vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

CUDA full GPU acceleration, KV cache in VRAM (#1827)

Browse source 
* Fixed CUDA RoPE

* ggml_cuda_mul_mat_vec_p021

* ggml_cuda_scale

* ggml_cuda_diag_mask_inf

* ggml_is_permuted

* ggml_cuda_cpy

* flatten rows for ggml_cuda_op

* Added a --low-vram option

* Fixed Windows performance

* Fixed LLAMA_CUDA_DMMV_Y > 1 for WizardLM
This commit is contained in:
Johannes Gäßler 2023-06-14 19:47:19 +02:00 • committed by GitHub
parent 9254920265
commit 254a7a7a5f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
11 changed files with 853 additions and 149 deletions
Download patch file Download diff file Expand all files Collapse all files

17
examples/common.h
View file

@ -21,15 +21,16 @@
int32_t get_num_physical_cores();
struct gpt_params {
int32_t seed = -1; // RNG seed
int32_t n_threads = get_num_physical_cores();
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 512; // context size
int32_t n_batch = 512; // batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_keep = 0; // number of tokens to keep from initial prompt
int32_t n_gpu_layers = 0; // number of layers to store in VRAM
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
int32_t seed = -1; // RNG seed
int32_t n_threads = get_num_physical_cores();
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 512; // context size
int32_t n_batch = 512; // batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_keep = 0; // number of tokens to keep from initial prompt
int32_t n_gpu_layers = 0; // number of layers to store in VRAM
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[LLAMA_MAX_DEVICES] = {0}; // how split tensors should be distributed across GPUs
bool low_vram = 0; // if true, reduce VRAM usage at the cost of performance
// sampling parameters
std::unordered_map<llama_token, float> logit_bias; // logit bias for specific tokens