matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 16:57:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: remove model loading for grammar (#10096)

Browse Source
This commit is contained in:
Parth Sareen
2025-04-24 11:51:19 -07:00
committed by GitHub
parent 40b10eee6d
commit a53d744b01
13 changed files with 521 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
llama/llama.cpp/src/llama-sampling.cpp vendored
View File

@@ -1465,7 +1465,7 @@ static void llama_sampler_grammar_reset(struct llama_sampler * smpl) {
trigger_patterns_c.push_back(trigger_pattern.pattern.c_str());
}
auto * grammar_new = llama_grammar_init_impl(ctx->grammar->vocab, ctx->grammar_str.c_str(), ctx->grammar_root.c_str(),
auto * grammar_new = llama_grammar_init_impl(ctx->grammar->vocab, nullptr, ctx->grammar_str.c_str(), ctx->grammar_root.c_str(),
ctx->grammar->lazy, trigger_patterns_c.data(), trigger_patterns_c.size(),
ctx->grammar->trigger_tokens.data(), ctx->grammar->trigger_tokens.size());
@@ -1547,7 +1547,7 @@ static struct llama_sampler * llama_sampler_init_grammar_impl(
/* .vocab = */ vocab,
/* .grammar_str = */ grammar_str,
/* .grammar_root = */ grammar_root,
/* .grammar = */ llama_grammar_init_impl(vocab, grammar_str, grammar_root, lazy, trigger_patterns, num_trigger_patterns, trigger_tokens, num_trigger_tokens),
/* .grammar = */ llama_grammar_init_impl(vocab, nullptr, grammar_str, grammar_root, lazy, trigger_patterns, num_trigger_patterns, trigger_tokens, num_trigger_tokens),
};
if (!ctx->grammar) {
delete ctx;