Georgi Gerganov
1 year ago
parent
d1da35de06
commit
3b1aacbe6d
@ -0,0 +1 @@
|
||||
eleven-labs.py
|
||||
@ -0,0 +1,7 @@
|
||||
if (WHISPER_SUPPORT_SDL2)
|
||||
# talk
|
||||
set(TARGET talk)
|
||||
add_executable(${TARGET} talk.cpp gpt-2.cpp)
|
||||
target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
|
||||
target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
|
||||
endif ()
|
||||
@ -0,0 +1,33 @@
|
||||
# talk
|
||||
|
||||
Talk with an Artificial Intelligence in your terminal
|
||||
|
||||
[Demo Talk](https://user-images.githubusercontent.com/1991296/206805012-48e71cc2-588d-4745-8798-c1c70ea3b40d.mp4)
|
||||
|
||||
Web version: [examples/talk.wasm](/examples/talk.wasm)
|
||||
|
||||
## Building
|
||||
|
||||
The `talk` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
|
||||
|
||||
```bash
|
||||
# Install SDL2 on Linux
|
||||
sudo apt-get install libsdl2-dev
|
||||
|
||||
# Install SDL2 on Mac OS
|
||||
brew install sdl2
|
||||
|
||||
# Build the "talk" executable
|
||||
make talk
|
||||
|
||||
# Run it
|
||||
./talk -p Santa
|
||||
```
|
||||
|
||||
To run this, you will need a ggml GPT-2 model: [instructions](https://github.com/ggerganov/ggml/tree/master/examples/gpt-2#downloading-and-converting-the-original-models)
|
||||
|
||||
Alternatively, you can simply download the smallest ggml GPT-2 117M model (240 MB) like this:
|
||||
|
||||
```
|
||||
wget --quiet --show-progress -O models/ggml-gpt-2-117M.bin https://ggml.ggerganov.com/ggml-model-gpt-2-117M.bin
|
||||
```
|
||||
@ -0,0 +1,925 @@
|
||||
#include "ggml.h"
|
||||
#include "gpt-2.h"
|
||||
|
||||
#include <cmath>
|
||||
#include <cstdio>
|
||||
#include <cstring>
|
||||
#include <fstream>
|
||||
#include <map>
|
||||
#include <string>
|
||||
#include <thread>
|
||||
#include <vector>
|
||||
#include <regex>
|
||||
#include <random>
|
||||
|
||||
/////////////////////// GPT-2 BEGIN /////////////////////////
|
||||
|
||||
//
|
||||
// Vocab utils
|
||||
//
|
||||
|
||||
std::vector<gpt_vocab::id> gpt_tokenize(const gpt_vocab & vocab, const std::string & text) {
|
||||
std::vector<std::string> words;
|
||||
|
||||
// first split the text into words
|
||||
{
|
||||
std::string str = text;
|
||||
std::string pat = R"('s|'t|'re|'ve|'m|'ll|'d| ?[[:alpha:]]+| ?[[:digit:]]+| ?[^\s[:alpha:][:digit:]]+|\s+(?!\S)|\s+)";
|
||||
|
||||
std::regex re(pat);
|
||||
std::smatch m;
|
||||
|
||||
while (std::regex_search(str, m, re)) {
|
||||
for (auto x : m) {
|
||||
words.push_back(x);
|
||||
}
|
||||
str = m.suffix();
|
||||
}
|
||||
}
|
||||
|
||||
// find the longest tokens that form the words:
|
||||
std::vector<gpt_vocab::id> tokens;
|
||||
for (const auto & word : words) {
|
||||
if (word.size() == 0) continue;
|
||||
|
||||
int i = 0;
|
||||
int n = word.size();
|
||||
while (i < n) {
|
||||
int j = n;
|
||||
while (j > i) {
|
||||
auto it = vocab.token_to_id.find(word.substr(i, j-i));
|
||||
if (it != vocab.token_to_id.end()) {
|
||||
tokens.push_back(it->second);
|
||||
i = j;
|
||||
break;
|
||||
}
|
||||
--j;
|
||||
}
|
||||
if (i == n) {
