mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-11-01 09:01:57 +00:00
speculative : add infill mode
ggml-ci
This commit is contained in:
@@ -10,6 +10,8 @@ struct common_speculative_params {
|
||||
int n_reuse = 256;
|
||||
|
||||
float p_min = 0.9f; // min probabiliy required to accept a token in the draft
|
||||
|
||||
bool infill = false; // use infill sampling (useful for FIM)
|
||||
};
|
||||
|
||||
struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);
|
||||
|
||||
Reference in New Issue
Block a user