vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-12 18:09:42 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge commit 'f7625019c5' into concedo_experimental

Browse source 
# Conflicts:
#	.github/ISSUE_TEMPLATE/bug.md
#	.github/workflows/build.yml
#	CMakeLists.txt
#	Makefile
#	README.md
#	tests/test-backend-ops.cpp
#	tests/test-opt.cpp
#	tests/test-quantize-fns.cpp
This commit is contained in:
Concedo 2024-02-26 10:44:23 +08:00
commit 3ccaf8e09a
37 changed files with 3514 additions and 698 deletions
Download patch file Download diff file Expand all files Collapse all files

4
common/common.h
View file

@ -55,7 +55,7 @@ struct gpt_params {
float p_split = 0.1f; // speculative decoding split probability
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t n_gpu_layers_draft = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
llama_split_mode split_mode = LLAMA_SPLIT_LAYER; // how to split the model across GPUs
llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
int32_t n_beams = 0; // if non-zero then use beam search of given width.
@ -69,7 +69,7 @@ struct gpt_params {
float yarn_beta_fast = 32.0f; // YaRN low correction dim
float yarn_beta_slow = 1.0f; // YaRN high correction dim
int32_t yarn_orig_ctx = 0; // YaRN original context length
int32_t rope_scaling_type = LLAMA_ROPE_SCALING_UNSPECIFIED;
int32_t rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED;
ggml_numa_strategy numa = GGML_NUMA_STRATEGY_DISABLED;
// sampling parameters