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

sampling : support for llguidance grammars (#10224)

Browse source 
* initial porting of previous LLG patch

* update for new APIs

* build: integrate llguidance as an external project

* use '%llguidance' as marker to enable llg lark syntax

* add some docs

* clarify docs

* code style fixes

* remove llguidance.h from .gitignore

* fix tests when llg is enabled

* pass vocab not model to llama_sampler_init_llg()

* copy test-grammar-integration.cpp to test-llguidance.cpp

* clang fmt

* fix ref-count bug

* build and run test

* gbnf -> lark syntax

* conditionally include llguidance test based on LLAMA_LLGUIDANCE flag

* rename llguidance test file to test-grammar-llguidance.cpp

* add gh action for llg test

* align tests with LLG grammar syntax and JSON Schema spec

* llama_tokenizer() in fact requires valid utf8

* update llg

* format file

* add $LLGUIDANCE_LOG_LEVEL support

* fix whitespace

* fix warning

* include <cmath> for INFINITY

* add final newline

* fail llama_sampler_init_llg() at runtime

* Link gbnf_to_lark.py script; fix links; refer to llg docs for lexemes

* simplify #includes

* improve doc string for LLAMA_LLGUIDANCE

* typo in merge

* bump llguidance to 0.6.12
This commit is contained in:
Michał Moskal 2025-02-01 23:55:32 -08:00 • committed by GitHub
parent 0cec062a63
commit ff227703d6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 1555 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

22
common/sampling.cpp
View file

@ -156,13 +156,25 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
for (const auto & str : params.grammar_trigger_words) {
trigger_words.push_back(str.word.c_str());
}
struct llama_sampler * grmr;
if (params.grammar.compare(0, 11, "%llguidance") == 0) {
#ifdef LLAMA_USE_LLGUIDANCE
grmr = llama_sampler_init_llg(vocab, "lark", params.grammar.c_str());
#else
GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
#endif // LLAMA_USE_LLGUIDANCE
} else {
grmr = params.grammar_lazy
? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
trigger_words.data(), trigger_words.size(),
params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
: llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
}
auto * result = new common_sampler {
/* .params = */ params,
/* .grmr = */ params.grammar_lazy
? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
trigger_words.data(), trigger_words.size(),
params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
: llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root"),
/* .grmr = */ grmr,
/* .chain = */ llama_sampler_chain_init(lparams),
/* .prev = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
/* .cur = */ {},