tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 20:45:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

ubatch : new splitting logic (#14217)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-20 10:14:14 +03:00
committed by GitHub
parent 9eaa51e7f0
commit 4c9fdfbe15
19 changed files with 992 additions and 915 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama-graph.h
View File

@ -95,14 +95,14 @@ public:
class llm_graph_input_pos : public llm_graph_input_i {
public:
llm_graph_input_pos(int64_t n_pos_per_embd) : n_pos_per_embd(n_pos_per_embd) {}
llm_graph_input_pos(uint32_t n_pos_per_embd) : n_pos_per_embd(n_pos_per_embd) {}
virtual ~llm_graph_input_pos() = default;
void set_input(const llama_ubatch * ubatch) override;
ggml_tensor * pos = nullptr; // I32 [n_batch]
const int64_t n_pos_per_embd = 1;
const uint32_t n_pos_per_embd = 1;
};
// temperature tuning, used by llama4
@ -464,8 +464,6 @@ struct llm_graph_context {
llm_graph_context(const llm_graph_params & params);
int64_t n_pos_per_embd() const;
void cb(ggml_tensor * cur, const char * name, int il) const;
//