vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 09:34:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

workaround for deepseek not working

Browse source
This commit is contained in:
Concedo 2024-07-05 22:39:53 +08:00
parent 6b0756506b
commit 388a2aff00
3 changed files with 17 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-opencl.cpp
View file

@ -1812,6 +1812,10 @@ static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
cl_kernel* to_fp32_cl = ggml_get_to_fp32_cl(type); cl_kernel* to_fp32_cl = ggml_get_to_fp32_cl(type);
cl_kernel* dmmv = ggml_get_dequantize_mul_mat_vec_cl(type); cl_kernel* dmmv = ggml_get_dequantize_mul_mat_vec_cl(type);
if(to_fp32_cl==nullptr)
{
printf("\nOpenCL: Unsupported Tensor Type Detected: %d\n",type);
}
GGML_ASSERT(to_fp32_cl != nullptr); GGML_ASSERT(to_fp32_cl != nullptr);
const size_t global_denom = ggml_cl_global_denom(type); const size_t global_denom = ggml_cl_global_denom(type);

2
src/llama.cpp
View file

@ -5607,12 +5607,14 @@ static bool llm_load_tensors(
int64_t i_gpu_start = std::max((int64_t) hparams.n_layer - n_gpu_layers, (int64_t) 0); int64_t i_gpu_start = std::max((int64_t) hparams.n_layer - n_gpu_layers, (int64_t) 0);
bool use_mmap_buffer = true; bool use_mmap_buffer = true;
#if defined(GGML_USE_CLBLAST)
if(clblast_offload_fallback_mode) if(clblast_offload_fallback_mode)
{ {
printf("\nOpenCL GPU Offload Fallback..."); printf("\nOpenCL GPU Offload Fallback...");
clblast_offload_fallback_layers = n_gpu_layers; clblast_offload_fallback_layers = n_gpu_layers;
i_gpu_start = std::max((int64_t) hparams.n_layer, (int64_t) 0); i_gpu_start = std::max((int64_t) hparams.n_layer, (int64_t) 0);
} }
#endif
// there is very little benefit to offloading the input layer, so always keep it on the CPU // there is very little benefit to offloading the input layer, so always keep it on the CPU
model.buft_input = llama_default_buffer_type_cpu(true); model.buft_input = llama_default_buffer_type_cpu(true);

12
src/unicode.cpp
View file

@ -189,9 +189,19 @@ static std::unordered_map<std::string, uint8_t> unicode_utf8_to_byte_map() {
return map; return map;
} }
static bool unicode_wstring_from_utf8_failed_once = false;
static inline std::wstring unicode_wstring_from_utf8(const std::string & s) { static inline std::wstring unicode_wstring_from_utf8(const std::string & s) {
std::wstring_convert<std::codecvt_utf8<wchar_t>> conv; std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
return conv.from_bytes(s); try {
return conv.from_bytes(s);
} catch(const std::exception & e) {
if(!unicode_wstring_from_utf8_failed_once)
{
unicode_wstring_from_utf8_failed_once = true;
printf("\nunicode_wstring_from_utf8 failed: %s\n", e.what());
}
return L"";
}
} }
static std::vector<std::string> unicode_byte_encoding_process(const std::vector<std::string> & bpe_words) { static std::vector<std::string> unicode_byte_encoding_process(const std::vector<std::string> & bpe_words) {