CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

memory : handle kv_unified for hybrid models (#15050)

Browse Source
This commit is contained in:
compilade
2025-08-03 15:43:07 -04:00
committed by GitHub
parent 97366dc6ab
commit 11a3811164
3 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/llama-memory-hybrid.cpp
View File

@@ -25,6 +25,7 @@ llama_memory_hybrid::llama_memory_hybrid(
/* common */
uint32_t n_seq_max,
bool offload,
bool unified,
/* layer filters */
layer_filter_cb && filter_attn,
layer_filter_cb && filter_recr) :
@@ -38,7 +39,7 @@ llama_memory_hybrid::llama_memory_hybrid(
type_v,
v_trans,
offload,
1,
unified,
kv_size,
n_seq_max,
n_pad,

1
src/llama-memory-hybrid.h
View File

@@ -39,6 +39,7 @@ public:
/* common */
uint32_t n_seq_max,
bool offload,
bool unified,
/* layer filters */
layer_filter_cb && filter_attn = nullptr,
layer_filter_cb && filter_recr = nullptr);

1
src/llama-model.cpp
View File

@@ -17598,6 +17598,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
/* recurrent_kv_size */ std::max((uint32_t) 1, cparams.n_seq_max),
/* n_seq_max */ cparams.n_seq_max,
/* offload */ cparams.offload_kqv,
/* unified */ cparams.kv_unified,
/* filter_attn */ (arch == LLM_ARCH_FALCON_H1) ? [&](int32_t) { return true; } : (llama_memory_hybrid::layer_filter_cb)nullptr,
/* filter_recr */ (arch == LLM_ARCH_FALCON_H1) ? [&](int32_t) { return true; } : (llama_memory_hybrid::layer_filter_cb)nullptr);
} else {