tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

speculative : update default params

Browse Source
This commit is contained in:
Georgi Gerganov
2025-02-19 08:20:10 +02:00
parent 9626d9351a
commit 965ad1c08a
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
common/common.h
View File

@ -178,10 +178,10 @@ struct common_params_speculative {
int32_t n_ctx = 0; // draft context size int32_t n_ctx = 0; // draft context size
int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding
int32_t n_min = 5; // minimum number of draft tokens to use for speculative decoding int32_t n_min = 0; // minimum number of draft tokens to use for speculative decoding
int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default) int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
float p_split = 0.1f; // speculative decoding split probability float p_split = 0.1f; // speculative decoding split probability
float p_min = 0.9f; // minimum speculative decoding probability (greedy) float p_min = 0.75f; // minimum speculative decoding probability (greedy)
struct cpu_params cpuparams; struct cpu_params cpuparams;
struct cpu_params cpuparams_batch; struct cpu_params cpuparams_batch;

2
common/speculative.h
View File

@ -9,7 +9,7 @@ struct common_speculative_params {
int n_draft = 16; // max drafted tokens int n_draft = 16; // max drafted tokens
int n_reuse = 256; int n_reuse = 256;
float p_min = 0.9f; // min probability required to accept a token in the draft float p_min = 0.75f; // min probability required to accept a token in the draft
}; };
struct common_speculative * common_speculative_init(struct llama_context * ctx_dft); struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);