|
|
@ -93,6 +93,7 @@ struct LLamaPrivate {
|
|
|
|
llama_context *ctx = nullptr;
|
|
|
|
llama_context *ctx = nullptr;
|
|
|
|
llama_context_params params;
|
|
|
|
llama_context_params params;
|
|
|
|
int64_t n_threads = 0;
|
|
|
|
int64_t n_threads = 0;
|
|
|
|
|
|
|
|
std::vector<LLModel::Token> end_tokens;
|
|
|
|
};
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
LLamaModel::LLamaModel()
|
|
|
|
LLamaModel::LLamaModel()
|
|
|
@ -176,6 +177,8 @@ bool LLamaModel::loadModel(const std::string &modelPath)
|
|
|
|
return false;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
d_ptr->end_tokens = {llama_token_eos(d_ptr->ctx)};
|
|
|
|
|
|
|
|
|
|
|
|
#ifdef GGML_USE_KOMPUTE
|
|
|
|
#ifdef GGML_USE_KOMPUTE
|
|
|
|
if (ggml_vk_has_device()) {
|
|
|
|
if (ggml_vk_has_device()) {
|
|
|
|
std::cerr << "llama.cpp: using Vulkan on " << ggml_vk_current_device().name << std::endl;
|
|
|
|
std::cerr << "llama.cpp: using Vulkan on " << ggml_vk_current_device().name << std::endl;
|
|
|
@ -259,8 +262,7 @@ int32_t LLamaModel::contextLength() const
|
|
|
|
|
|
|
|
|
|
|
|
const std::vector<LLModel::Token> &LLamaModel::endTokens() const
|
|
|
|
const std::vector<LLModel::Token> &LLamaModel::endTokens() const
|
|
|
|
{
|
|
|
|
{
|
|
|
|
static const std::vector<LLModel::Token> fres = {llama_token_eos(d_ptr->ctx)};
|
|
|
|
return d_ptr->end_tokens;
|
|
|
|
return fres;
|
|
|
|
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#if defined(GGML_USE_KOMPUTE)
|
|
|
|
#if defined(GGML_USE_KOMPUTE)
|
|
|
|