upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-01 18:37:28 -05:00
Code Issues Packages Projects Releases Wiki Activity

mtmd: support multi-row batching for deepseek-ocr

Browse Source
This commit is contained in:
Xuan Son Nguyen
2026-07-27 01:34:10 +02:00
parent d4d057b6dd
commit 6b0b1ea112
3 changed files with 74 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/models/deepseekocr.cpp
+62 -26
View File
@@ -253,6 +253,9 @@ ggml_cgraph * clip_graph_deepseekocr::build() {
bool is_overview = img.add_viewsep;
int n_tiles_per_row = 0;
// number of separate "row" images batched together in this graph call
// (captured now, before n_batch below gets repurposed as the SAM/ViT batch size)
const int n_rows_batch = n_batch;
// note: we expect either a batch of rows or a batch of overviews, but not a mix of both
@@ -272,16 +275,30 @@ ggml_cgraph * clip_graph_deepseekocr::build() {
GGML_ASSERT(img.ny() % img.nx() == 0);
n_tiles_per_row = img.ny() / img.nx();
// input shape: [tile_size, tile_size * n_tiles_per_row, 3]
// we want to reshape it to [tile_size, tile_size, 3, n_tiles_per_row]
inp_raw = ggml_reshape_4d(ctx0, inp_raw, img.nx(), img.nx(), n_tiles_per_row, 3);
inp_raw = ggml_cont(ctx0, ggml_permute(ctx0, inp_raw, 0, 1, 3, 2));
// each entry is one "row" image of shape [tile_size, tile_size * n_tiles_per_row, 3];
// reshape+concat all rows into one combined SAM input of shape
// [tile_size, tile_size, 3, n_tiles_per_row * n_rows_batch] (tile fast, row slow)
ggml_tensor * combined = nullptr;
for (int r = 0; r < n_rows_batch; r++) {
ggml_tensor * row = ggml_view_4d(ctx0, inp_raw, img.nx(), img.ny(), 3, 1,
inp_raw->nb[1], inp_raw->nb[2], inp_raw->nb[3],
r * inp_raw->nb[3]);
row = ggml_cont(ctx0, row);
// input shape: [tile_size, tile_size * n_tiles_per_row, 3, 1]
// we want to reshape it to [tile_size, tile_size, 3, n_tiles_per_row]
row = ggml_reshape_4d(ctx0, row, img.nx(), img.nx(), n_tiles_per_row, 3);
row = ggml_cont(ctx0, ggml_permute(ctx0, row, 0, 1, 3, 2));
combined = combined ? ggml_concat(ctx0, combined, row, 3) : row;
}
inp_raw = combined;
}
ggml_tensor * sam_out = build_sam(inp_raw);
if (!is_overview) {
n_batch = n_tiles_per_row;
n_batch = n_tiles_per_row * n_rows_batch;
}
const int clip_n_patches = sam_out->ne[0] * sam_out->ne[1];
@@ -354,34 +371,53 @@ ggml_cgraph * clip_graph_deepseekocr::build() {
const auto w = h;
const auto n_dim = cur->ne[0];
ggml_tensor * imgnl = ggml_repeat_4d(ctx0, model.image_newline, n_dim, 1, h, 1);
cur = ggml_reshape_3d(ctx0, cur, n_dim, w, h);
cur = ggml_reshape_2d(ctx0, ggml_concat(ctx0, cur, imgnl, 1), n_dim, (w + 1) * h);
