sleepy/llama.cpp
1
0
Fork 0
Code Issues 8 Pull Requests Actions 170 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
common/sampling.cpp
+39 -6
View File
@@ -1,13 +1,16 @@
#include "sampling.h"
#include "common.h"
#include "ggml.h"
#include "log.h"
#include "reasoning-budget.h"
#include <algorithm>
#include <cctype>
#include <cmath>
#include <cstring>
#include <unordered_map>
#include <vector>
// the ring buffer works similarly to std::deque, but with a fixed capacity
// TODO: deduplicate with llama-impl.h
@@ -189,9 +192,10 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
std::vector<llama_sampler *> samplers;
if (params.grammar.compare(0, 11, "%llguidance") == 0) {
const std::string & grammar_str = common_grammar_value(params.grammar);
if (grammar_str.compare(0, 11, "%llguidance") == 0) {
#ifdef LLAMA_USE_LLGUIDANCE
grmr = llama_sampler_init_llg(vocab, "lark", params.grammar.c_str());
grmr = llama_sampler_init_llg(vocab, "lark", grammar_str.c_str());
#else
GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
#endif // LLAMA_USE_LLGUIDANCE
@@ -240,17 +244,46 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
trigger_patterns_c.push_back(regex.c_str());
}
if (!params.grammar.empty()) {
if (!grammar_str.empty()) {
if (params.grammar_lazy) {
grmr = llama_sampler_init_grammar_lazy_patterns(vocab, params.grammar.c_str(), "root",
grmr = llama_sampler_init_grammar_lazy_patterns(vocab, grammar_str.c_str(), "root",
trigger_patterns_c.data(), trigger_patterns_c.size(),
trigger_tokens.data(), trigger_tokens.size());
} else {
grmr = llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
grmr = llama_sampler_init_grammar(vocab, grammar_str.c_str(), "root");
}
}
}
// Feed generation prompt tokens to the grammar sampler so it advances past
// tokens the template already placed in the prompt.
// Only applies to output-format and tool-call grammars; user-supplied grammars must not be prefilled.
std::vector<llama_token> prefill_tokens;
if (!params.generation_prompt.empty() && common_grammar_needs_prefill(params.grammar)) {
GGML_ASSERT(vocab != nullptr);
prefill_tokens = common_tokenize(vocab, params.generation_prompt, false, true);
if (!prefill_tokens.empty()) {
std::string first_token = common_token_to_piece(vocab, prefill_tokens[0], true);
if (std::isspace(first_token[0]) && !std::isspace(params.generation_prompt[0])) {
// Some tokenizers will add a space before the first special token, need to remove
prefill_tokens = std::vector<llama_token>(prefill_tokens.begin() + 1, prefill_tokens.end());
}
}
if (grmr) {
try {
for (const auto & token : prefill_tokens) {
llama_sampler_accept(grmr, token);
LOG_DBG("%s: accepted prefill token (%d)\n", __func__, token);
}
} catch (std::exception &e) {
LOG_ERR("%s: error initializing grammar sampler for grammar:\n%s\n\nGeneration prompt:\n'%s'\n", __func__,
common_grammar_value(params.grammar).c_str(), params.generation_prompt.c_str());
throw e;
}
}
}
// reasoning budget sampler — added first so it can force tokens before other samplers
if (params.reasoning_budget_tokens >= 0 && !params.reasoning_budget_forced.empty()) {
samplers.push_back(common_reasoning_budget_init(
@@ -259,7 +292,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
params.reasoning_budget_end,
params.reasoning_budget_forced,
params.reasoning_budget_tokens,
params.reasoning_budget_activate_immediately ? REASONING_BUDGET_COUNTING : REASONING_BUDGET_IDLE));
prefill_tokens));
}
if (params.has_logit_bias()) {