tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-13 20:07:41 -04:00
Code Issues Packages Projects Releases Wiki Activity

graph : fix stack-use-after-return (#14960)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-07-30 13:52:11 +03:00
committed by GitHub
parent a118d80233
commit 1e15bfd42c
1 changed files with 15 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/llama-graph.h
View File

@@ -144,7 +144,7 @@ public:
ggml_tensor * pos_bucket = nullptr; // I32 [n_batch, n_batch]
const llama_hparams & hparams;
const llama_hparams hparams;
};
class llm_graph_input_pos_bucket_kv : public llm_graph_input_i {
@@ -158,7 +158,7 @@ public:
ggml_tensor * pos_bucket = nullptr; // I32 [n_kv, n_batch]
const llama_hparams & hparams;
const llama_hparams hparams;
const llama_kv_cache_unified_context * mctx;
};
@@ -177,8 +177,8 @@ public:
ggml_tensor * out_ids; // I32 [n_outputs]
const llama_hparams & hparams;
const llama_cparams & cparams;
const llama_hparams hparams;
const llama_cparams cparams;
const uint32_t n_outputs;
};
@@ -192,7 +192,7 @@ public:
ggml_tensor * mean; // F32 [n_batch, n_batch]
const llama_cparams & cparams;
const llama_cparams cparams;
};
class llm_graph_input_cls : public llm_graph_input_i {
@@ -204,7 +204,7 @@ public:
ggml_tensor * cls; // I32 [n_batch]
const llama_cparams & cparams;
const llama_cparams cparams;
};
class llm_graph_input_rs : public llm_graph_input_i {
@@ -247,8 +247,8 @@ public:
ggml_tensor * kq_mask = nullptr; // F32 [n_tokens, n_batch, 1, 1]
ggml_tensor * kq_mask_cnv = nullptr; // [n_tokens, n_batch, 1, 1]
const llama_hparams & hparams;
const llama_cparams & cparams;
const llama_hparams hparams;
const llama_cparams cparams;
};
class llm_graph_input_attn_kv_unified : public llm_graph_input_i {
@@ -278,8 +278,11 @@ public:
ggml_tensor * self_kq_mask = nullptr; // F32 [n_kv, n_batch/n_stream, 1, n_stream]
ggml_tensor * self_kq_mask_cnv = nullptr; // [n_kv, n_batch/n_stream, 1, n_stream]
const llama_hparams & hparams;
const llama_cparams & cparams;
// note: these have to be copies because in order to be able to reuse a graph, its inputs
// need to carry these parameters with them. otherwise, they can point to freed
// llm_graph_params from a previous batch, causing stack-use-after-return
const llama_hparams hparams;
const llama_cparams cparams;
const llama_kv_cache_unified_context * mctx;
};
@@ -318,8 +321,8 @@ public:
ggml_tensor * self_kq_mask_swa = nullptr; // F32 [n_kv, n_batch/n_stream, 1, n_stream]
ggml_tensor * self_kq_mask_swa_cnv = nullptr; // [n_kv, n_batch/n_stream, 1, n_stream]
const llama_hparams & hparams;
const llama_cparams & cparams;
const llama_hparams hparams;
const llama_cparams cparams;
const llama_kv_cache_unified_iswa_context * mctx;
};