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

llama : model-based max number of graph nodes

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-07-22 09:50:57 +03:00
parent 45f2c19cc5
commit 7e27c17572
No known key found for this signature in database
GPG key ID: BF970631944C16B7
1 changed files with 53 additions and 43 deletions
Download patch file Download diff file Expand all files Collapse all files

96
src/llama.cpp
View file

@ -113,7 +113,6 @@
#endif #endif
// bump if necessary // bump if necessary
#define LLAMA_MAX_NODES 8192
#define LLAMA_MAX_LAYERS 512 #define LLAMA_MAX_LAYERS 512
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2 #define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
@ -3657,6 +3656,15 @@ namespace GGUFMeta {
using llama_buf_map = std::unordered_map<uint32_t, ggml_backend_buffer_t>; using llama_buf_map = std::unordered_map<uint32_t, ggml_backend_buffer_t>;
// TODO: update when needed or think of some clever automatic way to do this
static size_t llama_model_max_nodes(const llama_model & model) {
if (model.arch == LLM_ARCH_LLAMA && model.hparams.n_layer > 400) { // llama-3 405B
return 32768;
}
return 8192;
}
struct llama_model_loader { struct llama_model_loader {
int n_kv = 0; int n_kv = 0;
int n_tensors = 0; int n_tensors = 0;
@ -8495,7 +8503,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_k_shift() { struct ggml_cgraph * build_k_shift() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
GGML_ASSERT(kv_self.size == n_ctx); GGML_ASSERT(kv_self.size == n_ctx);
@ -8526,7 +8534,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_s_copy() { struct ggml_cgraph * build_s_copy() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
GGML_ASSERT(kv_self.recurrent); GGML_ASSERT(kv_self.recurrent);
@ -8549,7 +8557,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_defrag(const std::vector<uint32_t> & ids) { struct ggml_cgraph * build_defrag(const std::vector<uint32_t> & ids) {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
for (uint32_t i = 0; i < ids.size(); ++i) { for (uint32_t i = 0; i < ids.size(); ++i) {
const uint32_t id = ids[i]; const uint32_t id = ids[i];
@ -8790,7 +8798,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_llama() { struct ggml_cgraph * build_llama() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -8933,7 +8941,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_baichuan() { struct ggml_cgraph * build_baichuan() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -9048,7 +9056,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_xverse() { struct ggml_cgraph * build_xverse() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -9151,7 +9159,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_falcon() { struct ggml_cgraph * build_falcon() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -9271,7 +9279,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_grok() { struct ggml_cgraph * build_grok() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -9428,7 +9436,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_dbrx() { struct ggml_cgraph * build_dbrx() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -9554,7 +9562,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_starcoder() { struct ggml_cgraph * build_starcoder() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -9658,7 +9666,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_refact() { struct ggml_cgraph * build_refact() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -9752,7 +9760,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_bert() { struct ggml_cgraph * build_bert() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -9946,7 +9954,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_bloom() { struct ggml_cgraph * build_bloom() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -10047,7 +10055,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_mpt() { struct ggml_cgraph * build_mpt() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -10337,7 +10345,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_qwen() { struct ggml_cgraph * build_qwen() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -10449,7 +10457,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_qwen2() { struct ggml_cgraph * build_qwen2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -10561,7 +10569,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_qwen2moe() { struct ggml_cgraph * build_qwen2moe() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -10707,7 +10715,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_phi2() { struct ggml_cgraph * build_phi2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -10828,7 +10836,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_phi3() { struct ggml_cgraph * build_phi3() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -11060,7 +11068,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_gpt2() { struct ggml_cgraph * build_gpt2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -11165,7 +11173,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_codeshell() { struct ggml_cgraph * build_codeshell() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -11276,7 +11284,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_orion() { struct ggml_cgraph * build_orion() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -11394,7 +11402,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_internlm2() { struct ggml_cgraph * build_internlm2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -11515,7 +11523,7 @@ struct llm_build_context {
// https://github.com/ggerganov/llama.cpp/issues/5276#issuecomment-1925774738 // https://github.com/ggerganov/llama.cpp/issues/5276#issuecomment-1925774738
