extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-08-09 15:56:13 +02:00
Code Issues Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-01 14:07:36 +03:00
parent 3f46282cbe
commit 7fd6fa8097
22 changed files with 4265 additions and 3552 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
examples/talk-llama/llama-context.h
View File

@ -18,6 +18,9 @@ struct llama_kv_cache;
class llama_io_read_i;
class llama_io_write_i;
class llama_memory_i;
class llama_memory_state_i;
struct llama_context {
// init scheduler and compute buffers, reserve worst-case graphs
llama_context(
@ -47,7 +50,9 @@ struct llama_context {
llama_kv_cache * get_kv_self();
const llama_kv_cache * get_kv_self() const;
void kv_self_update();
// return true of the KV cache was updated
// TODO: remove
bool kv_self_update();
enum llama_pooling_type pooling_type() const;
@ -88,6 +93,16 @@ struct llama_context {
int32_t il_start,
int32_t il_end);
// process a single ubatch with a specific graph type
// if memory_state is provided, it will be applied first to the context's memory
// ret contains the status of the graph computation
// returns nullptr only if ret != GGML_STATUS_SUCCESS
llm_graph_result_ptr process_ubatch(
const llama_ubatch & ubatch,
llm_graph_type gtype,
llama_memory_state_i * mstate,
ggml_status & ret);
int encode(llama_batch & inp_batch);
int decode(llama_batch & inp_batch);
@ -180,16 +195,18 @@ public:
ggml_cgraph * graph_init();
// returns the result of ggml_backend_sched_graph_compute_async execution
ggml_status graph_compute(
ggml_cgraph * gf,
bool batched);
ggml_status graph_compute(ggml_cgraph * gf, bool batched);
// reserve a graph with a dummy ubatch of the specified size
ggml_cgraph * graph_reserve(uint32_t n_tokens, uint32_t n_seqs, uint32_t n_outputs, const llama_memory_state_i * mstate);
private:
llm_graph_result_ptr graph_build(
ggml_context * ctx,
ggml_cgraph * gf,
const llama_ubatch & ubatch,
llm_graph_type gtype);
ggml_context * ctx,
ggml_cgraph * gf,
const llama_ubatch & ubatch,
llm_graph_type gtype,
const llama_memory_state_i * mstate);
llm_graph_cb graph_get_cb() const;