LLM-Inference/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-12 22:31:11 +04:00
Code Issues Packages Projects Releases Wiki Activity

mtmd: stop feeding the text stream again during Qwen3-TTS generation (#26706)

Browse Source 
The reference implementation has two mutually exclusive prompt layouts.
In non streaming mode the prefill carries the whole utterance text plus
tts_eos summed with codec_pad, and the trailing text hidden collapses to
a single tts_pad row. In streaming mode the prefill carries only the
first text token and the trailing rows stream the rest of the text
followed by tts_eos.

The pipeline built the non streaming prefill but the streaming overlay,
so the talker saw the utterance a second time during generation and read
it twice before emitting codec_eos.

The overlay is now the single tts_pad row that matches the prefill.
This commit is contained in:
Pascal
2026-08-07 13:32:52 +02:00
committed by GitHub
parent cb26014d96
commit 217df17ac3
1 changed files with 5 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/mtmd-helper-gen.cpp
+5 -11
View File
@@ -112,7 +112,6 @@ public:
c2w_state.clear();
audio_pcm.clear();
overlay.clear();
overlay_idx = 0;
h_state_buf.clear();
out_buf.clear();
prompt_embd_buf.clear();
@@ -205,11 +204,9 @@ public:
top_p = inp->top_p > 0 ? inp->top_p : 1.0f;
out_type = inp->out_type;
// the text stream keeps flowing during generation: after frame k, the input adds
// trailing text row k on top of the codes embedding, then tts_eos, then tts_pad
for (int i = 3; i < n_ids - 5; i++) overlay.push_back(row(ids[(size_t) i]));
overlay.push_back(row(tts_eos));
overlay.push_back(row(tts_pad));
// the prompt above holds the whole text stream up to tts_eos, so every generated
// frame adds tts_pad on top of the codes embedding
overlay = row(tts_pad);
return 0;
}
@@ -265,9 +262,7 @@ public:
}
std::vector<float> fb(out.embd, out.embd + n_embd);
const auto & ov = overlay[std::min(overlay_idx, overlay.size() - 1)];
for (int i = 0; i < n_embd; i++) fb[(size_t) i] += ov[(size_t) i];
overlay_idx++;
for (int i = 0; i < n_embd; i++) fb[(size_t) i] += overlay[(size_t) i];
const int n_pos_per_embd = mrope ? 4 : 1;
decode_embd_batch batch_embd(fb.data(), 1, n_pos_per_embd, n_embd);
@@ -437,8 +432,7 @@ private:
std::vector<int32_t> codes_buf;
std::vector<uint8_t> c2w_state;
std::vector<float> audio_pcm;
std::vector<std::vector<float>> overlay;
size_t overlay_idx = 0;
std::vector<float> overlay;
std::vector<float> h_state_buf;
mtmd_helper_gen_audio_outtype out_type = MTMD_HELPER_GEN_AUDIO_OUTTYPE_WAV;
std::vector<char> out_buf;