mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 31 Packages Projects Releases Wiki Activity

mtmd: add chunks and fix preproc for qwen3a (#23073)

Browse Source 
* mtmd: add chunks and fix preproc for qwen3a

* add attn_mask

* limit mtmd_chunk size (avoid blow up memory)

* correct audio tokens

* re-order the set_input case

* remove attn_mask
This commit is contained in:
Xuan-Son Nguyen
2026-05-15 19:32:47 +02:00
committed by GitHub
parent 8be1786707
commit 72e60f500d
7 changed files with 200 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/mtmd.cpp
+6 -1
View File
@@ -515,7 +515,6 @@ struct mtmd_context {
// set preprocessor
switch (proj) {
case PROJECTOR_TYPE_QWEN2A:
case PROJECTOR_TYPE_QWEN3A:
case PROJECTOR_TYPE_QWEN25O:
{
// <|audio_bos|> ... (embeddings) ... <|audio_eos|>
@@ -523,6 +522,12 @@ struct mtmd_context {
aud_end = "<|audio_eos|>";
audio_preproc = std::make_unique<mtmd_audio_preprocessor_whisper>(ctx_a);
} break;
case PROJECTOR_TYPE_QWEN3A:
{
aud_beg = "<|audio_start|>";
aud_end = "<|audio_end|>";
audio_preproc = std::make_unique<mtmd_audio_preprocessor_qwen3a>(ctx_a);
} break;
case PROJECTOR_TYPE_VOXTRAL:
{
// [BEGIN_AUDIO] ... (embeddings) ...