Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-26 12:21:40 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : fix t5 segfault

Browse Source
This commit is contained in:
Francis Couture-Harpin 2024-07-17 15:36:56 -04:00
parent 1fb5d4fdee
commit 1725de768e
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama.cpp
View File

@ -15377,11 +15377,13 @@ static int llama_encode_internal(
ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float)); ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float));
GGML_ASSERT(!ubatch.equal_seqs); // TODO: use batch.n_seqs instead of failing
// remember the sequence ids used during the encoding - needed for cross attention later // remember the sequence ids used during the encoding - needed for cross attention later
lctx.seq_ids_enc.resize(n_tokens); lctx.seq_ids_enc.resize(n_tokens);
for (uint32_t i = 0; i < n_tokens; i++) { for (uint32_t i = 0; i < n_tokens; i++) {
for (int s = 0; s < batch.n_seq_id[i]; s++) { for (int s = 0; s < ubatch.n_seq_id[i]; s++) {
llama_seq_id seq_id = batch.seq_id[i][s]; llama_seq_id seq_id = ubatch.seq_id[i][s];
lctx.seq_ids_enc[i].insert(seq_id); lctx.seq_ids_enc[i].insert(seq_id);
} }
} }