CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : remove obsolete map for layer counting

Browse Source
This commit is contained in:
Georgi Gerganov
2023-10-29 13:39:04 +02:00
parent 79ad734417
commit 210e6e5d02
1 changed files with 1 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
llama.cpp
View File

@@ -5469,9 +5469,6 @@ static struct ggml_cgraph * llama_build_graph(
// check if we should build the worst-case graph (for memory measurement) // check if we should build the worst-case graph (for memory measurement)
const bool worst_case = ggml_allocr_is_measure(lctx.alloc); const bool worst_case = ggml_allocr_is_measure(lctx.alloc);
// count the number of times a tensor with a given name has been offloaded
std::unordered_map<std::string, int> offload_n;
// keep track of the input that has already been allocated // keep track of the input that has already been allocated
bool alloc_inp_tokens = false; bool alloc_inp_tokens = false;
bool alloc_inp_embd = false; bool alloc_inp_embd = false;
@@ -5654,7 +5651,7 @@ static struct ggml_cgraph * llama_build_graph(
break; break;
case OFFLOAD_FUNC: case OFFLOAD_FUNC:
if (n_gpu_layers < n_layer) { if (n_gpu_layers < n_layer) {
if (offload_n[name]++ < i_gpu_start) { if (il < i_gpu_start) {
func_e = OFFLOAD_FUNC_NOP; func_e = OFFLOAD_FUNC_NOP;
} }
} }