|
||||
break;
|
||||
}
|
||||
if (j == i) {
|
||||
auto sub = word.substr(i, 1);
|
||||
if (vocab.token_to_id.find(sub) != vocab.token_to_id.end()) {
|
||||
tokens.push_back(vocab.token_to_id.at(sub));
|
||||
} else {
|
||||
fprintf(stderr, "%s: unknown token '%s'\n", __func__, sub.data());
|
||||
}
|
||||
++i;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return tokens;
|
||||
}
|
||||
|
||||
gpt_vocab::id gpt_sample_top_k_top_p(
|
||||
const gpt_vocab & vocab,
|
||||
const float * logits,
|
||||
int top_k,
|
||||
double top_p,
|
||||
double temp,
|
||||
std::mt19937 & rng) {
|
||||
int n_logits = vocab.id_to_token.size();
|
||||
|
||||
std::vector<std::pair<double, gpt_vocab::id>> logits_id;
|
||||
logits_id.reserve(n_logits);
|
||||
|
||||
for (int i = 0; i < n_logits; i++) {
|
||||
logits_id.push_back(std::make_pair(logits[i], i));
|
||||
}
|
||||
|
||||
// find the top K tokens
|
||||
std::partial_sort(
|
||||
logits_id.begin(),
|
||||
logits_id.begin() + top_k, logits_id.end(),
|
||||
[](const std::pair<double, gpt_vocab::id> & a, const std::pair<double, gpt_vocab::id> & b) {
|
||||
return a.first > b.first;
|
||||
});
|
||||
|
||||
logits_id.resize(top_k);
|
||||
|
||||
// normalize
|
||||
{
|
||||
double sum = 0.0f;
|
||||
for (int i = 0; i < (int)logits_id.size(); i++) {
|
||||
sum += logits_id[i].first;
|
||||
}
|
||||
|
||||
sum = 1.0/sum;
|
||||
for (int i = 0; i < (int)logits_id.size(); i++) {
|
||||
logits_id[i].first *= sum;
|
||||
}
|
||||
}
|
||||
|
||||
if (top_p < 1.0f) {
|
||||
{
|
||||
double cumsum = 0.0f;
|
||||
for (int i = 0; i < top_k; i++) {
|
||||
cumsum += logits_id[i].first;
|
||||
if (cumsum >= top_p) {
|
||||
logits_id.resize(i+1);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// normalize again
|
||||
{
|
||||
double sum = 0.0f;
|
||||
for (int i = 0; i < (int)logits_id.size(); i++) {
|
||||
sum += logits_id[i].first;
|
||||
}
|
||||
|
||||
sum = 1.0/sum;
|
||||
for (int i = 0; i < (int)logits_id.size(); i++) {
|
||||
logits_id[i].first *= sum;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
//printf("\n");
|
||||
//for (int i = 0; i < (int)logits_id.size(); i++) {
|
||||
// printf("%d: '%s' %f\n", i, vocab.id_to_token.at(logits_id[i].second).c_str(), logits_id[i].first);
|
||||
//}
|
||||
//exit(0);
|
||||
|
||||
// sample from the obtained distribution
|
||||
std::vector<double> probs;
|
||||
probs.reserve(logits_id.size());
|
||||
|
||||
for (int i = 0; i < (int) logits_id.size(); i++) {
|
||||
probs.push_back(logits_id[i].first);
|
||||
}
|
||||
|
||||
std::discrete_distribution<> dist(probs.begin(), probs.end());
|
||||
int idx = dist(rng);
|
||||
|
||||
return logits_id[idx].second;
|
||||
}
|
||||
|
||||
// default hparams (GPT-2 117M)
|
||||
struct gpt2_hparams {
|
||||
int32_t n_vocab = 50257;
|
||||
int32_t n_ctx = 1024;
|
||||
int32_t n_embd = 768;
|
||||
int32_t n_head = 12;
|
||||
int32_t n_layer = 12;
|
||||
int32_t f16 = 1;
|
||||
};
|
||||
|
||||
struct gpt2_layer {
|
||||
// normalization
|
||||
struct ggml_tensor * ln_1_g;
|
||||
