upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-29 09:27:33 -05:00
Code Issues Packages Projects Releases Wiki Activity

hrm : fix layer placement of z_l_init weight (#29512)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-09-27 10:16:23 +03:00
committed by GitHub
parent 7ac59a6e3a
commit 85ca3b52c3
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
+1 -1
View File
@@ -722,7 +722,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_TOKEN_EMBD, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
{LLM_TENSOR_POS_EMBD, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
{LLM_TENSOR_TOKEN_TYPES, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
{LLM_TENSOR_HRM_Z_L_INIT, {LLM_TENSOR_LAYER_INPUT, GGML_OP_ADD}},
{LLM_TENSOR_HRM_Z_L_INIT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
{LLM_TENSOR_TOKEN_EMBD_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, // do the norms on the first layer (not the input layer)
{LLM_TENSOR_OUTPUT, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
{LLM_TENSOR_CLS, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
src/models/hrm-text.cpp
+1 -1
View File
@@ -43,7 +43,7 @@ void llama_model_hrm_text::load_arch_tensors(llama_model_loader &) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, TENSOR_DUPLICATED);
}
hrm_z_l_init = create_tensor(tn(LLM_TENSOR_HRM_Z_L_INIT), { n_embd }, 0);
hrm_z_l_init = create_tensor(tn(LLM_TENSOR_HRM_Z_L_INIT, 0), { n_embd }, 0);
const int lps = hparams.n_hrm_layers_per_stack;