-
Notifications
You must be signed in to change notification settings - Fork 48
/
bark.cpp
362 lines (274 loc) · 13.4 KB
/
bark.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
#include "bark.h"
#include "ggml.h"
#include "util.h"
#include <cassert>
#include <cmath>
#include <cstdio>
#include <cstring>
#include <fstream>
#include <map>
#include <string>
bool gpt_model_load(std::ifstream & fin, gpt_model & model) {
// load hparams
{
auto & hparams = model.hparams;
read_safe(fin, hparams.n_layer);
read_safe(fin, hparams.n_head);
read_safe(fin, hparams.n_embd);
read_safe(fin, hparams.block_size);
read_safe(fin, hparams.n_in_vocab);
read_safe(fin, hparams.n_out_vocab);
read_safe(fin, hparams.n_lm_heads);
read_safe(fin, hparams.n_wtes);
printf("%s: n_in_vocab = %d\n", __func__, hparams.n_in_vocab);
printf("%s: n_out_vocab = %d\n", __func__, hparams.n_out_vocab);
printf("%s: block_size = %d\n", __func__, hparams.block_size);
printf("%s: n_embd = %d\n", __func__, hparams.n_embd);
printf("%s: n_head = %d\n", __func__, hparams.n_head);
printf("%s: n_layer = %d\n", __func__, hparams.n_layer);
printf("%s: n_lm_heads = %d\n", __func__, hparams.n_lm_heads);
printf("%s: n_wtes = %d\n", __func__, hparams.n_wtes);
}
// TODO: load vocab
{
}
// for the big tensors, we have the option to store the data in 16-bit floats or quantized
// in order to save memory and also to speed up the computation
// ggml_type wtype = ggml_ftype_to_ggml_type((ggml_ftype) (model.hparams.ftype));
// if (wtype == GGML_TYPE_COUNT) {
// fprintf(stderr, "%s: invalid model file '%s' (bad ftype value %d)\n",
// __func__, fname.c_str(), model.hparams.ftype);
// return false;
// }
ggml_type wtype = GGML_TYPE_F32;
auto & ctx = model.ctx;
size_t ctx_size = 0;
{
const auto & hparams = model.hparams;
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int block_size = hparams.block_size;
const int n_in_vocab = hparams.n_in_vocab;
const int n_out_vocab = hparams.n_out_vocab;
const int n_lm_heads = hparams.n_lm_heads;
const int n_wtes = hparams.n_wtes;
ctx_size += n_embd*ggml_type_sizef(GGML_TYPE_F32); // ln_f_g
ctx_size += n_embd*ggml_type_sizef(GGML_TYPE_F32); // ln_f_b
ctx_size += n_wtes*n_in_vocab*n_embd*ggml_type_sizef(wtype); // wte
ctx_size += block_size*n_embd*ggml_type_sizef(GGML_TYPE_F32); // wpe
ctx_size += n_lm_heads*n_out_vocab*n_embd*ggml_type_sizef(wtype); // lm_head
ctx_size += n_layer*(n_embd*ggml_type_sizef(GGML_TYPE_F32)); // ln_1_g
ctx_size += n_layer*(n_embd*ggml_type_sizef(GGML_TYPE_F32)); // ln_1_b
ctx_size += n_layer*(n_embd*ggml_type_sizef(GGML_TYPE_F32)); // ln_2_g
ctx_size += n_layer*(n_embd*ggml_type_sizef(GGML_TYPE_F32)); // ln_2_b
ctx_size += n_layer*(3*n_embd*n_embd*ggml_type_sizef(wtype)); // c_attn_attn_w
ctx_size += n_layer*( 3*n_embd*ggml_type_sizef(GGML_TYPE_F32)); // c_attn_attn_b
ctx_size += n_layer*(n_embd*n_embd*ggml_type_sizef(wtype)); // c_attn_proj_w
ctx_size += n_layer*( n_embd*ggml_type_sizef(GGML_TYPE_F32)); // c_attn_proj_b
ctx_size += n_layer*(4*n_embd*n_embd*ggml_type_sizef(wtype)); // c_mlp_fc_w
ctx_size += n_layer*( 4*n_embd*ggml_type_sizef(GGML_TYPE_F32)); // c_mlp_fc_b
ctx_size += n_layer*(4*n_embd*n_embd*ggml_type_sizef(wtype)); // c_mlp_proj_w
ctx_size += n_layer*( n_embd*ggml_type_sizef(GGML_TYPE_F32)); // c_mlp_proj_b
ctx_size += block_size*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_k
ctx_size += block_size*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_v
ctx_size += (6 + 12*n_layer)*512; // object overhead
printf("%s: ggml tensor size = %d bytes\n", __func__, (int) sizeof(ggml_tensor));
printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0));
}
// create the ggml context
{
struct ggml_init_params params = {
/*.mem_size =*/ ctx_size,