struct ggml_tensor * ln_1_b;
|
||||
|
||||
struct ggml_tensor * ln_2_g;
|
||||
struct ggml_tensor * ln_2_b;
|
||||
|
||||
// attention
|
||||
struct ggml_tensor * c_attn_attn_w;
|
||||
struct ggml_tensor * c_attn_attn_b;
|
||||
|
||||
struct ggml_tensor * c_attn_proj_w;
|
||||
struct ggml_tensor * c_attn_proj_b;
|
||||
|
||||
// mlp
|
||||
struct ggml_tensor * c_mlp_fc_w;
|
||||
struct ggml_tensor * c_mlp_fc_b;
|
||||
|
||||
struct ggml_tensor * c_mlp_proj_w_trans; // transposed for efficiency
|
||||
struct ggml_tensor * c_mlp_proj_b;
|
||||
};
|
||||
|
||||
struct gpt2_model {
|
||||
gpt2_hparams hparams;
|
||||
|
||||
// normalization
|
||||
struct ggml_tensor * ln_f_g;
|
||||
struct ggml_tensor * ln_f_b;
|
||||
|
||||
struct ggml_tensor * wte; // position embedding
|
||||
struct ggml_tensor * wpe; // token embedding
|
||||
|
||||
std::vector<gpt2_layer> layers;
|
||||
|
||||
// key + value memory
|
||||
struct ggml_tensor * memory_k;
|
||||
struct ggml_tensor * memory_v;
|
||||
|
||||
//
|
||||
struct ggml_context * ctx;
|
||||
std::map<std::string, struct ggml_tensor *> tensors;
|
||||
};
|
||||
|
||||
// load the model's weights from a file
|
||||
bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab & vocab) {
|
||||
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 != 0x67676d6c) {
|
||||
fprintf(stderr, "%s: invalid model file '%s' (bad magic)\n", __func__, fname.c_str());
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
// load hparams
|
||||
{
|
||||
auto & hparams = model.hparams;
|
||||
|
||||
fin.read((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
|
||||
fin.read((char *) &hparams.n_ctx, sizeof(hparams.n_ctx));
|
||||
fin.read((char *) &hparams.n_embd, sizeof(hparams.n_embd));
|
||||
fin.read((char *) &hparams.n_head, sizeof(hparams.n_head));
|
||||
fin.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
|
||||
fin.read((char *) &hparams.f16, sizeof(hparams.f16));
|
||||
|
||||
printf("%s: n_vocab = %d\n", __func__, hparams.n_vocab);
|
||||
printf("%s: n_ctx = %d\n", __func__, hparams.n_ctx);
|
||||
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: f16 = %d\n", __func__, hparams.f16);
|
||||
}
|
||||
|
||||
// load vocab
|
||||
{
|
||||
int32_t n_vocab = 0;
|
||||
fin.read((char *) &n_vocab, sizeof(n_vocab));
|
||||
|
||||
if (n_vocab != model.hparams.n_vocab) {
|
||||
fprintf(stderr, "%s: invalid model file '%s' (bad vocab size %d != %d)\n",
|
||||
__func__, fname.c_str(), n_vocab, model.hparams.n_vocab);
|
||||
return false;
|
||||
}
|
||||
|
||||
std::string word;
|
||||
for (int i = 0; i < n_vocab; i++) {
|
||||
uint32_t len;
|
||||
fin.read((char *) &len, sizeof(len));
|
||||
|
||||
word.resize(len);
|
||||
fin.read((char *) word.data(), len);
|
||||
|
||||
vocab.token_to_id[word] = i;
|
||||
vocab.id_to_token[i] = word;
|
||||
}
|
||||
}
|
||||
|
||||
// for the big tensors, we have the option to store the data in 16-bit floats
|
||||
// in order to save memory and also to speed up the computation
|
||||
const ggml_type wtype = model.hparams.f16 ? GGML_TYPE_F16 : 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 n_ctx = hparams.n_ctx;
