tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-06 10:13:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

mamba : fix mismatched new and delete size for llm_build_mamba

Browse Source 
Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON
This commit is contained in:
Francis Couture-Harpin
2025-06-26 17:52:28 -04:00
parent afdb669206
commit dc1d109da8
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/llama-model.cpp
View File

@ -9198,9 +9198,7 @@ struct llm_build_starcoder2 : public llm_graph_context {
};
struct llm_build_mamba : public llm_graph_context {
const llama_model & model;
llm_build_mamba(const llama_model & model, const llm_graph_params & params, ggml_cgraph * gf) : llm_graph_context(params), model(model) {
llm_build_mamba(const llama_model & model, const llm_graph_params & params, ggml_cgraph * gf) : llm_graph_context(params) {
ggml_tensor * cur;
ggml_tensor * inpL;
@ -9219,9 +9217,9 @@ struct llm_build_mamba : public llm_graph_context {
cb(cur, "attn_norm", il);
if (model.arch == LLM_ARCH_MAMBA2) {
cur = build_mamba2_layer(rs_inp, gf, cur, ubatch, il);
cur = build_mamba2_layer(rs_inp, gf, cur, model, ubatch, il);
} else {
cur = build_mamba_layer(rs_inp, gf, cur, ubatch, il);
cur = build_mamba_layer(rs_inp, gf, cur, model, ubatch, il);
}
if (il == n_layer - 1 && inp_out_ids) {
@ -9260,6 +9258,7 @@ struct llm_build_mamba : public llm_graph_context {
llm_graph_input_rs * inp,
ggml_cgraph * gf,
ggml_tensor * cur,
const llama_model & model,
const llama_ubatch & ubatch,
int il) const {
const auto * mctx_cur = static_cast<const llama_memory_recurrent_context *>(mctx);
@ -9398,6 +9397,7 @@ struct llm_build_mamba : public llm_graph_context {
llm_graph_input_rs * inp,
ggml_cgraph * gf,
ggml_tensor * cur,
const llama_model & model,
const llama_ubatch & ubatch,
int il) const {
const auto * mctx_cur = static_cast<const llama_memory_recurrent_context *>(mctx);