root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 12:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

hann window

Browse Source
This commit is contained in:
Georgi Gerganov 2024-12-11 10:52:07 +02:00
parent 0150e413b2
commit dcf2230afb
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
3 changed files with 11 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
examples/tts/convert_pt_to_hf.py
View File

@ -101,6 +101,9 @@ def flatten_state_dict(state_dict, parent_key='', sep='.'):
if new_key.endswith("gamma"): if new_key.endswith("gamma"):
new_key = new_key.replace("gamma", "gamma.weight") new_key = new_key.replace("gamma", "gamma.weight")
if new_key == "head.istft.window":
new_key = "head.istft.window.weight"
size_mb = value.element_size() * value.nelement() / (1024 * 1024) size_mb = value.element_size() * value.nelement() / (1024 * 1024)
print(f"{size_mb:8.2f} MB - {new_key}: {value.shape}") print(f"{size_mb:8.2f} MB - {new_key}: {value.shape}")

1
gguf-py/gguf/tensor_mapping.py
View File

@ -28,7 +28,6 @@ class TensorNameMap:
"transformer.token_embeddings", # openelm "transformer.token_embeddings", # openelm
"shared", # t5 "shared", # t5
"rwkv.embeddings", # rwkv "rwkv.embeddings", # rwkv
"feature_extractor.encodec.quantizer.vq.layers.0._codebook.embed" # outetts
), ),
# Token type embeddings # Token type embeddings

12
src/llama.cpp
View File

@ -3070,6 +3070,8 @@ struct llama_model {
struct ggml_tensor * conv_1d = nullptr; struct ggml_tensor * conv_1d = nullptr;
struct ggml_tensor * conv_1d_b = nullptr; struct ggml_tensor * conv_1d_b = nullptr;
struct ggml_tensor * hann_window = nullptr;
// resnet 0 // resnet 0
struct ggml_tensor * posnet_0_norm1 = nullptr; struct ggml_tensor * posnet_0_norm1 = nullptr;
struct ggml_tensor * posnet_0_norm1_b = nullptr; struct ggml_tensor * posnet_0_norm1_b = nullptr;
@ -5121,8 +5123,7 @@ struct llama_model_loader {
void done_getting_tensors() const { void done_getting_tensors() const {
if (n_created != n_tensors) { if (n_created != n_tensors) {
// TODO: TEMPORARY DISABLED [OUTETTS] throw std::runtime_error(format("%s: wrong number of tensors; expected %d, got %d", __func__, n_tensors, n_created));
//throw std::runtime_error(format("%s: wrong number of tensors; expected %d, got %d", __func__, n_tensors, n_created));
} }
} }
@ -7461,6 +7462,7 @@ static const std::map<llm_tensor, llm_tensor_info> llm_tensor_info_mapping = {
{LLM_TENSOR_CONV_NEXT_PW1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_CONV_NEXT_PW1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_CONV_NEXT_PW2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_CONV_NEXT_PW2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_CONV_NEXT_GAMMA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, {LLM_TENSOR_CONV_NEXT_GAMMA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_HANN_WINDOW, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
}; };
// checks if the weight tensor can be used with the specified buffer type and device // checks if the weight tensor can be used with the specified buffer type and device
@ -9634,8 +9636,10 @@ static bool llm_load_tensors(
model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {768}, 0); model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {768}, 0);
model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {768}, 0); model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {768}, 0);
model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {768, 1282}, llama_model_loader::TENSOR_NOT_REQUIRED); model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {768, 1282}, 0);
model.output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {1282}, llama_model_loader::TENSOR_NOT_REQUIRED); model.output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {1282}, 0);
model.hann_window = create_tensor(tn(LLM_TENSOR_HANN_WINDOW, "weight"), {1280}, 0);
} break; } break;
default: default:
throw std::runtime_error("unknown architecture"); throw std::runtime_error("unknown architecture");