|
||||
const int n_vocab = hparams.n_vocab;
|
||||
|
||||
ctx_size += n_embd*ggml_type_size(GGML_TYPE_F32); // ln_f_g
|
||||
ctx_size += n_embd*ggml_type_size(GGML_TYPE_F32); // ln_f_b
|
||||
|
||||
ctx_size += n_vocab*n_embd*ggml_type_size(wtype); // wte
|
||||
ctx_size += n_ctx*n_embd*ggml_type_size(GGML_TYPE_F32); // wpe
|
||||
|
||||
ctx_size += n_layer*(n_embd*ggml_type_size(GGML_TYPE_F32)); // ln_1_g
|
||||
ctx_size += n_layer*(n_embd*ggml_type_size(GGML_TYPE_F32)); // ln_1_b
|
||||
|
||||
ctx_size += n_layer*(n_embd*ggml_type_size(GGML_TYPE_F32)); // ln_2_g
|
||||
ctx_size += n_layer*(n_embd*ggml_type_size(GGML_TYPE_F32)); // ln_2_b
|
||||
|
||||
ctx_size += n_layer*(3*n_embd*n_embd*ggml_type_size(wtype)); // c_attn_attn_w
|
||||
ctx_size += n_layer*( 3*n_embd*ggml_type_size(GGML_TYPE_F32)); // c_attn_attn_b
|
||||
|
||||
ctx_size += n_layer*(n_embd*n_embd*ggml_type_size(wtype)); // c_attn_proj_w
|
||||
ctx_size += n_layer*( n_embd*ggml_type_size(GGML_TYPE_F32)); // c_attn_proj_b
|
||||
|
||||
ctx_size += n_layer*(4*n_embd*n_embd*ggml_type_size(wtype)); // c_mlp_fc_w
|
||||
ctx_size += n_layer*( 4*n_embd*ggml_type_size(GGML_TYPE_F32)); // c_mlp_fc_b
|
||||
|
||||
ctx_size += n_layer*(4*n_embd*n_embd*ggml_type_size(wtype)); // c_mlp_proj_w
|
||||
ctx_size += n_layer*( n_embd*ggml_type_size(GGML_TYPE_F32)); // c_mlp_proj_b
|
||||
|
||||
ctx_size += n_ctx*n_layer*n_embd*ggml_type_size(GGML_TYPE_F32); // memory_k
|
||||
ctx_size += n_ctx*n_layer*n_embd*ggml_type_size(GGML_TYPE_F32); // memory_v
|
||||
|
||||
ctx_size += (6 + 12*n_layer)*256; // object overhead
|
||||
|
||||
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,
|
||||
};
|
||||
|
||||
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 n_ctx = hparams.n_ctx;
|
||||
const int n_vocab = hparams.n_vocab;
|
||||
|
||||
model.layers.resize(n_layer);
|
||||
|
||||
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.wte = ggml_new_tensor_2d(ctx, wtype, n_embd, n_vocab);
|
||||
model.wpe = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, n_embd, n_ctx);
|
||||
|
||||
// map by name
|
||||
model.tensors["model/ln_f/g"] = model.ln_f_g;
|
||||
model.tensors["model/ln_f/b"] = model.ln_f_b;
|
||||
|
||||
model.tensors["model/wte"] = model.wte;
|
||||
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, 3*n_embd, 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, 4*n_embd, n_embd);
|
||||
layer.c_mlp_fc_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 4*n_embd);
|
||||
|
||||
layer.c_mlp_proj_w_trans = 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_trans;
|
||||
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 n_ctx = hparams.n_ctx;
|
||||
|
||||
const int n_mem = n_layer*n_ctx;
|
||||
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;
|
||||
|
||||
while (true) {
|
||||
int32_t n_dims;
|
||||
int32_t length;
|
||||
int32_t ftype;
|
||||
|
||||
fin.read(reinterpret_cast<char *>(&n_dims), sizeof(n_dims));
|
||||
fin.read(reinterpret_cast<char *>(&length), sizeof(length));