// based on the original build_llama() function // based on the original build_llama() function
struct ggml_cgraph * build_minicpm() { struct ggml_cgraph * build_minicpm() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -11659,7 +11667,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_gemma() { struct ggml_cgraph * build_gemma() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head_k = hparams.n_embd_head_k; const int64_t n_embd_head_k = hparams.n_embd_head_k;
@ -11767,7 +11775,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_gemma2() { struct ggml_cgraph * build_gemma2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head_k = hparams.n_embd_head_k; const int64_t n_embd_head_k = hparams.n_embd_head_k;
@ -11902,7 +11910,7 @@ struct llm_build_context {
struct ggml_cgraph * build_starcoder2() { struct ggml_cgraph * build_starcoder2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -12021,7 +12029,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_mamba() { struct ggml_cgraph * build_mamba() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t d_model = n_embd; const int64_t d_model = n_embd;
const int64_t d_conv = hparams.ssm_d_conv; const int64_t d_conv = hparams.ssm_d_conv;
@ -12170,7 +12178,7 @@ struct llm_build_context {
struct ggml_cgraph * build_command_r() { struct ggml_cgraph * build_command_r() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -12324,7 +12332,7 @@ struct llm_build_context {
// * removed bias // * removed bias
// * removed MoE // * removed MoE
struct ggml_cgraph * build_olmo() { struct ggml_cgraph * build_olmo() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -12448,7 +12456,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_openelm() { struct ggml_cgraph * build_openelm() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -12573,7 +12581,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_gptneox() { struct ggml_cgraph * build_gptneox() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -12715,7 +12723,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_arctic() { struct ggml_cgraph * build_arctic() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -12847,7 +12855,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_deepseek2() { struct ggml_cgraph * build_deepseek2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -13075,7 +13083,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_bitnet() { struct ggml_cgraph * build_bitnet() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@ -13215,7 +13223,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_t5() { struct ggml_cgraph * build_t5() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
// mutable variable, needed during the last layer of the computation to skip unused tokens // mutable variable, needed during the last layer of the computation to skip unused tokens
int32_t n_tokens = this->n_tokens; int32_t n_tokens = this->n_tokens;
@ -13532,7 +13540,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_jais() { struct ggml_cgraph * build_jais() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -13624,7 +13632,7 @@ struct llm_build_context {
} }
struct ggml_cgraph * build_chatglm() { struct ggml_cgraph * build_chatglm() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false); struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
@ -14951,9 +14959,9 @@ static void llama_kv_cache_defrag_internal(struct llama_context & lctx) {
// each move requires 6*n_layer tensors (see build_defrag) // each move requires 6*n_layer tensors (see build_defrag)
// - source view, destination view, copy operation // - source view, destination view, copy operation
// - x2 for keys and values // - x2 for keys and values
//const uint32_t max_moves = LLAMA_MAX_NODES/(6*n_layer); //const uint32_t max_moves = llama_model_max_nodes(model)/(6*n_layer);
// TODO: tmp fix https://github.com/ggerganov/llama.cpp/issues/6685#issuecomment-2057579516 // TODO: tmp fix https://github.com/ggerganov/llama.cpp/issues/6685#issuecomment-2057579516
const uint32_t max_moves = (LLAMA_MAX_NODES - 2*n_layer)/(6*n_layer); const uint32_t max_moves = (llama_model_max_nodes(lctx.model) - 2*n_layer)/(6*n_layer);
// determine which KV cells to move where // determine which KV cells to move where
// //
@ -19351,8 +19359,10 @@ struct llama_context * llama_new_context_with_model(
} }
} }
const size_t max_nodes = llama_model_max_nodes(*model);
// buffer used to store the computation graph and the tensor meta data // buffer used to store the computation graph and the tensor meta data
ctx->buf_compute_meta.resize(ggml_tensor_overhead()*LLAMA_MAX_NODES + ggml_graph_overhead_custom(LLAMA_MAX_NODES, false)); ctx->buf_compute_meta.resize(ggml_tensor_overhead()*max_nodes + ggml_graph_overhead_custom(max_nodes, false));
// enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary // enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary
bool pipeline_parallel = bool pipeline_parallel =
@ -19365,7 +19375,7 @@ struct llama_context * llama_new_context_with_model(
// currently this is only implemented in the CUDA backend // currently this is only implemented in the CUDA backend
pipeline_parallel = false; pipeline_parallel = false;
#endif #endif
ctx->sched = ggml_backend_sched_new(ctx->backends.data(), backend_buft.data(), ctx->backends.size(), LLAMA_MAX_NODES, pipeline_parallel); ctx->sched = ggml_backend_sched_new(ctx->backends.data(), backend_buft.data(), ctx->backends.size(), max_nodes, pipeline_parallel);
if (pipeline_parallel) { if (pipeline_parallel) {
LLAMA_LOG_INFO("%s: pipeline parallelism enabled (n_copies=%d)\n", __func__, ggml_backend_sched_get_n_copies(ctx->sched)); LLAMA_LOG_INFO("%s: pipeline parallelism enabled (n_copies=%d)\n", __func__, ggml_backend_sched_get_n_copies(ctx->sched));