ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

mtmd : add ultravox audio input (#13623)

Browse source 
* convert ok, load ok

* warmup ok

* test

* still does not work?

* fix padding

* temporary give up

* fix merge conflict

* build_ultravox()

* rm test

* fix merge conflict

* add necessary mtmd APIs

* first working version (only 4s of audio)

* will this monster compile?

* fix compile

* please compile

* fPIC

* fix windows

* various fixes

* clean up audio_helpers

* fix conversion

* add some debug stuff

* long audio input ok

* adapt the api

* add --audio arg

* final touch UX

* add miniaudio to readme

* fix typo

* refactor kv metadata

* mtmd_default_marker()
This commit is contained in:
Xuan-Son Nguyen 2025-05-22 20:42:48 +02:00 • committed by GitHub
parent ab86335760
commit 797990c4bc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
21 changed files with 95401 additions and 259 deletions
Download patch file Download diff file Expand all files Collapse all files

62
gguf-py/gguf/tensor_mapping.py
View file

@ -1110,6 +1110,68 @@ class TensorNameMap:
MODEL_TENSOR.V_MM_PATCH_MERGER: (
"multi_modal_projector.patch_merger.merging_layer", # mistral small 3.1
),
# audio (mtmd)
MODEL_TENSOR.A_ENC_EMBD_POS: (
"audio_tower.embed_positions", # ultravox
),
MODEL_TENSOR.A_ENC_CONV1D: (
"audio_tower.conv{bid}", # ultravox
),
MODEL_TENSOR.A_PRE_NORM: (),
MODEL_TENSOR.A_POST_NORM: (
"audio_tower.layer_norm", # ultravox
),
MODEL_TENSOR.A_ENC_ATTN_Q: (
"audio_tower.layers.{bid}.self_attn.q_proj", # ultravox
),
MODEL_TENSOR.A_ENC_ATTN_K: (
"audio_tower.layers.{bid}.self_attn.k_proj", # ultravox
),
MODEL_TENSOR.A_ENC_ATTN_V: (
"audio_tower.layers.{bid}.self_attn.v_proj", # ultravox
),
MODEL_TENSOR.A_ENC_INPUT_NORM: (
"audio_tower.layers.{bid}.self_attn_layer_norm", # ultravox
),
MODEL_TENSOR.A_ENC_OUTPUT: (
"audio_tower.layers.{bid}.self_attn.out_proj", # ultravox
),
MODEL_TENSOR.A_ENC_OUTPUT_NORM: (
"audio_tower.layers.{bid}.final_layer_norm", # ultravox
),
MODEL_TENSOR.A_ENC_FFN_UP: (
"audio_tower.layers.{bid}.fc1", # ultravox
),
MODEL_TENSOR.A_ENC_FFN_GATE: (),
MODEL_TENSOR.A_ENC_FFN_DOWN: (
"audio_tower.layers.{bid}.fc2", # ultravox
),
MODEL_TENSOR.A_MMPROJ: (
"audio.multi_modal_projector.linear_{bid}", # ultravox
),
MODEL_TENSOR.A_MM_NORM_PRE: (
"audio.multi_modal_projector.ln_pre", # ultravox
),
MODEL_TENSOR.A_MM_NORM_MID: (
"audio.multi_modal_projector.ln_mid", # ultravox
),
}
# architecture-specific block mappings