cur = ggml_concat(ctx0, cur, model.view_seperator, 1); // (n_dim, h*(w+1) + 1)
ggml_tensor * imgnl = ggml_repeat_4d(ctx0, model.image_newline, n_dim, 1, h, n_batch);
cur = ggml_reshape_4d(ctx0, cur, n_dim, w, h, n_batch);
cur = ggml_reshape_3d(ctx0, ggml_concat(ctx0, cur, imgnl, 1), n_dim, (w + 1) * h, n_batch);
ggml_tensor * vs = ggml_repeat_4d(ctx0, model.view_seperator, n_dim, 1, n_batch, 1);
cur = ggml_concat(ctx0, cur, vs, 1); // (n_dim, h*(w+1) + 1, n_batch)
} else {
// tile row: interleave tiles within each row, add newline per row
const int grid_x = static_cast<int>(std::sqrt(static_cast<float>(clip_n_patches)));
const int grid_y = grid_x;
const auto n_dim = cur->ne[0];
// this weave is done per-row (cheap reshuffle ops) since a single row's
// interleave already uses all 4 ggml tensor axes, leaving no spare axis to
// also carry the n_rows_batch dimension through in one shot
const int grid_x = static_cast<int>(std::sqrt(static_cast<float>(clip_n_patches)));
const int grid_y = grid_x;
const auto n_dim = cur->ne[0];
// (n_dim, clip_n_patches, n_batch) -> (n_dim, grid_x, grid_y, n_batch)
cur = ggml_reshape_4d(ctx0, cur, n_dim, grid_x, grid_y, n_batch);
// (n_dim, clip_n_patches, n_tiles_per_row * n_rows_batch) -> (n_dim, clip_n_patches, n_tiles_per_row, n_rows_batch)
ggml_tensor * cur4 = ggml_reshape_4d(ctx0, cur, n_dim, clip_n_patches, n_tiles_per_row, n_rows_batch);
// tiles: re-order from A.row0 A.row1 B.row0 B.row1 ...
// to A.row0 B.row0 A.row1 B.row1 ...
// then add nl: A.row0 B.row0 [nl] A.row1 B.row1 [nl] ...
// interleave tiles: (n_dim, grid_x, grid_y, n_batch) -> (n_dim, grid_x, n_batch, grid_y)
cur = ggml_cont(ctx0, ggml_permute(ctx0, cur, 0, 1, 3, 2));
ggml_tensor * rows_out = nullptr;
for (int r = 0; r < n_rows_batch; r++) {
ggml_tensor * row = ggml_view_3d(ctx0, cur4, n_dim, clip_n_patches, n_tiles_per_row,
cur4->nb[1], cur4->nb[2], r * cur4->nb[3]);
row = ggml_cont(ctx0, row);
// merge: (n_dim, grid_x, n_batch, grid_y) -> (n_dim, grid_x*n_batch, grid_y, 1)
cur = ggml_reshape_4d(ctx0, cur, n_dim, grid_x * n_batch, grid_y, 1);
// (n_dim, clip_n_patches, n_tiles_per_row) -> (n_dim, grid_x, grid_y, n_tiles_per_row)
row = ggml_reshape_4d(ctx0, row, n_dim, grid_x, grid_y, n_tiles_per_row);
// append newline per row: (n_dim, grid_x*n_batch+1, grid_y, 1)
ggml_tensor * imgnl = ggml_repeat_4d(ctx0, model.image_newline, n_dim, 1, grid_y, 1);
cur = ggml_concat(ctx0, cur, imgnl, 1);
// tiles: re-order from A.row0 A.row1 B.row0 B.row1 ...
// to A.row0 B.row0 A.row1 B.row1 ...
// then add nl: A.row0 B.row0 [nl] A.row1 B.row1 [nl] ...
