tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-29 05:33:37 -04:00
Code Issues Packages Projects Releases Wiki Activity

batch : auto-gen positions + verify multi-sequence input (#14177)

Browse Source 
* batch : verify multi-sequence input batches

ggml-ci

* cont : auto-gen positions + verify multi-seq input

ggml-ci

* cont : first print debug info, then perform validation

ggml-ci

* cont : fix position auto-gen + add comments

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-15 09:18:37 +03:00
committed by GitHub
parent 00ba772610
commit b9912ac570
5 changed files with 155 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama-context.cpp
View File

@@ -727,9 +727,8 @@ int llama_context::encode(const llama_batch & batch_inp) {
return -1;
}
// temporary allocate memory for the input batch if needed
// note: during encode, we always pass the full sequence starting from pos = 0
if (!batch_allocr->init(batch_inp, model.vocab, batch_inp.pos ? -1 : 0)) {
if (!batch_allocr->init(batch_inp, model.vocab, nullptr)) {
LLAMA_LOG_ERROR("%s: failed to initialize batch\n", __func__);
return -1;
}
@@ -895,8 +894,7 @@ int llama_context::decode(const llama_batch & batch_inp) {
return -1;
}
// temporary allocate memory for the input batch if needed
if (!batch_allocr->init(batch_inp, model.vocab, batch_inp.pos ? -1 : memory->seq_pos_max(0) + 1)) {
if (!batch_allocr->init(batch_inp, model.vocab, memory.get())) {
LLAMA_LOG_ERROR("%s: failed to initialize batch\n", __func__);
return -1;
}