tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-19 09:08:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama-context: add ability to get logits (#14672)

Browse Source
This commit is contained in:
Aman Gupta
2025-07-14 21:01:41 +08:00
committed by GitHub
parent 494c5899cb
commit 9c9e4fc635
1 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
src/llama-context.cpp
View File

@ -731,7 +731,8 @@ int llama_context::encode(const llama_batch & batch_inp) {
const auto & hparams = model.hparams;
const int64_t n_embd = hparams.n_embd;
const int64_t n_embd = hparams.n_embd;
const int32_t n_vocab = model.vocab.n_tokens();
// note: during encode, we always pass the full sequence starting from pos = 0
if (!balloc->init(batch_inp, model.vocab, nullptr, n_embd, true)) {
@ -791,10 +792,20 @@ int llama_context::encode(const llama_batch & batch_inp) {
}
}
auto * t_logits = res->get_logits();
auto * t_embd = res->get_embd_pooled() ? res->get_embd_pooled() : res->get_embd();
// extract logits
if (logits && t_logits) {
ggml_backend_t backend_res = ggml_backend_sched_get_tensor_backend(sched.get(), t_logits);
GGML_ASSERT(backend_res != nullptr);
GGML_ASSERT(logits != nullptr);
ggml_backend_tensor_get_async(backend_res, t_logits, logits, 0, n_tokens*n_vocab*sizeof(float));
}
// extract embeddings
if (t_embd) {
if (embd && t_embd) {
ggml_backend_t backend_embd = ggml_backend_sched_get_tensor_backend(sched.get(), t_embd);
GGML_ASSERT(backend_embd != nullptr);