|
||||
fin.read(reinterpret_cast<char *>(&ftype), sizeof(ftype));
|
||||
|
||||
if (fin.eof()) {
|
||||
break;
|
||||
}
|
||||
|
||||
int32_t nelements = 1;
|
||||
int32_t ne[2] = { 1, 1 };
|
||||
for (int i = 0; i < n_dims; ++i) {
|
||||
fin.read(reinterpret_cast<char *>(&ne[i]), sizeof(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(), tensor->ne[0], tensor->ne[1], ne[0], ne[1]);
|
||||
return false;
|
||||
}
|
||||
|
||||
const size_t bpe = (ftype == 0) ? sizeof(float) : sizeof(ggml_fp16_t);
|
||||
|
||||
if (nelements*bpe != 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("%24s - [%5d, %5d], type = %6s, %6.2f MB\n", name.data(), ne[0], ne[1], ftype == 0 ? "float" : "f16", 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;
|
||||
}
|
||||
|
||||
// evaluate the transformer
|
||||
//
|
||||
// - model: the model
|
||||
// - n_threads: number of threads to use
|
||||
// - n_past: the context size so far
|
||||
// - embd_inp: the embeddings of the tokens in the context
|
||||
// - embd_w: the predicted probabilities of the next token
|
||||
//
|
||||
bool gpt2_eval(
|
||||
const gpt2_model & model,
|
||||
const int n_threads,
|
||||
const int n_past,
|
||||
const std::vector<gpt_vocab::id> & embd_inp,
|
||||
std::vector<float> & embd_w,
|
||||
size_t & mem_per_token) {
|
||||
const int N = embd_inp.size();
|
||||
|
||||
const auto & hparams = model.hparams;
|
||||
|
||||
const int n_embd = hparams.n_embd;
|
||||
const int n_layer = hparams.n_layer;
|
||||
const int n_ctx = hparams.n_ctx;
|
||||
const int n_head = hparams.n_head;
|
||||
const int n_vocab = hparams.n_vocab;
|
||||
|
||||
static size_t buf_size = 5640ull*1024*1024;
|
||||
static void * buf = malloc(buf_size);
|
||||
|
||||
if (mem_per_token > 0 && mem_per_token*N > buf_size) {
|
||||
const size_t buf_size_new = 1.1*(mem_per_token*N); // add 10% to account for ggml object overhead
|
||||
printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
|
||||
|
||||
// reallocate
|
||||
buf_size = buf_size_new;
|
||||
buf = realloc(buf, buf_size);
|
||||
if (buf == nullptr) {
|
||||
fprintf(stderr, "%s: failed to allocate %zu bytes\n", __func__, buf_size);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
struct ggml_init_params params = {
|
||||
.mem_size = buf_size,
|
||||
.mem_buffer = buf,
|
||||
};
|
||||
|
||||
struct ggml_context * ctx0 = ggml_init(params);
|
||||
struct ggml_cgraph gf = { .n_threads = n_threads };
|
||||
|
||||
struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
|
||||
memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));
|
||||
|
||||
struct ggml_tensor * position = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
|
||||
for (int i = 0; i < N; ++i) {
|
||||
((int32_t *) position->data)[i] = n_past + i;
|
||||
}
|
||||
|
||||
// wte + wpe
|
||||
struct ggml_tensor * inpL =
|
||||
ggml_add(ctx0,
|
||||
ggml_get_rows(ctx0, model.wte, embd),
|
||||
ggml_get_rows(ctx0, model.wpe, position));
|
||||
|
||||
for (int il = 0; il < n_layer; ++il) {
|
||||
struct ggml_tensor * cur;
|
||||
