vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-10-03 03:25:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

metadata forcing

Browse source
This commit is contained in:
Concedo 2026-09-24 00:20:36 +08:00
parent 444b9e4419
commit 60acd9b92f
1 changed files with 47 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

61
otherarch/acestep/ace-qwen3.cpp
View file

@ -589,10 +589,21 @@ struct MetadataFSM {
value_acc.clear();
}
// Force FSM to only allow a specific language value
void force_language(BPETokenizer & bpe, const std::string & lang) {
language_tree = PrefixTree();
build_value_tree(bpe, language_tree, "language:", {lang});
// Force a user-supplied metadata value. The constrained token sequence is
// emitted into the CoT, so the LM sees it before generating lyrics/codes.
void force_field(BPETokenizer & bpe, State value_state, const std::string & value) {
PrefixTree * tree = nullptr;
const char * prefix = nullptr;
switch (value_state) {
case BPM_VALUE: tree = &bpm_tree; prefix = "bpm:"; break;
case DURATION_VALUE: tree = &duration_tree; prefix = "duration:"; break;
case KEYSCALE_VALUE: tree = &keyscale_tree; prefix = "keyscale:"; break;
case LANGUAGE_VALUE: tree = &language_tree; prefix = "language:"; break;
case TIMESIG_VALUE: tree = &timesig_tree; prefix = "timesignature:"; break;
default: return;
}
*tree = PrefixTree();
build_value_tree(bpe, *tree, prefix, {value});
}
const std::vector<int> * current_name_tokens() const {
@ -787,11 +798,15 @@ static void parse_phase1_into_aces(
if (!parse_cot_and_lyrics(texts[i], &parsed))
fprintf(stderr, "WARNING: batch %d CoT parse incomplete\n", i);
aces[i] = base;
if (parsed.bpm > 0) aces[i].bpm = parsed.bpm;
if (parsed.duration > 0) aces[i].duration = parsed.duration;
if (!parsed.keyscale.empty()) aces[i].keyscale = parsed.keyscale;
if (!parsed.timesignature.empty()) aces[i].timesignature = parsed.timesignature;
if (!parsed.vocal_language.empty()) aces[i].vocal_language = parsed.vocal_language;
// User-supplied metadata is authoritative. Phase 1 only fills fields
// that were absent from the request.
if (parsed.bpm > 0 && base.bpm <= 0) aces[i].bpm = parsed.bpm;
if (parsed.duration > 0 && base.duration <= 0) aces[i].duration = parsed.duration;
if (!parsed.keyscale.empty() && base.keyscale.empty()) aces[i].keyscale = parsed.keyscale;
if (!parsed.timesignature.empty() && base.timesignature.empty()) aces[i].timesignature = parsed.timesignature;
if (!parsed.vocal_language.empty() &&
(base.vocal_language.empty() || base.vocal_language == "unknown"))
aces[i].vocal_language = parsed.vocal_language;
if (!parsed.caption.empty()) aces[i].caption = parsed.caption;
if (merge_lyrics && !parsed.lyrics.empty()) aces[i].lyrics = parsed.lyrics;
if (aces[i].duration <= 0) aces[i].duration = 120.0f;
@ -1391,6 +1406,24 @@ std::string acestep_prepare_request(const music_generation_inputs inputs)
ace.timesignature = req.timesignature;
ace.vocal_language = req.vocal_language;
auto force_known_metadata = [&](MetadataFSM & target, const AcePrompt & values) {
if (values.bpm > 0)
target.force_field(acestep_bpe, MetadataFSM::BPM_VALUE,
std::to_string(values.bpm));
if (values.duration > 0)
target.force_field(acestep_bpe, MetadataFSM::DURATION_VALUE,
std::to_string((int)values.duration));
if (!values.keyscale.empty())
target.force_field(acestep_bpe, MetadataFSM::KEYSCALE_VALUE,
values.keyscale);
if (!values.vocal_language.empty() && values.vocal_language != "unknown")
target.force_field(acestep_bpe, MetadataFSM::LANGUAGE_VALUE,
values.vocal_language);
if (!values.timesignature.empty())
target.force_field(acestep_bpe, MetadataFSM::TIMESIG_VALUE,
values.timesignature);
};
bool user_has_codes = !req.audio_codes.empty();
bool need_lm_codes = inputs.gen_codes && !user_has_codes;
@ -1421,10 +1454,10 @@ std::string acestep_prepare_request(const music_generation_inputs inputs)
+ std::string(req.instrumental ? "true" : "false");
prompt = build_custom_prompt(acestep_bpe, sys, user_msg.c_str());
// FSM: reset then optionally force language (shared for both paths)
// Inject supplied metadata into the generated CoT so lyrics are planned
// for the requested duration/tempo/key rather than newly guessed values.
fsm.reset();
if (use_fsm && ace.vocal_language != "unknown" && !ace.vocal_language.empty())
fsm.force_language(acestep_bpe, ace.vocal_language);
if (use_fsm) force_known_metadata(fsm, ace);
// Phase 1: N lyrics + metadata generations (always batched, N=batch_size)
fprintf(stderr, "[Simple] %zu tokens, N=%d, seeds: %lld..%lld\n",
@ -1446,8 +1479,7 @@ std::string acestep_prepare_request(const music_generation_inputs inputs)
MetadataFSM caption_fsm;
caption_fsm.init(acestep_bpe, acestep_llm.cfg.vocab_size, true);
caption_fsm.reset();
if (aces[0].vocal_language != "unknown" && !aces[0].vocal_language.empty())
caption_fsm.force_language(acestep_bpe, aces[0].vocal_language);
if (use_fsm) force_known_metadata(caption_fsm, aces[0]);
auto caption_texts = generate_phase1_batch(
&acestep_llm, &acestep_bpe, prompt, 768,
@ -1485,6 +1517,7 @@ std::string acestep_prepare_request(const music_generation_inputs inputs)
prompt.size(), phase1cfg, batch_size, seed, seed + batch_size - 1);
fsm.reset();
if (use_fsm) force_known_metadata(fsm, ace);
auto phase1_texts = generate_phase1_batch(
&acestep_llm, &acestep_bpe, prompt, 2048, temperature, top_p, top_k, rep_pen,
seed, batch_size, use_fsm ? &fsm : nullptr, false,