tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-09 01:52:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

kv-cache : restore find_slot impl

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-07-04 19:08:31 +03:00
parent 5c00eb22f2
commit 31feaee04d
1 changed files with 5 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/llama-kv-cache-unified.cpp
View File

@ -789,7 +789,7 @@ llama_kv_cache_unified::slot_info llama_kv_cache_unified::find_slot(const llama_
res.s1 = std::max<llama_seq_id>(res.s1, seq_to_stream[seq_id]); res.s1 = std::max<llama_seq_id>(res.s1, seq_to_stream[seq_id]);
res.strm[s] = seq_to_stream[seq_id]; res.strm[s] = seq_to_stream[seq_id];
res.idxs[s].resize(n_tokens); res.idxs[s].reserve(n_tokens);
const auto & cells = v_cells[seq_to_stream[seq_id]]; const auto & cells = v_cells[seq_to_stream[seq_id]];
@ -806,7 +806,6 @@ llama_kv_cache_unified::slot_info llama_kv_cache_unified::find_slot(const llama_
return { }; return { };
} }
uint32_t n_found = 0;
uint32_t n_tested = 0; uint32_t n_tested = 0;
// for continuous slots, we test that all tokens in the ubatch fit, starting from the current head // for continuous slots, we test that all tokens in the ubatch fit, starting from the current head
@ -857,9 +856,7 @@ llama_kv_cache_unified::slot_info llama_kv_cache_unified::find_slot(const llama_
} }
if (can_use) { if (can_use) {
res.idxs[s][n_found] = idx; res.idxs[s].push_back(idx);
n_found++;
} else { } else {
if (cont) { if (cont) {
break; break;
@ -867,12 +864,12 @@ llama_kv_cache_unified::slot_info llama_kv_cache_unified::find_slot(const llama_
} }
} }
if (n_found == n_tokens) { if (res.idxs[s].size() == n_tokens) {
break; break;
} }
if (cont) { if (cont) {
n_found = 0; res.idxs[s].clear();
} }
if (n_tested >= cells.size()) { if (n_tested >= cells.size()) {
@ -882,7 +879,7 @@ llama_kv_cache_unified::slot_info llama_kv_cache_unified::find_slot(const llama_
} }
// we didn't find a suitable slot - return empty result // we didn't find a suitable slot - return empty result
if (n_found < n_tokens) { if (res.idxs[s].size() < n_tokens) {
return { }; return { };
} }
} }