write_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) const {
GGML_UNUSED(flags);