1#pragma once
2
3#include "llama.h"
4
5#include "common.h"
6
7#include <string>
8#include <vector>
9
10// common_sampler extends llama_sampler with additional functionality:
11//
12// - grammar support
13// - custom sampler logic based on the parameters
14// - history of the last accepted tokens
15// - performance metrics
16//
17// This goal is to have a common implementation of the sampling logic shared across the examples.
18// For example, depending on the temperature, the sampling chain can be very simple (greedy) or more
19// complex (top-k, top-p, etc).
20//
21// Another example is related to the grammar. In general, the grammar constraints applied on the full
22// vocabulary can be very taxing. To improve performance, the grammar can be applied only to the sampled
23// token in order to verify if it fits the grammar. And only if the token doesn't fit the grammar, the
24// grammar constraints are applied to the full vocabulary and the token is resampled.
25//
26// The common_sampler also maintains a container with the last accepted tokens. In the future, this can
27// be moved into the core llama library.
28//
29// For convenience, the common_sampler also maintains a container with the current candidate tokens.
30// This can be used to access the probabilities of the rest of the non-sampled tokens.
31//
32// TODO: measure grammar performance
33//
34
35struct common_sampler;
36
37// llama_sampler API overloads
38
39// note: can mutate params in some cases
40struct common_sampler * common_sampler_init(const struct llama_model * model, struct common_params_sampling & params);
41
42void common_sampler_free(struct common_sampler * gsmpl);
43
44// if accept_grammar is true, the token is accepted both by the sampling chain and the grammar
45void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, bool accept_grammar);
46void common_sampler_reset (struct common_sampler * gsmpl);
47struct common_sampler * common_sampler_clone (struct common_sampler * gsmpl);
48
49// arguments can be nullptr to skip printing
50void common_perf_print(const struct llama_context * ctx, const struct common_sampler * gsmpl);
51
52// get the underlying llama_sampler_chain
53struct llama_sampler * common_sampler_get(const struct common_sampler * gsmpl);
54
55// extended sampling implementation:
56//
57// - set logits
58// - apply the configured sampler chain
59// - check if the token fits the grammar (if any)
60// - if not: resample by first applying the grammar constraints and then sampling again (slower path)
61//
62// if grammar_first is true, the grammar is applied before the samplers (slower)
63// useful in cases where all the resulting candidates (not just the sampled one) must fit the grammar
64//
65llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_context * ctx, int idx, bool grammar_first = false);
66
67// generalized version of common_sampler_sample
68//
69// will cross-reference the sampled tokens with a batch of draft tokens and accept those that match
70// if the sampler disagrees at some point, we stop and return the accepted tokens up to now
71//
72// common_sampler_sample_n(gsmpl, ctx, { idx }, {});
73//
74// is equivalent to
75//
76// common_sampler_sample(gsmpl, ctx, idx);
77// common_sampler_accept(gsmpl, token, true);
78//
79// requires: idxs.size() == draft.size() + 1
80//
81// returns at least 1 token, up to idxs.size()
82//
83std::vector<llama_token> common_sampler_sample_and_accept_n(struct common_sampler * gsmpl, struct llama_context * ctx, const std::vector<int> & idxs, const llama_tokens & draft, bool grammar_first = false);
84
85// assume idxs == [ 0, 1, 2, ..., draft.size() ]
86std::vector<llama_token> common_sampler_sample_and_accept_n(struct common_sampler * gsmpl, struct llama_context * ctx, const llama_tokens & draft, bool grammar_first = false);
87
88uint32_t common_sampler_get_seed(const struct common_sampler * gsmpl);
89
90// helpers
91
92// access the internal list of current candidate tokens
93// if do_sort == true, the candidates are guaranteed to be sorted afterwards (in descending order of probability)
94// the .sorted flag of the result indicates whether the returned candidates are sorted
95llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl, bool do_sort);
96
97// get the last accepted token
98llama_token common_sampler_last(const struct common_sampler * gsmpl);
99
100// print the sampler chain into a string
101std::string common_sampler_print(const struct common_sampler * gsmpl);
102
103// get a string representation of the last accepted tokens
104std::string common_sampler_prev_str(common_sampler * gsmpl, llama_context * ctx, int n);
105
106char common_sampler_type_to_chr(enum common_sampler_type cnstr);
107std::string common_sampler_type_to_str(enum common_sampler_type cnstr);
108
109std::vector<enum common_sampler_type> common_sampler_types_from_names(const std::vector<std::string> & names, bool allow_alt_names);
110std::vector<enum common_sampler_type> common_sampler_types_from_chars(const std::string & chars);
111
112llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab,
113 const char * grammar_kind, const char * grammar_data);
114
115struct common_sampler_deleter {
116 void operator()(common_sampler * s) { common_sampler_free(s); }
117};
118
119typedef std::unique_ptr<common_sampler, common_sampler_deleter> common_sampler_ptr;