CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +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

70
src/llama-kv-cache-unified-iswa.cpp
View File

@@ -95,7 +95,7 @@ llama_pos llama_kv_cache_unified_iswa::seq_pos_max(llama_seq_id seq_id) const {
return kv_swa->seq_pos_max(seq_id);
}
llama_memory_state_ptr llama_kv_cache_unified_iswa::init_batch(llama_batch_allocr & balloc, uint32_t n_ubatch, bool embd_all) {
llama_memory_context_ptr llama_kv_cache_unified_iswa::init_batch(llama_batch_allocr & balloc, uint32_t n_ubatch, bool embd_all) {
GGML_UNUSED(embd_all);
// first try simple split
@@ -125,7 +125,7 @@ llama_memory_state_ptr llama_kv_cache_unified_iswa::init_batch(llama_batch_alloc
assert(heads_base.size() == heads_swa.size());
return std::make_unique<llama_kv_cache_unified_iswa_state>(
return std::make_unique<llama_kv_cache_unified_iswa_context>(
this, std::move(heads_base), std::move(heads_swa), std::move(ubatches));
} while (false);
@@ -156,22 +156,22 @@ llama_memory_state_ptr llama_kv_cache_unified_iswa::init_batch(llama_batch_alloc
assert(heads_base.size() == heads_swa.size());
return std::make_unique<llama_kv_cache_unified_iswa_state>(
return std::make_unique<llama_kv_cache_unified_iswa_context>(
this, std::move(heads_base), std::move(heads_swa), std::move(ubatches));
} while (false);
// TODO: if we fail again, we should attempt different splitting strategies
// but to do that properly, we first have to refactor the batches to be more flexible
return std::make_unique<llama_kv_cache_unified_iswa_state>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
return std::make_unique<llama_kv_cache_unified_iswa_context>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
}
llama_memory_state_ptr llama_kv_cache_unified_iswa::init_full() {
return std::make_unique<llama_kv_cache_unified_iswa_state>(this);
llama_memory_context_ptr llama_kv_cache_unified_iswa::init_full() {
return std::make_unique<llama_kv_cache_unified_iswa_context>(this);
}
llama_memory_state_ptr llama_kv_cache_unified_iswa::init_update(llama_context * lctx, bool optimize) {
return std::make_unique<llama_kv_cache_unified_iswa_state>(this, lctx, optimize);
llama_memory_context_ptr llama_kv_cache_unified_iswa::init_update(llama_context * lctx, bool optimize) {
return std::make_unique<llama_kv_cache_unified_iswa_context>(this, lctx, optimize);
}
bool llama_kv_cache_unified_iswa::get_can_shift() const {
@@ -197,46 +197,46 @@ llama_kv_cache_unified * llama_kv_cache_unified_iswa::get_swa() const {
}
//
// llama_kv_cache_unified_iswa_state
// llama_kv_cache_unified_iswa_context
//
llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(llama_memory_status status) : status(status) {}
llama_kv_cache_unified_iswa_context::llama_kv_cache_unified_iswa_context(llama_memory_status status) : status(status) {}
llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
llama_kv_cache_unified_iswa_context::llama_kv_cache_unified_iswa_context(
llama_kv_cache_unified_iswa * kv) :
state_base(kv->get_base()->init_full()),
state_swa (kv->get_swa ()->init_full()),
status(llama_memory_status_combine(state_base->get_status(), state_swa->get_status())) {
ctx_base(kv->get_base()->init_full()),
ctx_swa (kv->get_swa ()->init_full()),
status(llama_memory_status_combine(ctx_base->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
llama_kv_cache_unified_iswa_context::llama_kv_cache_unified_iswa_context(
llama_kv_cache_unified_iswa * kv,
llama_context * lctx,
bool optimize) :
state_base(kv->get_base()->init_update(lctx, optimize)),
state_swa (kv->get_swa ()->init_update(lctx, optimize)),
status(llama_memory_status_combine(state_base->get_status(), state_swa->get_status())) {
ctx_base(kv->get_base()->init_update(lctx, optimize)),
ctx_swa (kv->get_swa ()->init_update(lctx, optimize)),
status(llama_memory_status_combine(ctx_base->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
llama_kv_cache_unified_iswa_context::llama_kv_cache_unified_iswa_context(
llama_kv_cache_unified_iswa * kv,
std::vector<uint32_t> heads_base,
std::vector<uint32_t> heads_swa,
std::vector<llama_ubatch> ubatches) :
ubatches(std::move(ubatches)),
// note: here we copy the ubatches. not sure if this is ideal
state_base(new llama_kv_cache_unified_state(kv->get_base(), std::move(heads_base), this->ubatches)),
state_swa (new llama_kv_cache_unified_state(kv->get_swa (), std::move(heads_swa), this->ubatches)),
status(llama_memory_status_combine(state_base->get_status(), state_swa->get_status())) {
ctx_base(new llama_kv_cache_unified_context(kv->get_base(), std::move(heads_base), this->ubatches)),
ctx_swa (new llama_kv_cache_unified_context(kv->get_swa (), std::move(heads_swa), this->ubatches)),
status(llama_memory_status_combine(ctx_base->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_unified_iswa_state:: ~llama_kv_cache_unified_iswa_state() = default;
llama_kv_cache_unified_iswa_context:: ~llama_kv_cache_unified_iswa_context() = default;
bool llama_kv_cache_unified_iswa_state::next() {
bool llama_kv_cache_unified_iswa_context::next() {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
state_base->next();
state_swa ->next();
ctx_base->next();
ctx_swa ->next();
if (++i_next >= ubatches.size()) {
return false;
@@ -245,35 +245,35 @@ bool llama_kv_cache_unified_iswa_state::next() {
return true;
}
bool llama_kv_cache_unified_iswa_state::apply() {
bool llama_kv_cache_unified_iswa_context::apply() {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
bool res = true;
res = res & state_base->apply();
res = res & state_swa ->apply();
res = res & ctx_base->apply();
res = res & ctx_swa ->apply();
return res;
}
llama_memory_status llama_kv_cache_unified_iswa_state::get_status() const {
llama_memory_status llama_kv_cache_unified_iswa_context::get_status() const {
return status;
}
const llama_ubatch & llama_kv_cache_unified_iswa_state::get_ubatch() const {
const llama_ubatch & llama_kv_cache_unified_iswa_context::get_ubatch() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return ubatches[i_next];
}
const llama_kv_cache_unified_state * llama_kv_cache_unified_iswa_state::get_base() const {
const llama_kv_cache_unified_context * llama_kv_cache_unified_iswa_context::get_base() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return static_cast<const llama_kv_cache_unified_state *>(state_base.get());
return static_cast<const llama_kv_cache_unified_context *>(ctx_base.get());
}
const llama_kv_cache_unified_state * llama_kv_cache_unified_iswa_state::get_swa() const {
const llama_kv_cache_unified_context * llama_kv_cache_unified_iswa_context::get_swa() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return static_cast<const llama_kv_cache_unified_state *>(state_swa.get());
return static_cast<const llama_kv_cache_unified_context *>(ctx_swa.get());
}