/*.mem_buffer =*/ NULL,
/*.no_alloc =*/ false,
};
model.ctx = ggml_init(params);
if (!model.ctx) {
fprintf(stderr, "%s: ggml_init() failed\n", __func__);
return false;
}
}
// prepare memory for the weights
{
const auto & hparams = model.hparams;
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int block_size = hparams.block_size;
const int n_in_vocab = hparams.n_in_vocab;
const int n_out_vocab = hparams.n_out_vocab;
const int n_lm_heads = hparams.n_lm_heads;
const int n_wtes = hparams.n_wtes;
model.layers.resize(n_layer);
model.lm_heads.resize(n_lm_heads);
model.wtes.resize(n_wtes);
model.ln_f_g = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
model.ln_f_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
model.wpe = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, n_embd, block_size);
for (int i = 0; i < n_wtes; i++) {
model.wtes[i] = ggml_new_tensor_2d(ctx, wtype, n_embd, n_in_vocab);
model.tensors["model/wte/" + std::to_string(i)] = model.wtes[i];
}
for (int i = 0; i < n_lm_heads; i++) {
model.lm_heads[i] = ggml_new_tensor_2d(ctx, wtype, n_embd, n_out_vocab);
model.tensors["model/lm_head/" + std::to_string(i)] = model.lm_heads[i];
}
model.tensors["model/ln_f/g"] = model.ln_f_g;
model.tensors["model/ln_f/b"] = model.ln_f_b;
model.tensors["model/wpe"] = model.wpe;
for (int i = 0; i < n_layer; ++i) {
auto & layer = model.layers[i];
layer.ln_1_g = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.ln_1_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.ln_2_g = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.ln_2_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.c_attn_attn_w = ggml_new_tensor_2d(ctx, wtype, n_embd, 3*n_embd);
layer.c_attn_attn_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 3*n_embd);
layer.c_attn_proj_w = ggml_new_tensor_2d(ctx, wtype, n_embd, n_embd);
layer.c_attn_proj_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.c_mlp_fc_w = ggml_new_tensor_2d(ctx, wtype, n_embd, 4*n_embd);
layer.c_mlp_fc_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 4*n_embd);
layer.c_mlp_proj_w = ggml_new_tensor_2d(ctx, wtype, 4*n_embd, n_embd);
layer.c_mlp_proj_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
// map by name
model.tensors["model/h" + std::to_string(i) + "/ln_1/g"] = layer.ln_1_g;
model.tensors["model/h" + std::to_string(i) + "/ln_1/b"] = layer.ln_1_b;
model.tensors["model/h" + std::to_string(i) + "/ln_2/g"] = layer.ln_2_g;
model.tensors["model/h" + std::to_string(i) + "/ln_2/b"] = layer.ln_2_b;
model.tensors["model/h" + std::to_string(i) + "/attn/c_attn/w"] = layer.c_attn_attn_w;
model.tensors["model/h" + std::to_string(i) + "/attn/c_attn/b"] = layer.c_attn_attn_b;
model.tensors["model/h" + std::to_string(i) + "/attn/c_proj/w"] = layer.c_attn_proj_w;
model.tensors["model/h" + std::to_string(i) + "/attn/c_proj/b"] = layer.c_attn_proj_b;
model.tensors["model/h" + std::to_string(i) + "/mlp/c_fc/w"] = layer.c_mlp_fc_w;
model.tensors["model/h" + std::to_string(i) + "/mlp/c_fc/b"] = layer.c_mlp_fc_b;
model.tensors["model/h" + std::to_string(i) + "/mlp/c_proj/w"] = layer.c_mlp_proj_w;
model.tensors["model/h" + std::to_string(i) + "/mlp/c_proj/b"] = layer.c_mlp_proj_b;
}
}
// key + value memory
{
const auto & hparams = model.hparams;
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int block_size = hparams.block_size;
const int n_mem = n_layer*block_size;
const int n_elements = n_embd*n_mem;
model.memory_k = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
model.memory_v = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v);
printf("%s: memory size = %8.2f MB, n_mem = %d\n", __func__, memory_size/1024.0/1024.0, n_mem);
}
// load weights
{
size_t total_size = 0;
int32_t n_tensors;