// interleave tiles: (n_dim, grid_x, grid_y, n_tiles_per_row) -> (n_dim, grid_x, n_tiles_per_row, grid_y)
row = ggml_cont(ctx0, ggml_permute(ctx0, row, 0, 1, 3, 2));
// flatten: (n_dim, (grid_x*n_batch+1)*grid_y)
cur = ggml_reshape_2d(ctx0, cur, n_dim, (grid_x * n_batch + 1) * grid_y);
// merge: (n_dim, grid_x, n_tiles_per_row, grid_y) -> (n_dim, grid_x*n_tiles_per_row, grid_y, 1)
row = ggml_reshape_4d(ctx0, row, n_dim, grid_x * n_tiles_per_row, grid_y, 1);
// append newline per row: (n_dim, grid_x*n_tiles_per_row+1, grid_y, 1)
ggml_tensor * imgnl = ggml_repeat_4d(ctx0, model.image_newline, n_dim, 1, grid_y, 1);
row = ggml_concat(ctx0, row, imgnl, 1);
// flatten: (n_dim, (grid_x*n_tiles_per_row+1)*grid_y, 1)
row = ggml_reshape_3d(ctx0, row, n_dim, (grid_x * n_tiles_per_row + 1) * grid_y, 1);
rows_out = rows_out ? ggml_concat(ctx0, rows_out, row, 2) : row;
}
// (n_dim, (grid_x*n_tiles_per_row+1)*grid_y, n_rows_batch)
cur = rows_out;
}
cb(cur, "dsocr_output", -1);
tools/mtmd/models/deepseekocr2.cpp
+10 -6
View File
@@ -14,8 +14,9 @@ ggml_cgraph * clip_graph_deepseekocr2::build() {
{
ggml_tensor * inp;
inp = ggml_reshape_2d(ctx0, sam_out, sam_out->ne[0] * sam_out->ne[1], sam_out->ne[2]); // H*W, C
inp = ggml_cont(ctx0, ggml_permute(ctx0, inp, 1, 0, 2, 3));
// H*W, C, B
inp = ggml_reshape_3d(ctx0, sam_out, sam_out->ne[0] * sam_out->ne[1], sam_out->ne[2], sam_out->ne[3]);
inp = ggml_cont(ctx0, ggml_permute(ctx0, inp, 1, 0, 2, 3)); // C, H*W, B
auto num_image_tokens = inp->ne[1]; // H*W
GGML_ASSERT(num_image_tokens == 144 || num_image_tokens == 256);
@@ -32,8 +33,10 @@ ggml_cgraph * clip_graph_deepseekocr2::build() {
num_queries = 144;
}
// (B, num_image_tokens + num_queries, C)
inp = ggml_concat(ctx0, inp, ggml_cast(ctx0, query_embed, inp->type), 1);
// repeat the query embedding per batch item, then append: (C, num_image_tokens + num_queries, B)
query_embed = ggml_cast(ctx0, query_embed, inp->type);
query_embed = ggml_repeat_4d(ctx0, query_embed, query_embed->ne[0], num_queries, inp->ne[2], 1);
inp = ggml_concat(ctx0, inp, query_embed, 1);
auto seq_len = inp->ne[1];
@@ -57,7 +60,7 @@ ggml_cgraph * clip_graph_deepseekocr2::build() {
/* learned_pos_embd */ nullptr, add_rope, vit_opts);
cur = ggml_cont(ctx0,
ggml_view_2d(ctx0, cur, cur->ne[0], num_queries, cur->nb[1],
ggml_view_3d(ctx0, cur, cur->ne[0], num_queries, cur->ne[2], cur->nb[1], cur->nb[2],
cur->nb[1] * (cur->ne[1] - num_queries))); // only take query tokens for output
ggml_build_forward_expand(gf, cur);
@@ -71,7 +74,8 @@ ggml_cgraph * clip_graph_deepseekocr2::build() {
// view_seperator only after the global view
if (img.add_viewsep) {
cur = ggml_concat(ctx0, cur, model.view_seperator, 1); // (n_dim, 257)
ggml_tensor * vs = ggml_repeat_4d(ctx0, model.view_seperator, model.view_seperator->ne[0], 1, cur->ne[2], 1);
cur = ggml_concat(ctx0, cur, vs, 1); // (n_dim, 257, n_batch)
}
cb(cur, "dsocr2_output", -1);
tools/mtmd/models/models.h
+2 -1
View File
@@ -127,12 +127,13 @@ struct clip_graph_deepseekocr : clip_graph {
clip_graph_deepseekocr(clip_ctx * ctx, const clip_image_f32 & img) : clip_graph(ctx, img) {}
ggml_cgraph * build() override;
ggml_tensor * build_sam(ggml_tensor * inp); // build the SAM model
// bool support_batch() const override { return true; } // TODO: support batch for DeepSeek-OCR v1
bool support_batch() const override { return true; }
};
struct clip_graph_deepseekocr2 : clip_graph_deepseekocr {
clip_graph_deepseekocr2(clip_ctx * ctx, const clip_image_f32 & img) : clip_graph_deepseekocr(ctx, img) {}
ggml_cgraph * build() override; // reuses build_sam() from base
bool support_batch() const override { return true; }
};
struct clip_graph_conformer : clip_graph {