vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'upstream' into concedo_experimental

Browse source 
# Conflicts:
#	CMakeLists.txt
This commit is contained in:
Concedo 2024-06-03 18:33:23 +08:00
commit 94753ad103
6 changed files with 73395 additions and 13846 deletions
Download patch file Download diff file Expand all files Collapse all files

14
llama.cpp
View file

@ -5194,12 +5194,10 @@ static bool llm_load_tensors(
// output
{
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
if (model.arch != LLM_ARCH_MINICPM){
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
}
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
}
}
@ -10282,7 +10280,7 @@ struct llm_build_context {
cb(cur, "lmhead_scaling", -1);
// lm_head
cur = ggml_mul_mat(ctx0, model.tok_embd, cur);
cur = ggml_mul_mat(ctx0, model.output, cur);
cb(cur, "result_output", -1);
ggml_build_forward_expand(gf, cur);
@ -16677,7 +16675,7 @@ struct llama_context * llama_new_context_with_model(
return nullptr;
}
if (model->split_mode == LLAMA_SPLIT_MODE_NONE) {
ggml_backend_t backend = ggml_backend_vk_init(0);
ggml_backend_t backend = ggml_backend_vk_init(model->main_gpu);
if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize Vulkan backend\n", __func__);
llama_free(ctx);