vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 17:44:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

don't build rope factors from https://github.com/ggerganov/llama.cpp/pull/8676 for CLBlast as it segfaults

Browse source
This commit is contained in:
Concedo 2024-07-28 17:27:09 +08:00
parent edbdfbced2
commit e47477fd4d
1 changed files with 5 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

6
src/llama.cpp
View file

@ -5921,7 +5921,7 @@ static bool llm_load_tensors(
#if defined(GGML_USE_CLBLAST) #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...\n");
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);
} }
@ -8784,7 +8784,11 @@ struct llm_build_context {
// self-attention // self-attention
{ {
// rope freq factors for llama3; may return nullptr for llama2 and other models // rope freq factors for llama3; may return nullptr for llama2 and other models
#if defined(GGML_USE_CLBLAST)
struct ggml_tensor * rope_factors = nullptr; //clblast does not work with rope_factors
#else
struct ggml_tensor * rope_factors = build_rope_factors(il); struct ggml_tensor * rope_factors = build_rope_factors(il);
#endif
// compute Q and K and RoPE them // compute Q and K and RoPE them
struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur); struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);