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

mtmd : support Qwen 2.5 Omni (input audio+vision, no audio output) (#13784)

Browse source 
* mtmd : allow multiple modalities at the same time

* refactor mtmd tokenizer

* fix compile

* ok, missing SinusoidsPositionEmbedding

* first working version

* fix style

* more strict validate of n_embd

* refactor if..else to switch

* fix regression

* add test for 3B

* update docs

* fix tokenizing with add_special

* add more tests

* fix test case "huge"

* rm redundant code

* set_position_mrope_1d rm n_tokens
This commit is contained in:
Xuan-Son Nguyen 2025-05-27 14:06:10 +02:00 • committed by GitHub
parent 72b090da2c
commit bc583e3c63
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 1148 additions and 744 deletions
Download patch file Download diff file Expand all files Collapse all files

44
tools/mtmd/mtmd-helper.cpp
View file

@ -66,7 +66,8 @@ struct decode_embd_batch {
}
}
void set_position_mrope(llama_pos pos_0, int nx, int ny, llama_seq_id seq_id) {
// M-RoPE for image
void set_position_mrope_2d(llama_pos pos_0, int nx, int ny, llama_seq_id seq_id) {
GGML_ASSERT(n_pos_per_embd == 4);
seq_id_0[0] = seq_id;
for (int y = 0; y < ny; y++) {
@ -85,6 +86,23 @@ struct decode_embd_batch {
}
}
// M-RoPE for audio
void set_position_mrope_1d(llama_pos pos_0, llama_seq_id seq_id) {
GGML_ASSERT(n_pos_per_embd == 4);
seq_id_0[0] = seq_id;
for (int i = 0; i < batch.n_tokens; i++) {
pos[i ] = pos_0 + i;
pos[i + batch.n_tokens ] = pos_0 + i;
pos[i + batch.n_tokens * 2] = pos_0 + i;
pos[i + batch.n_tokens * 3] = 0; // last pos dim is unused
}
for (int i = 0; i < batch.n_tokens; i++) {
batch.n_seq_id[i] = 1;
batch.seq_id [i] = seq_id_0.data();
batch.logits [i] = false;
}
}
llama_batch get_view(int offset, int n_tokens) {
llama_pos * pos_ptr;
pos_view.clear();
@ -146,18 +164,20 @@ int32_t mtmd_helper_decode_image_chunk(
decode_embd_batch batch_embd(encoded_embd, n_tokens, n_pos_per_embd, n_mmproj_embd);
if (mtmd_decode_use_mrope(ctx)) {
const auto image_tokens = mtmd_input_chunk_get_tokens_image(chunk);
if (chunk_type != MTMD_INPUT_CHUNK_TYPE_IMAGE) {
LOG_ERR("failed to decode chunk: M-RoPE only accepts image chunk\n");
return -1;
if (chunk_type == MTMD_INPUT_CHUNK_TYPE_IMAGE) {
const auto image_tokens = mtmd_input_chunk_get_tokens_image(chunk);
if (!image_tokens) {
LOG_ERR("failed to decode chunk: image tokens are null\n");
return -1;
}
const int nx = mtmd_image_tokens_get_nx(image_tokens);
const int ny = mtmd_image_tokens_get_ny(image_tokens);
batch_embd.set_position_mrope_2d(n_past, nx, ny, seq_id);
} else if (chunk_type == MTMD_INPUT_CHUNK_TYPE_AUDIO) {
batch_embd.set_position_mrope_1d(n_past, seq_id);
} else {
GGML_ABORT("invalid chunk type for M-RoPE");
}
if (!image_tokens) {
LOG_ERR("failed to decode chunk: image tokens are null\n");
return -1;
}
const int nx = mtmd_image_tokens_get_nx(image_tokens);
const int ny = mtmd_image_tokens_get_ny(image_tokens);
batch_embd.set_position_mrope(n_past, nx, ny, seq_id);
} else {
batch_embd.set_position_normal(n_past, seq_id);
}