upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-07-29 22:30:43 -05:00
Code Issues Packages Projects Releases Wiki Activity

add a few print tensor; tweak in sample again

Browse Source
This commit is contained in:
bssrdf
2024-02-13 21:44:17 -05:00
parent c122507fa1
commit 539a94adc7
4 changed files with 26 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
clip.hpp
View File

@@ -963,22 +963,22 @@ struct CLIPVisionModel {
ggml_set_name(x, "id_pixel");
ggml_set_name(temp, "temp_input");
int64_t* ne = patch_embeddings->ne;
struct ggml_tensor *patch_embeddings_f16 = ggml_reshape_3d(ctx0, patch_embeddings, ne[0], ne[1], ne[2]*ne[3]);
patch_embeddings_f16 = ggml_cast(ctx0, patch_embeddings_f16, GGML_TYPE_F16);
patch_embeddings_f16 = ggml_reshape_4d(ctx0, patch_embeddings_f16, ne[0], ne[1], ne[2], ne[3]);
// struct ggml_tensor *patch_embeddings_f16 = ggml_reshape_3d(ctx0, patch_embeddings, ne[0], ne[1], ne[2]*ne[3]);
// patch_embeddings_f16 = ggml_cast(ctx0, patch_embeddings_f16, GGML_TYPE_F16);
// patch_embeddings_f16 = ggml_reshape_4d(ctx0, patch_embeddings_f16, ne[0], ne[1], ne[2], ne[3]);
struct ggml_tensor *patch_embeddings_f16 = ggml_cast(ctx0, patch_embeddings, GGML_TYPE_F16);
struct ggml_tensor * inp = ggml_conv_2d(ctx0, patch_embeddings_f16, x, patch_size, patch_size, 0, 0, 1, 1);
ggml_set_name(inp, "inp_conv_2d");
// print_ggml_tensor(inp, true, "inp_conv_2d");
print_ggml_tensor(inp, true, "inp_conv_2d");
inp = ggml_reshape_3d(ctx0, inp, num_patches, hidden_size, batch_size);
ggml_set_name(inp, "inp_reshape_3d");
// print_ggml_tensor(inp, true, "inp_reshape_3d");
print_ggml_tensor(inp, true, "inp_reshape_3d");
// inp = ggml_cont(ctx0, ggml_permute(ctx0, inp, 1, 0, 2, 3));
// print_ggml_tensor(ggml_permute(ctx0, inp, 2, 0, 1, 3), true, "inp_permute");
inp = ggml_cont(ctx0, ggml_permute(ctx0, inp, 2, 0, 1, 3));
// print_ggml_tensor(inp, true, "inp_cont");
print_ggml_tensor(inp, true, "inp_cont");
ggml_set_name(inp, "inp_cont");
ggml_set_name(class_embedding, "class_embedding");
@@ -996,7 +996,7 @@ struct CLIPVisionModel {
ggml_set_name(class_embedding_rep, "class_embedding_rep");
// print_ggml_tensor(class_embedding_rep, true, "class_embedding_rep");
// class_embedding_rep = ggml_cast(ctx0, class_embedding_rep, inp->type);
// print_ggml_tensor(class_embedding_rep, true, "class_embedding_rep_aft_casting");
print_ggml_tensor(class_embedding_rep, true, "class_embedding_rep_aft_casting");
struct ggml_tensor *embeddings = ggml_concat(ctx0, class_embedding_rep, inp);
ggml_set_name(embeddings, "embeddings_after_concat");
// print_ggml_tensor(embeddings, true, "embeddings_after_concat");
@@ -1027,10 +1027,12 @@ struct CLIPVisionModel {
embeddings = resblocks[i].forward(ctx0, embeddings); // [N, n_token, hidden_size]
}
ggml_set_name(embeddings, "embeddings_after_transformer");
print_ggml_tensor(embeddings, true, "embeddings_after_encoder");
// get the output of cls token, e.g., 0th index
embeddings = ggml_get_rows(ctx0, ggml_reshape_2d(ctx0, embeddings, hidden_size, num_positions * batch_size), cls);
ggml_set_name(embeddings, "embeddings_after_cls_token");
print_ggml_tensor(embeddings, true, "embeddings_pooled_after_encoder");
// post-layernorm
embeddings = ggml_nn_layer_norm(ctx0, embeddings, post_ln_w, post_ln_b);

4
model.cpp
View File

@@ -789,8 +789,8 @@ bool ModelLoader::init_from_safetensors_file(const std::string& file_path, const
ne[i] = shape[i].get<int64_t>();
}
TensorStorage tensor_storage(prefix + name, type, ne, n_dims, file_index, ST_HEADER_SIZE_LEN + header_size_ + begin);
// if(strcmp(prefix.c_str(), "pmid") == 0)
// printf(" %s:%s [%zu, %zu, %zu, %zu]\n", (prefix + name).c_str(), ggml_type_name(type), ne[0], ne[1],ne[2], ne[3]);
// if(starts_with(prefix, "pmi"))
// printf(" %s:%s [%zu, %zu, %zu, %zu]\n", (prefix + name).c_str(), ggml_type_name(type), ne[0], ne[1],ne[2], ne[3]);
tensor_storage.reverse_ne();
size_t tensor_data_size = end - begin;

5
pmid.hpp
View File

@@ -552,7 +552,10 @@ struct PhotoMakerLoraModel : public GGMLModule {
ModelLoader model_loader;
bool load_failed = false;
PhotoMakerLoraModel(const std::string file_path = "")
PhotoMakerLoraModel(){
}
PhotoMakerLoraModel(const std::string file_path)
: file_path(file_path) {
name = "photomaker lora";
if (!model_loader.init_from_file(file_path, "pmid.")) {

12
stable-diffusion.cpp
View File

@@ -633,9 +633,17 @@ public:
prompts_embeds->type,
total_hidden_size,
cond_stage_model.text_model.max_position_embeddings);
pmid_model.alloc_compute_buffer(work_ctx, init_img, prompts_embeds, class_tokens_mask);
pmid_model.compute(n_threads, init_img, prompts_embeds, class_tokens_mask, res);
pmid_model.free_compute_buffer();
// printf("[");
// int j = 70;
// for(int i = 0; i < 100; i++){
// float val = *((float *)(res->data)+j*total_hidden_size+i);
// printf("%f, ", val);
// }
// printf("]\n");
return res;
}
@@ -825,8 +833,8 @@ public:
diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c, control_net.controls, control_strength, t_emb, c_vector);
}
else{
// diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c_id, control_net.controls, control_strength, t_emb, c_vec_id);
diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c_id, control_net.controls, control_strength, t_emb, c_vector);
diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c_id, control_net.controls, control_strength, t_emb, c_vec_id);
// diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c_id, control_net.controls, control_strength, t_emb, c_vector);
}
}else{
diffusion_model.compute(out_cond, n_threads, noised_input, NULL, c, control_net.controls, control_strength, t_emb, c_vector);