read_safe(fin, n_tensors);
for(int i = 0; i < n_tensors; i++) {
int32_t n_dims;
int32_t length;
int32_t ttype;
read_safe(fin, n_dims);
read_safe(fin, length);
read_safe(fin, ttype);
int32_t nelements = 1;
int32_t ne[2] = { 1, 1 };
for (int i = 0; i < n_dims; ++i) {
read_safe(fin, ne[i]);
nelements *= ne[i];
}
std::string name(length, 0);
fin.read(&name[0], length);
if (model.tensors.find(name.data()) == model.tensors.end()) {
fprintf(stderr, "%s: unknown tensor '%s' in model file\n", __func__, name.data());
return false;
}
auto tensor = model.tensors[name.data()];
if (ggml_nelements(tensor) != nelements) {
fprintf(stderr, "%s: tensor '%s' has wrong size in model file\n", __func__, name.data());
return false;
}
if (tensor->ne[0] != ne[0] || tensor->ne[1] != ne[1]) {
fprintf(stderr, "%s: tensor '%s' has wrong shape in model file: got [%d, %d], expected [%d, %d]\n",
__func__, name.data(), (int) tensor->ne[0], (int) tensor->ne[1], ne[0], ne[1]);
return false;
}
const size_t bpe = ggml_type_size(ggml_type(ttype));
if ((nelements*bpe)/ggml_blck_size(tensor->type) != ggml_nbytes(tensor)) {
fprintf(stderr, "%s: tensor '%s' has wrong size in model file: got %zu, expected %zu\n",
__func__, name.data(), ggml_nbytes(tensor), nelements*bpe);
return false;
}
fin.read(reinterpret_cast<char *>(tensor->data), ggml_nbytes(tensor));
printf("%48s - [%5d, %5d], type = %6s, %6.2f MB\n", name.data(), ne[0], ne[1], "float", ggml_nbytes(tensor)/1024.0/1024.0);
total_size += ggml_nbytes(tensor);
}
printf("%s: model size = %8.2f MB\n", __func__, total_size/1024.0/1024.0);
}
fin.close();
return true;
}
bool bark_model_load(const std::string & fname, bark_model & model) {
printf("%s: loading model from '%s'\n", __func__, fname.c_str());
auto fin = std::ifstream(fname, std::ios::binary);
if (!fin) {
fprintf(stderr, "%s: failed to open '%s'\n", __func__, fname.c_str());
return false;
}
// verify magic
{
uint32_t magic;
fin.read((char *) &magic, sizeof(magic));
if (magic != GGML_FILE_MAGIC) {
fprintf(stderr, "%s: invalid model file '%s' (bad magic)\n", __func__, fname.c_str());
return false;
}
}
printf("\n%s: reading bark text model\n", __func__);
if(!gpt_model_load(fin, model.text_model)) {
fprintf(stderr, "%s: invalid model file '%s' (bad text)\n", __func__, fname.c_str());
return false;
}
printf("%s: reading bark coarse model\n\n", __func__);
if(!gpt_model_load(fin, model.coarse_model)) {
fprintf(stderr, "%s: invalid model file '%s' (bad coarse)\n", __func__, fname.c_str());
return false;
}
printf("\n%s: reading bark fine model\n", __func__);
if(!gpt_model_load(fin, model.fine_model)) {
fprintf(stderr, "%s: invalid model file '%s' (bad fine)\n", __func__, fname.c_str());
return false;
}
printf("\n%s: reading bark codec model\n", __func__);
if(!encodec_model_load(fin, model.codec_model)) {
fprintf(stderr, "%s: invalid model file '%s' (bad codec)\n", __func__, fname.c_str());
return false;
}
return true;
}
int main(int argc, char **argv) {
const int64_t t_main_start_us = ggml_time_us();
int64_t t_load_us = 0;
bark_model model;
std::string fname = "./ggml_weights/ggml-model.bin";
// load the model
{
const int64_t t_start_us = ggml_time_us();
if(!bark_model_load(fname, model)) {
fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, fname.c_str());
return 1;
}
t_load_us = ggml_time_us() - t_start_us;
}
// report timing
{
const int64_t t_main_end_us = ggml_time_us();
printf("\n\n");
printf("%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f);
printf("%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f);
}
// TODO: write wrapper
ggml_free(model.coarse_model.ctx);
ggml_free(model.fine_model.ctx);
ggml_free(model.text_model.ctx);
ggml_free(model.codec_model.ctx);
return 0;
}