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

sampling : refactor init to use llama_sampling_params

Browse source
This commit is contained in:
Georgi Gerganov 2023-10-20 14:58:20 +03:00
parent 8cf19d60dc
commit cd1e937821
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
12 changed files with 110 additions and 142 deletions
Download patch file Download diff file Expand all files Collapse all files

5
common/sampling.h
View file

@ -22,6 +22,7 @@ typedef struct llama_sampling_params {
int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
float mirostat_tau = 5.00f; // target entropy
float mirostat_eta = 0.10f; // learning rate
int32_t n_prev = 256; // number of previous tokens to remember
bool penalize_nl = true; // consider newlines as a repeatable token
@ -34,6 +35,8 @@ typedef struct llama_sampling_params {
std::unordered_map<llama_token, float> logit_bias; // logit bias for specific tokens
std::string grammar = ""; // optional BNF-like grammar to constrain sampling
} llama_sampling_params;
// general sampler context
@ -58,7 +61,7 @@ struct llama_sampling_context {
#include "common.h"
// Create a new sampling context instance.
struct llama_sampling_context * llama_sampling_init(const struct gpt_params & params);
struct llama_sampling_context * llama_sampling_init(const struct llama_sampling_params & params);
void llama_sampling_free(struct llama_sampling_context * ctx);