model : add LiquidAI/d1-omni-600M decision model (#30114)

* model : add LiquidAI/d1-omni-600M decision model

Assisted-by: Claude Opus 5.5

* mtmd : keep conformer GLU sigmoid on CUDA

Assisted-by: Claude Opus 5.5

* server : take d1omni audio through images and input_audio, scope memory-less lfm2 to non-causal

Assisted-by: Claude Opus 5.5

* common : rename decision type d1omni to lfm2-d1-omni, server : make images an alias of files

Assisted-by: Claude Opus 5.5
This commit is contained in:
Tarek Dakhran authored and GitHub committed 2026-10-08 02:07:49 +02:00
1 parent aa5e0092fd
commit a657f7e981
21 files changed
+833 -35

No files matched your search

+9
View File
@@ -871,6 +871,11 @@ struct mtmd_context {
ov_img_first = false;
image_preproc = std::make_unique<mtmd_image_preprocessor_lfm2>(ctx_v);
} break;
case PROJECTOR_TYPE_D1OMNI_V:
{
// same tiles and thumbnail as lfm2, without separator tokens
image_preproc = std::make_unique<mtmd_image_preprocessor_lfm2>(ctx_v);
} break;
case PROJECTOR_TYPE_GLM4V:
{
// <|begin_of_image|> ... (image embeddings) ... <|end_of_image|>
@@ -982,6 +987,10 @@ struct mtmd_context {
{
audio_preproc = std::make_unique<mtmd_audio_preprocessor_conformer>(ctx_a);
} break;
case PROJECTOR_TYPE_D1OMNI_A:
{
audio_preproc = std::make_unique<mtmd_audio_preprocessor_d1omni>(ctx_a);
} break;
case PROJECTOR_TYPE_GRANITE_SPEECH:
{
audio_preproc = std::make_unique<mtmd_audio_preprocessor_granite_speech>(ctx_a);