kanshan/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-06-26 06:10:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

cli : fix not copying preserved tokens (#24258)

Browse Source
This commit is contained in:
Michael Wand
2026-06-14 02:52:15 -07:00
committed by GitHub
parent 8ed274ef46
commit 6e14286eda
1 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/cli/cli.cpp
+10 -3
View File
@@ -97,11 +97,18 @@ struct cli_context {
task.params.chat_parser_params.parser.load(chat_params.parser);
}
// Copy the preserved tokens into the sampling params
const llama_vocab * vocab = llama_model_get_vocab(
llama_get_model(ctx_server.get_llama_context()));
for (const auto & token : chat_params.preserved_tokens) {
auto ids = common_tokenize(vocab, token, false, true);
if (ids.size() == 1) {
task.params.sampling.preserved_tokens.insert(ids[0]);
}
}
// reasoning budget sampler
if (!chat_params.thinking_end_tag.empty()) {
const llama_vocab * vocab = llama_model_get_vocab(
llama_get_model(ctx_server.get_llama_context()));
task.params.sampling.reasoning_budget_tokens = defaults.sampling.reasoning_budget_tokens;
task.params.sampling.generation_prompt = chat_params.generation_prompt;