happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

squash! sampling : simplify backend sampling logic decode 

Fix condition to check if backend actually sampled tokens, not just that
backend samplers are available.
This commit is contained in:
Daniel Bevenius 2025-11-19 11:29:26 +01:00
parent 7e98ebcc6b
commit d74eb61aa7
No known key found for this signature in database
1 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/llama-context.cpp
View File

@ -1361,7 +1361,9 @@ int llama_context::decode(const llama_batch & batch_inp) {
// ggml_graph_dump_dot(gf, NULL, "llama.dot");
//}
if (has_backend_samplers) {
const bool backend_has_sampled = !res->t_sampled_tokens.empty() || !res->t_sampled_probs.empty() || !res->t_sampled_logits.empty();
if (has_backend_samplers && backend_has_sampled) {
const auto seq_to_batch_idx = build_seq_to_batch_idx(ubatch);
// If a backend sampler has sampled a token we only want to copy the
@ -1381,7 +1383,9 @@ int llama_context::decode(const llama_batch & batch_inp) {
// 2) CPU samplers to associate filtered logits with their token ids.
copy_tensor_async_token_ids(res->t_sampled_token_ids, sampled_token_ids_map, seq_to_batch_idx, sched.get());
} else {
}
if (!backend_has_sampled) {
auto * t_logits = res->get_logits();
auto * t_embd = cparams.embeddings ? res->get_embd() : nullptr;