mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-27 08:27:30 -05:00
Oliver Simons
a6aa6f5450
* Resolve -1 to 1024 instead of ctx-len for samplers Because of backend-sampling we initialize samplers before the complete llama_context is there. Therefore, we cannot infer the resolved context length yet at the time we construct the samplers. * Shared default of 64 for history-based samplers, remove context_size
42 lines
824 B
C++
42 lines
824 B
C++
#pragma once
|
|
|
|
#include "llama.h"
|
|
|
|
#include <vector>
|
|
|
|
struct llama_vocab;
|
|
struct llama_grammar;
|
|
|
|
// sampler chain
|
|
|
|
struct llama_sampler_chain {
|
|
llama_sampler_chain_params params;
|
|
|
|
// has .backend_init() been called?
|
|
bool is_init = false;
|
|
|
|
struct info {
|
|
bool is_backend;
|
|
|
|
llama_sampler * ptr;
|
|
};
|
|
|
|
std::vector<info> samplers;
|
|
|
|
// pre-allocated buffer for llama_sampler_sample to avoid repeated allocations
|
|
std::vector<llama_token_data> cur;
|
|
|
|
// timing
|
|
|
|
mutable int64_t t_sample_us;
|
|
|
|
mutable int32_t n_sample;
|
|
};
|
|
|
|
struct llama_sampler * llama_sampler_init_dry_testing(
|
|
float dry_multiplier,
|
|
float dry_base,
|
|
int32_t dry_allowed_length,
|
|
int32_t dry_penalty_last_n,
|
|
const std::vector<std::vector<llama_token>> & seq_breakers);
|