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

cont [no ci]

Browse source
This commit is contained in:
Georgi Gerganov 2024-12-16 20:39:46 +02:00
parent 35259e5335
commit 2033fb7eef
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 6 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

13
src/llama.cpp
View file

@ -77,7 +77,6 @@
#endif #endif
// bump if necessary // bump if necessary
#define LLAMA_MAX_EMBD 8
#define LLAMA_MAX_LAYERS 512 #define LLAMA_MAX_LAYERS 512
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2 #define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
@ -3074,8 +3073,8 @@ struct llama_model {
struct ggml_tensor * cls_out = nullptr; struct ggml_tensor * cls_out = nullptr;
struct ggml_tensor * cls_out_b = nullptr; struct ggml_tensor * cls_out_b = nullptr;
struct ggml_tensor * conv_1d = nullptr; struct ggml_tensor * conv1d = nullptr;
struct ggml_tensor * conv_1d_b = nullptr; struct ggml_tensor * conv1d_b = nullptr;
std::vector<llama_layer> layers; std::vector<llama_layer> layers;
@ -9498,8 +9497,8 @@ static bool llm_load_tensors(
{ {
model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {hparams.n_embd_features, n_vocab}, 0); model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {hparams.n_embd_features, n_vocab}, 0);
model.conv_1d = create_tensor(tn(LLM_TENSOR_CONV1D, "weight"), {7, hparams.n_embd_features, hparams.posnet.n_embd}, 0); model.conv1d = create_tensor(tn(LLM_TENSOR_CONV1D, "weight"), {7, hparams.n_embd_features, hparams.posnet.n_embd}, 0);
model.conv_1d_b = create_tensor(tn(LLM_TENSOR_CONV1D, "bias"), {1, hparams.posnet.n_embd}, 0); model.conv1d_b = create_tensor(tn(LLM_TENSOR_CONV1D, "bias"), {1, hparams.posnet.n_embd}, 0);
// posnet // posnet
{ {
@ -17183,8 +17182,8 @@ struct llm_build_context {
cur = ggml_cont(ctx0, ggml_transpose(ctx0, inpL)); cur = ggml_cont(ctx0, ggml_transpose(ctx0, inpL));
cur = ggml_conv_1d_ph(ctx0, model.conv_1d, cur, 1, 1); cur = ggml_conv_1d_ph(ctx0, model.conv1d, cur, 1, 1);
cur = ggml_add(ctx0, cur, model.conv_1d_b); cur = ggml_add(ctx0, cur, model.conv1d_b);
// posnet // posnet
for (uint32_t il = 0; il < hparams.posnet.n_layer; ++il) { for (uint32_t il = 0; il < hparams.posnet.n_layer; ++il) {