| 1 | #include "models.h" |
| 2 | |
| 3 | ggml_cgraph * clip_graph_cogvlm::build() { |
| 4 | GGML_ASSERT(model.class_embedding != nullptr); |
| 5 | GGML_ASSERT(model.position_embeddings != nullptr); |
| 6 | |
| 7 | const int n_pos = n_patches + 1; // +1 for [CLS] |
| 8 | |
| 9 | // build input and concatenate class embedding |
| 10 | ggml_tensor * inp = build_inp(); |
| 11 | inp = ggml_concat(ctx0, inp, model.class_embedding, 1); |
| 12 | |
| 13 | inp = ggml_add(ctx0, inp, model.position_embeddings); |
| 14 | cb(inp, "inp_pos", -1); |
| 15 | |
| 16 | ggml_tensor * inpL = inp; |
| 17 | |
| 18 | for (int il = 0; il < n_layer; il++) { |
| 19 | auto & layer = model.layers[il]; |
| 20 | ggml_tensor * cur = inpL; |
| 21 | |
| 22 | cur = ggml_mul_mat(ctx0, layer.qkv_w, cur); |
| 23 | |
| 24 | cur = ggml_add(ctx0, cur, layer.qkv_b); |
| 25 | |
| 26 | ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, d_head, n_head, n_pos, d_head*sizeof(float), |
| 27 | cur->nb[1], 0); |
| 28 | ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, d_head, n_head, n_pos, d_head*sizeof(float), |
| 29 | cur->nb[1], n_embd * sizeof(float)); |
| 30 | ggml_tensor * Vcur = ggml_view_3d(ctx0, cur, d_head, n_head, n_pos, d_head*sizeof(float), |
| 31 | cur->nb[1], 2 * n_embd * sizeof(float)); |
| 32 | |
| 33 | cb(Qcur, "Qcur", il); |
| 34 | cb(Kcur, "Kcur", il); |
| 35 | cb(Vcur, "Vcur", il); |
| 36 | |
| 37 | cur = build_attn(layer.o_w, layer.o_b, |
| 38 | Qcur, Kcur, Vcur, nullptr, kq_scale, il); |
| 39 | cb(cur, "attn_out", il); |
| 40 | |
| 41 | cur = build_norm(cur, layer.ln_1_w, layer.ln_1_b, NORM_TYPE_NORMAL, eps, il); |
| 42 | cb(cur, "attn_post_norm", il); |
| 43 | |
| 44 | cur = ggml_add(ctx0, cur, inpL); |
| 45 | inpL = cur; |
| 46 | |
| 47 | cur = build_ffn(cur, |
| 48 | layer.ff_up_w, layer.ff_up_b, |
| 49 | layer.ff_gate_w, layer.ff_gate_b, |
| 50 | layer.ff_down_w, layer.ff_down_b, |
| 51 | hparams.ffn_op, il); |
| 52 | |
| 53 | cb(cur, "ffn_out", il); |
| 54 | |
| 55 | cur = build_norm(cur, layer.ln_2_w, layer.ln_2_b, NORM_TYPE_NORMAL, eps, il); |
| 56 | cb(cur, "ffn_post_norm", il); |
| 57 | |
| 58 | cur = ggml_add(ctx0, cur, inpL); |
| 59 | cb(cur, "layer_out", il); |
| 60 | inpL = cur; |
nothing calls this directly
no test coverage detected