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

grammars: nits (revert const grammar sig, fix comment)

Browse source
This commit is contained in:
ochafik 2024-04-28 15:34:56 +01:00
parent d3425f5cf1
commit c70037f2b3
2 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama.cpp
View file

@ -13545,7 +13545,7 @@ void llama_sample_repetition_penalties(
} }
} }
void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * candidates, struct llama_grammar * grammar) { void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * candidates, const struct llama_grammar * grammar) {
GGML_ASSERT(ctx); GGML_ASSERT(ctx);
const int64_t t_start_sample_us = ggml_time_us(); const int64_t t_start_sample_us = ggml_time_us();
@ -13557,7 +13557,7 @@ void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * c
} }
} }
// Store decoded codepoints when they are not cached. // Store decoded codepoints when they are not cached (happens when there's a partial utf8 string prefix).
std::vector<std::pair<std::vector<uint32_t>, llama_partial_utf8>> candidates_decoded; std::vector<std::pair<std::vector<uint32_t>, llama_partial_utf8>> candidates_decoded;
if (grammar->partial_utf8.n_remain > 0) { if (grammar->partial_utf8.n_remain > 0) {
candidates_decoded.reserve(candidates->size); candidates_decoded.reserve(candidates->size);

2
llama.h
View file

@ -961,7 +961,7 @@ extern "C" {
LLAMA_API void llama_sample_grammar( LLAMA_API void llama_sample_grammar(
struct llama_context * ctx, struct llama_context * ctx,
llama_token_data_array * candidates, llama_token_data_array * candidates,
struct llama_grammar * grammar); const struct llama_grammar * grammar);
/// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words. /// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
/// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text. /// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.