echodict/llama.cpp

1#include "models.h"2 3ggml_cgraph * clip_graph_pixtral::build() {4    const int n_merge = hparams.n_merge;5 6    // 2D input positions7    ggml_tensor * pos_h = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_patches);8    ggml_set_name(pos_h, "pos_h");9    ggml_set_input(pos_h);10 11    ggml_tensor * pos_w = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_patches);12    ggml_set_name(pos_w, "pos_w");13    ggml_set_input(pos_w);14 15    auto add_pos = [&](ggml_tensor * cur, const clip_layer &) {16        return build_rope_2d(ctx0, cur, pos_h, pos_w, hparams.rope_theta, true);17    };18 19    ggml_tensor * inp = build_inp();20    ggml_tensor * cur = build_vit(21                            inp, n_patches,22                            NORM_TYPE_RMS,23                            hparams.ffn_op,24                            nullptr, // no learned pos embd25                            add_pos);26 27    // mistral small 3.1 patch merger28    // ref: https://github.com/huggingface/transformers/blob/7a3e208892c06a5e278144eaf38c8599a42f53e7/src/transformers/models/mistral3/modeling_mistral3.py#L6729    if (model.mm_patch_merger_w) {30        GGML_ASSERT(hparams.n_merge > 0);31 32        cur = ggml_mul(ctx0, ggml_rms_norm(ctx0, cur, eps), model.mm_input_norm_w);33 34        // reshape image tokens to 2D grid35        cur = ggml_reshape_3d(ctx0, cur, n_embd, n_patches_x, n_patches_y);36        cur = ggml_permute(ctx0, cur, 2, 0, 1, 3); // [x, y, n_embd]37        cur = ggml_cont(ctx0, cur);38 39        // torch.nn.functional.unfold is just an im2col under the hood40        // we just need a dummy kernel to make it work41        ggml_tensor * kernel = ggml_view_3d(ctx0, cur, n_merge, n_merge, cur->ne[2], 0, 0, 0);42        cur = ggml_im2col(ctx0, kernel, cur, n_merge, n_merge, 0, 0, 1, 1, true, inp->type);43 44        // project to n_embd45        cur = ggml_reshape_2d(ctx0, cur, cur->ne[0], cur->ne[1] * cur->ne[2]);46        cur = build_mm(model.mm_patch_merger_w, cur);47    }48 49    // LlavaMultiModalProjector (always using GELU activation)50    {51        cur = build_ffn(cur,52            model.mm_1_w, model.mm_1_b,53            nullptr, nullptr,54            model.mm_2_w, model.mm_2_b,55            FFN_GELU,56            -1);57    }58 59    // arrangement of the [IMG_BREAK] token60    if (model.token_embd_img_break) {61        // not efficient, but works62        // the trick is to view the embeddings as a 3D tensor with shape [n_embd, n_patches_per_row, n_rows]63        // and then concatenate the [IMG_BREAK] token to the end of each row, aka n_patches_per_row dimension64        // after the concatenation, we have a tensor with shape [n_embd, n_patches_per_row + 1, n_rows]65 66        const int p_y             = n_merge > 0 ? n_patches_y / n_merge : n_patches_y;67        const int p_x             = n_merge > 0 ? n_patches_x / n_merge : n_patches_x;68        const int p_total         = p_x * p_y;69        const int n_embd_text     = cur->ne[0];70        const int n_tokens_output = p_total + p_y - 1; // one [IMG_BREAK] per row, except the last row71 72        ggml_tensor * tmp = ggml_reshape_3d(ctx0, cur, n_embd_text, p_x, p_y);73        ggml_tensor * tok = ggml_new_tensor_3d(ctx0, tmp->type, n_embd_text, 1, p_y);74        tok = ggml_scale(ctx0, tok, 0.0); // clear the tensor75        tok = ggml_add(ctx0, tok, model.token_embd_img_break);76        tmp = ggml_concat(ctx0, tmp, tok, 1);77        cur = ggml_view_2d(ctx0, tmp,78            n_embd_text, n_tokens_output,79            ggml_row_size(tmp->type, n_embd_text), 0);80    }81 82    // build the graph83    ggml_build_forward_expand(gf, cur);84 85    return gf;86}87