|
||||
// norm
|
||||
{
|
||||
// [ 768, N]
|
||||
cur = ggml_norm(ctx0, inpL);
|
||||
|
||||
// cur = ln_1_g*cur + ln_1_b
|
||||
// [ 768, N]
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_mul(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].ln_1_g, cur),
|
||||
cur),
|
||||
ggml_repeat(ctx0, model.layers[il].ln_1_b, cur));
|
||||
}
|
||||
|
||||
// attn
|
||||
// [2304, 768] - model.layers[il].c_attn_attn_w
|
||||
// [2304, 1] - model.layers[il].c_attn_attn_b
|
||||
// [ 768, N] - cur (in)
|
||||
// [2304, N] - cur (out)
|
||||
//
|
||||
// cur = attn_w*cur + attn_b
|
||||
// [2304, N]
|
||||
{
|
||||
cur = ggml_mul_mat(ctx0,
|
||||
ggml_transpose(ctx0, model.layers[il].c_attn_attn_w),
|
||||
cur);
|
||||
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].c_attn_attn_b, cur),
|
||||
cur);
|
||||
}
|
||||
|
||||
// self-attention
|
||||
{
|
||||
struct ggml_tensor * Qcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 0*sizeof(float)*n_embd);
|
||||
struct ggml_tensor * Kcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 1*sizeof(float)*n_embd);
|
||||
struct ggml_tensor * Vcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 2*sizeof(float)*n_embd);
|
||||
|
||||
// store key and value to memory
|
||||
if (N >= 1) {
|
||||
struct ggml_tensor * k = ggml_view_1d(ctx0, model.memory_k, N*n_embd, (ggml_element_size(model.memory_k)*n_embd)*(il*n_ctx + n_past));
|
||||
struct ggml_tensor * v = ggml_view_1d(ctx0, model.memory_v, N*n_embd, (ggml_element_size(model.memory_v)*n_embd)*(il*n_ctx + n_past));
|
||||
|
||||
ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Kcur, k));
|
||||
ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Vcur, v));
|
||||
}
|
||||
|
||||
// Q = Qcur.contiguous().view(n_embd/n_head, n_head, N).permute(0, 2, 1, 3)
|
||||
// [64, N, 12]
|
||||
struct ggml_tensor * Q =
|
||||
ggml_permute(ctx0,
|
||||
ggml_cpy(ctx0,
|
||||
Qcur,
|
||||
ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_embd/n_head, n_head, N)),
|
||||
0, 2, 1, 3);
|
||||
|
||||
// K = Kmem.view(n_embd/n_head, n_head, n_past + N).permute(0, 2, 1, 3)
|
||||
// [64, n_past + N, 12]
|
||||
struct ggml_tensor * K =
|
||||
ggml_permute(ctx0,
|
||||
ggml_reshape_3d(ctx0,
|
||||
ggml_view_1d(ctx0, model.memory_k, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_k)*n_embd),
|
||||
n_embd/n_head, n_head, n_past + N),
|
||||
0, 2, 1, 3);
|
||||
|
||||
// GG: flash attention
|
||||
//struct ggml_tensor * V =
|
||||
// ggml_cpy(ctx0,
|
||||
// ggml_permute(ctx0,
|
||||
// ggml_reshape_3d(ctx0,
|
||||
// ggml_view_1d(ctx0, model.memory_v, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_v)*n_embd),
|
||||
// n_embd/n_head, n_head, n_past + N),
|
||||
// 1, 2, 0, 3),
|
||||
// ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_past + N, n_embd/n_head, n_head));
|
||||
|
||||
//struct ggml_tensor * KQV = ggml_flash_attn(ctx0, Q, K, V, true);
|
||||
|
||||
// K * Q
|
||||
// [n_past + N, N, 12]
|
||||
struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q);
|
||||
|
||||
// KQ_scaled = KQ / sqrt(n_embd/n_head)
