tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 20:25:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

memory : rename interface to llama_memory_context_i (#14296)

Browse Source 
* memory : rename interface to llama_memory_context_i

ggml-ci

* cont : fix comments

* cont : use "mctx" for referencing a memory context

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-21 08:03:46 +03:00
committed by GitHub
parent b23fa0b3f4
commit 692e3cdd0a
14 changed files with 339 additions and 341 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
src/llama-kv-cache-unified.cpp
View File

@ -307,7 +307,7 @@ llama_pos llama_kv_cache_unified::seq_pos_max(llama_seq_id seq_id) const {
return cells.seq_pos_max(seq_id);
}
llama_memory_state_ptr llama_kv_cache_unified::init_batch(
llama_memory_context_ptr llama_kv_cache_unified::init_batch(
llama_batch_allocr & balloc,
uint32_t n_ubatch,
bool embd_all) {
@ -332,18 +332,18 @@ llama_memory_state_ptr llama_kv_cache_unified::init_batch(
break;
}
return std::make_unique<llama_kv_cache_unified_state>(
return std::make_unique<llama_kv_cache_unified_context>(
this, std::move(heads), std::move(ubatches));
} while (false);
return std::make_unique<llama_kv_cache_unified_state>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
return std::make_unique<llama_kv_cache_unified_context>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
}
llama_memory_state_ptr llama_kv_cache_unified::init_full() {
return std::make_unique<llama_kv_cache_unified_state>(this);
llama_memory_context_ptr llama_kv_cache_unified::init_full() {
return std::make_unique<llama_kv_cache_unified_context>(this);
}
llama_memory_state_ptr llama_kv_cache_unified::init_update(llama_context * lctx, bool optimize) {
llama_memory_context_ptr llama_kv_cache_unified::init_update(llama_context * lctx, bool optimize) {
bool do_shift = get_has_shift();
defrag_info dinfo;
@ -373,7 +373,7 @@ llama_memory_state_ptr llama_kv_cache_unified::init_update(llama_context * lctx,
}
}
return std::make_unique<llama_kv_cache_unified_state>(this, lctx, do_shift, std::move(dinfo));
return std::make_unique<llama_kv_cache_unified_context>(this, lctx, do_shift, std::move(dinfo));
}
llama_kv_cache_unified::ubatch_heads llama_kv_cache_unified::prepare(const std::vector<llama_ubatch> & ubatches) {
@ -1710,18 +1710,18 @@ bool llama_kv_cache_unified::state_read_data(llama_io_read_i & io, uint32_t cell
}
//
// llama_kv_cache_unified_state
// llama_kv_cache_unified_context
//
llama_kv_cache_unified_state::llama_kv_cache_unified_state(llama_memory_status status) : status(status) {}
llama_kv_cache_unified_context::llama_kv_cache_unified_context(llama_memory_status status) : status(status) {}
llama_kv_cache_unified_state::llama_kv_cache_unified_state(
llama_kv_cache_unified_context::llama_kv_cache_unified_context(
llama_kv_cache_unified * kv) : status(LLAMA_MEMORY_STATUS_SUCCESS), kv(kv) {
n_kv = kv->get_size();
head = 0;
}
llama_kv_cache_unified_state::llama_kv_cache_unified_state(
llama_kv_cache_unified_context::llama_kv_cache_unified_context(
llama_kv_cache_unified * kv,
llama_context * lctx,
bool do_shift,
@ -1731,15 +1731,15 @@ llama_kv_cache_unified_state::llama_kv_cache_unified_state(
}
}
llama_kv_cache_unified_state::llama_kv_cache_unified_state(
llama_kv_cache_unified_context::llama_kv_cache_unified_context(
llama_kv_cache_unified * kv,
llama_kv_cache_unified::ubatch_heads heads,
std::vector<llama_ubatch> ubatches) : status(LLAMA_MEMORY_STATUS_SUCCESS), kv(kv), heads(std::move(heads)), ubatches(std::move(ubatches)) {
}
llama_kv_cache_unified_state::~llama_kv_cache_unified_state() = default;
llama_kv_cache_unified_context::~llama_kv_cache_unified_context() = default;
bool llama_kv_cache_unified_state::next() {
bool llama_kv_cache_unified_context::next() {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
if (++i_next >= ubatches.size()) {
@ -1749,7 +1749,7 @@ bool llama_kv_cache_unified_state::next() {
return true;
}
bool llama_kv_cache_unified_state::apply() {
bool llama_kv_cache_unified_context::apply() {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
// no ubatches -> this is a KV cache update
@ -1767,45 +1767,45 @@ bool llama_kv_cache_unified_state::apply() {
return true;
}
llama_memory_status llama_kv_cache_unified_state::get_status() const {
llama_memory_status llama_kv_cache_unified_context::get_status() const {
return status;
}
const llama_ubatch & llama_kv_cache_unified_state::get_ubatch() const {
const llama_ubatch & llama_kv_cache_unified_context::get_ubatch() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return ubatches[i_next];
}
uint32_t llama_kv_cache_unified_state::get_n_kv() const {
uint32_t llama_kv_cache_unified_context::get_n_kv() const {
return n_kv;
}
ggml_tensor * llama_kv_cache_unified_state::get_k(ggml_context * ctx, int32_t il) const {
ggml_tensor * llama_kv_cache_unified_context::get_k(ggml_context * ctx, int32_t il) const {
return kv->get_k(ctx, il, n_kv);
}
ggml_tensor * llama_kv_cache_unified_state::get_v(ggml_context * ctx, int32_t il) const {
ggml_tensor * llama_kv_cache_unified_context::get_v(ggml_context * ctx, int32_t il) const {
return kv->get_v(ctx, il, n_kv);
}
ggml_tensor * llama_kv_cache_unified_state::cpy_k(ggml_context * ctx, ggml_tensor * k_cur, int32_t il) const {
ggml_tensor * llama_kv_cache_unified_context::cpy_k(ggml_context * ctx, ggml_tensor * k_cur, int32_t il) const {
return kv->cpy_k(ctx, k_cur, il, head);
}
ggml_tensor * llama_kv_cache_unified_state::cpy_v(ggml_context * ctx, ggml_tensor * v_cur, int32_t il) const {
ggml_tensor * llama_kv_cache_unified_context::cpy_v(ggml_context * ctx, ggml_tensor * v_cur, int32_t il) const {
return kv->cpy_v(ctx, v_cur, il, head);
}
void llama_kv_cache_unified_state::set_input_k_shift(ggml_tensor * dst) const {
void llama_kv_cache_unified_context::set_input_k_shift(ggml_tensor * dst) const {
kv->set_input_k_shift(dst);
}
void llama_kv_cache_unified_state::set_input_kq_mask(ggml_tensor * dst, const llama_ubatch * ubatch, bool causal_attn) const {
void llama_kv_cache_unified_context::set_input_kq_mask(ggml_tensor * dst, const llama_ubatch * ubatch, bool causal_attn) const {
kv->set_input_kq_mask(dst, ubatch, causal_attn);
}
void llama_kv_cache_unified_state::set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const {
void llama_kv_cache_unified_context::set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const {
kv->set_input_pos_bucket(dst, ubatch);
}