ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

llama : use n_swa + n_ubatch cells for SWA cache (#13833)

Browse source 
* llama : use n_swa + n_ubatch cells for SWA cache

ggml-ci

* llama : add warning about multi-sqeuence SWA contexts
This commit is contained in:
Georgi Gerganov 2025-05-31 15:57:44 +03:00 • committed by GitHub
parent c7e0a2054b
commit 3600cc2886
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 24 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

5
src/llama-context.cpp
View file

@ -123,6 +123,11 @@ llama_context::llama_context(
__func__, n_ctx_per_seq, hparams.n_ctx_train);
}
if (!params.swa_full && cparams.n_seq_max > 1) {
LLAMA_LOG_WARN("%s: requested n_seq_max (%u) > 1, but swa_full is not enabled -- performance may be degraded: %s\n",
__func__, cparams.n_seq_max, "https://github.com/ggml-org/llama.cpp/pull/13845#issuecomment-2924800573");
}
if (!hparams.vocab_only) {
// GPU backends
for (auto * dev : model.devices) {

4
src/llama-kv-cache.cpp
View file

@ -1731,14 +1731,14 @@ llama_kv_cache_unified_iswa::llama_kv_cache_unified_iswa(
bool swa_full,
uint32_t kv_size,
uint32_t n_seq_max,
uint32_t n_batch,
uint32_t n_ubatch,
uint32_t n_pad) : hparams(model.hparams) {
llama_kv_cache_unified::layer_filter_cb filter_base = [&](int32_t il) { return !model.hparams.is_swa(il); };
llama_kv_cache_unified::layer_filter_cb filter_swa = [&](int32_t il) { return model.hparams.is_swa(il); };
const uint32_t size_base = kv_size;
uint32_t size_swa = std::min(size_base, GGML_PAD(hparams.n_swa*n_seq_max + n_batch, n_pad));
uint32_t size_swa = std::min(size_base, GGML_PAD(hparams.n_swa*n_seq_max + n_ubatch, n_pad));
// when using full-size SWA cache, we set the SWA cache size to be equal to the base cache size
if (swa_full) {

2
src/llama-kv-cache.h
View file

@ -339,7 +339,7 @@ public:
bool swa_full,
uint32_t kv_size,
uint32_t n_seq_max,
uint32_t n_batch,
uint32_t n_ubatch,
uint32_t n_pad);
~llama_kv_cache_unified_iswa() = default;

6
src/llama-model.cpp
View file

@ -13230,7 +13230,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
params.swa_full,
cparams.n_ctx,
cparams.n_seq_max,
cparams.n_batch,
cparams.n_ubatch,
padding);
} else {
GGML_ASSERT(!hparams.is_swa_any());
@ -13593,6 +13593,10 @@ int32_t llama_model_n_head_kv(const llama_model * model) {
return model->hparams.n_head_kv();
}
int32_t llama_model_n_swa(const llama_model * model) {
return model->hparams.n_swa;
}
// deprecated
int32_t llama_n_ctx_train(const llama_model * model) {
return llama_model_n_ctx_train(model);