thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

context : fix output reorder with backend sampling (#19638)

Browse source
This commit is contained in:
Georgi Gerganov 2026-02-15 14:57:40 +02:00 • committed by GitHub
parent 08e6d914b8
commit 341bc7d23c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 30 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/llama-context.h
View file

@ -265,24 +265,26 @@ private:
std::unique_ptr<llama_memory_i> memory;
// decode output (2-dimensional array: [n_outputs][n_vocab])
struct buffer_view<float> logits = {nullptr, 0};
buffer_view<float> logits = {nullptr, 0};
// embeddings output (2-dimensional array: [n_outputs][n_embd])
// populated only when pooling_type == LLAMA_POOLING_TYPE_NONE
struct buffer_view<float> embd = {nullptr, 0};
buffer_view<float> embd = {nullptr, 0};
struct sampling_info {
// !samplers.empty() to check if any samplers are active
std::map<llama_seq_id, llama_sampler *> samplers;
struct buffer_view<float> logits = {nullptr, 0};
struct buffer_view<llama_token> sampled = {nullptr, 0};
struct buffer_view<float> probs = {nullptr, 0};
struct buffer_view<llama_token> candidates = {nullptr, 0};
buffer_view<float> logits = {nullptr, 0};
buffer_view<llama_token> sampled = {nullptr, 0};
buffer_view<float> probs = {nullptr, 0};
buffer_view<llama_token> candidates = {nullptr, 0};
std::vector<uint32_t> logits_count;
std::vector<uint32_t> probs_count;
std::vector<uint32_t> candidates_count;
// optimization
std::vector<llama_token> token_ids_full_vocab;
};