happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

model : fix wavtokenizer embedding notions (#19479)

This commit is contained in:
Georgi Gerganov 2026-02-11 07:52:20 +02:00 committed by GitHub
parent 89181c0b6d
commit 6d95707827
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/llama-hparams.h
View File

@ -42,7 +42,6 @@ struct llama_hparams {
uint32_t n_ctx_train; // context size the model was trained on uint32_t n_ctx_train; // context size the model was trained on
uint32_t n_embd; uint32_t n_embd;
uint32_t n_embd_features = 0;
uint32_t n_layer; uint32_t n_layer;
int32_t n_layer_kv_from_start = -1; // if non-negative, the first n_layer_kv_from_start layers have KV cache int32_t n_layer_kv_from_start = -1; // if non-negative, the first n_layer_kv_from_start layers have KV cache
uint32_t n_rot; uint32_t n_rot;

11
src/llama-model.cpp
View File

@ -523,7 +523,8 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_EXPERT_GROUP_USED_COUNT, hparams.n_group_used, false); ml.get_key(LLM_KV_EXPERT_GROUP_USED_COUNT, hparams.n_group_used, false);
if (arch == LLM_ARCH_WAVTOKENIZER_DEC) { if (arch == LLM_ARCH_WAVTOKENIZER_DEC) {
ml.get_key(LLM_KV_FEATURES_LENGTH, hparams.n_embd_features); ml.get_key(LLM_KV_FEATURES_LENGTH, hparams.n_embd);
ml.get_key(LLM_KV_EMBEDDING_LENGTH, hparams.n_embd_out_impl);
ml.get_key(LLM_KV_POSNET_EMBEDDING_LENGTH, hparams.posnet.n_embd); ml.get_key(LLM_KV_POSNET_EMBEDDING_LENGTH, hparams.posnet.n_embd);
ml.get_key(LLM_KV_POSNET_BLOCK_COUNT, hparams.posnet.n_layer); ml.get_key(LLM_KV_POSNET_BLOCK_COUNT, hparams.posnet.n_layer);
@ -6046,9 +6047,9 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
} break; } break;
case LLM_ARCH_WAVTOKENIZER_DEC: case LLM_ARCH_WAVTOKENIZER_DEC:
{ {
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {hparams.n_embd_features, n_vocab}, 0); tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {hparams.n_embd, n_vocab}, 0);
conv1d = create_tensor(tn(LLM_TENSOR_CONV1D, "weight"), {7, hparams.n_embd_features, hparams.posnet.n_embd}, 0); conv1d = create_tensor(tn(LLM_TENSOR_CONV1D, "weight"), {7, hparams.n_embd, hparams.posnet.n_embd}, 0);
conv1d_b = create_tensor(tn(LLM_TENSOR_CONV1D, "bias"), {1, hparams.posnet.n_embd}, 0); conv1d_b = create_tensor(tn(LLM_TENSOR_CONV1D, "bias"), {1, hparams.posnet.n_embd}, 0);
// posnet // posnet
@ -6144,8 +6145,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0);
} }
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {hparams.convnext.n_embd, n_embd}, 0); output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {hparams.convnext.n_embd, hparams.n_embd_out()}, 0);
output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {n_embd}, 0); output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {hparams.n_embd_out()}, 0);
} break; } break;
case LLM_ARCH_BAILINGMOE: case LLM_ARCH_BAILINGMOE:
{ {

2
tools/tts/tts.cpp
View File

@ -1036,7 +1036,7 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
#if 1 #if 1
// spectral operations // spectral operations
const int n_embd = llama_model_n_embd(model_cts); const int n_embd = llama_model_n_embd_out(model_cts);
const float * embd = llama_get_embeddings(ctx_cts); const float * embd = llama_get_embeddings(ctx_cts);
auto audio = embd_to_audio(embd, n_codes, n_embd, params.cpuparams.n_threads); auto audio = embd_to_audio(embd, n_codes, n_embd, params.cpuparams.n_threads);