mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 33 Packages Projects Releases Wiki Activity

common/parser: add proper reasoning tag prefill reading (#20424)

Browse Source 
* Implement proper prefill extraction

* Refactor cli parameters, update docs, move reasoning budget sampler part to common/reasoning-budget.cpp

* Update tools/server/server-task.cpp

* refactor: move grammars to variant, remove grammar_external, handle exception internally

* Make code less C++y

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Piotr Wilkin (ilintar)
2026-03-19 16:58:21 +01:00
committed by GitHub
parent c1258830b2
commit 5e54d51b19
33 changed files with 651 additions and 454 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test-chat-auto-parser.cpp
+3 -3
View File
@@ -1292,11 +1292,11 @@ static void test_nemotron_reasoning_detection(testing & t) {
// Check reasoning markers
t.assert_equal("reasoning_start should be '<think>'", "<think>", analysis.reasoning.start);
t.assert_equal("reasoning_end should be '</think>\\n'", "</think>\n", analysis.reasoning.end);
t.assert_equal("reasoning_end should be '</think>'", "</think>", analysis.reasoning.end);
// Check reasoning mode detection
// Nemotron uses forced closed reasoning with add_generation_prompt
t.assert_equal("reasoning should be FORCED_CLOSED", reasoning_mode::FORCED_CLOSED, analysis.reasoning.mode);
// Nemotron uses tag-based reasoning; prefill handles the template's forced markers
t.assert_equal("reasoning should be TAG_BASED", reasoning_mode::TAG_BASED, analysis.reasoning.mode);
// Make sure reasoning markers don't spill over to content markers
t.assert_equal("content start should be empty", "", analysis.content.start);