extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-08-09 02:54:46 +02:00
Code Issues Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

Browse Source
This commit is contained in:
Georgi Gerganov
2025-07-01 12:21:09 +03:00
parent c4ea72be9a
commit 1f816de7da
24 changed files with 1456 additions and 430 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
examples/talk-llama/llama-memory-hybrid.h
View File

@ -49,14 +49,14 @@ public:
// llama_memory_i
//
llama_memory_state_ptr init_batch(
llama_memory_context_ptr init_batch(
llama_batch_allocr & balloc,
uint32_t n_ubatch,
bool embd_all) override;
llama_memory_state_ptr init_full() override;
llama_memory_context_ptr init_full() override;
llama_memory_state_ptr init_update(llama_context * lctx, bool optimize) override;
llama_memory_context_ptr init_update(llama_context * lctx, bool optimize) override;
bool get_can_shift() const override;
@ -90,27 +90,27 @@ private:
const std::unique_ptr<llama_memory_recurrent> mem_recr;
};
class llama_memory_hybrid_state : public llama_memory_state_i {
class llama_memory_hybrid_context : public llama_memory_context_i {
public:
// init failure
explicit llama_memory_hybrid_state(llama_memory_status status);
explicit llama_memory_hybrid_context(llama_memory_status status);
// init full
explicit llama_memory_hybrid_state(llama_memory_hybrid * mem);
explicit llama_memory_hybrid_context(llama_memory_hybrid * mem);
// init update
explicit llama_memory_hybrid_state(
explicit llama_memory_hybrid_context(
llama_memory_hybrid * mem,
llama_context * lctx,
bool optimize);
// init success
llama_memory_hybrid_state(
llama_memory_hybrid_context(
llama_memory_hybrid * mem,
std::vector<uint32_t> heads_attn,
std::vector<llama_ubatch> ubatches);
~llama_memory_hybrid_state() = default;
~llama_memory_hybrid_context() = default;
bool next() override;
bool apply() override;
@ -119,11 +119,11 @@ public:
const llama_ubatch & get_ubatch() const override;
//
// llama_memory_hybrid_state
// llama_memory_hybrid_context
//
const llama_kv_cache_unified_state * get_state_attn() const;
const llama_memory_recurrent_state * get_state_recr() const;
const llama_kv_cache_unified_context * get_attn() const;
const llama_memory_recurrent_context * get_recr() const;
private:
// the index of the next ubatch to process
@ -131,8 +131,8 @@ private:
std::vector<llama_ubatch> ubatches;
const llama_memory_state_ptr state_attn;
const llama_memory_state_ptr state_recr;
const llama_memory_context_ptr ctx_attn;
const llama_memory_context_ptr ctx_recr;
const llama_memory_status status;
};