ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

speculative : update default params (#11954)

Browse source 
* speculative : update default params

* speculative : do not discard the last drafted token
This commit is contained in:
Georgi Gerganov 2025-02-19 13:29:42 +02:00 • committed by GitHub
parent 9626d9351a
commit abd4d0bc4f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 9 additions and 9 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_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)
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_batch;