|
||||
// [n_past + N, N, 12]
|
||||
struct ggml_tensor * KQ_scaled =
|
||||
ggml_scale(ctx0,
|
||||
KQ,
|
||||
ggml_new_f32(ctx0, 1.0f/sqrt(float(n_embd)/n_head))
|
||||
);
|
||||
|
||||
// KQ_masked = mask_past(KQ_scaled)
|
||||
// [n_past + N, N, 12]
|
||||
struct ggml_tensor * KQ_masked = ggml_diag_mask_inf(ctx0, KQ_scaled, n_past);
|
||||
|
||||
// KQ = soft_max(KQ_masked)
|
||||
// [n_past + N, N, 12]
|
||||
struct ggml_tensor * KQ_soft_max = ggml_soft_max(ctx0, KQ_masked);
|
||||
|
||||
// V_trans = Vmem.view(n_embd/n_head, n_head, n_past + N).permute(1, 2, 0, 3).contiguous()
|
||||
// [n_past + N, 64, 12]
|
||||
struct ggml_tensor * V_trans =
|
||||
ggml_permute(ctx0,
|
||||
ggml_reshape_3d(ctx0,
|
||||
ggml_view_1d(ctx0, model.memory_v, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_v)*n_embd),
|
||||
n_embd/n_head, n_head, n_past + N),
|
||||
1, 2, 0, 3);
|
||||
|
||||
// KQV = transpose(V) * KQ_soft_max
|
||||
// [64, N, 12]
|
||||
struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V_trans, KQ_soft_max);
|
||||
|
||||
// KQV_merged = KQV.permute(0, 2, 1, 3)
|
||||
// [64, 12, N]
|
||||
struct ggml_tensor * KQV_merged = ggml_permute(ctx0, KQV, 0, 2, 1, 3);
|
||||
|
||||
// cur = KQV_merged.contiguous().view(n_embd, N)
|
||||
// [768, N]
|
||||
cur = ggml_cpy(ctx0,
|
||||
KQV_merged,
|
||||
ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N));
|
||||
}
|
||||
|
||||
// projection
|
||||
// [ 768, 768] - model.layers[il].c_attn_proj_w
|
||||
// [ 768, 1] - model.layers[il].c_attn_proj_b
|
||||
// [ 768, N] - cur (in)
|
||||
// [ 768, N] - cur (out)
|
||||
//
|
||||
// cur = proj_w*cur + proj_b
|
||||
// [768, N]
|
||||
{
|
||||
cur = ggml_mul_mat(ctx0,
|
||||
ggml_transpose(ctx0, model.layers[il].c_attn_proj_w),
|
||||
cur);
|
||||
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].c_attn_proj_b, cur),
|
||||
cur);
|
||||
}
|
||||
|
||||
// add the input
|
||||
cur = ggml_add(ctx0, cur, inpL);
|
||||
|
||||
struct ggml_tensor * inpFF = cur;
|
||||
|
||||
// feed-forward network
|
||||
{
|
||||
// norm
|
||||
{
|
||||
cur = ggml_norm(ctx0, inpFF);
|
||||
|
||||
// cur = ln_2_g*cur + ln_2_b
|
||||
// [ 768, N]
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_mul(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].ln_2_g, cur),
|
||||
cur),
|
||||
ggml_repeat(ctx0, model.layers[il].ln_2_b, cur));
|
||||
}
|
||||
|
||||
// fully connected
|
||||
// [3072, 768] - model.layers[il].c_mlp_fc_w
|
||||
// [3072, 1] - model.layers[il].c_mlp_fc_b
|
||||
// [ 768, N] - cur (in)
|
||||
// [3072, N] - cur (out)
|
||||
//
|
||||
// cur = fc_w*cur + fc_b
|
||||
// [3072, N]
|
||||
cur = ggml_mul_mat(ctx0,
|
||||
ggml_transpose(ctx0, model.layers[il].c_mlp_fc_w),
|
||||
cur);
|
||||
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].c_mlp_fc_b, cur),
|
||||
cur);
|
||||
|
||||
// GELU activation
|
||||
// [3072, N]
|
||||
cur = ggml_gelu(ctx0, cur);
|
||||
|
||||
// projection
|
||||
// [ 768, 3072] - model.layers[il].c_mlp_proj_w
|
||||
// [ 768, 1] - model.layers[il].c_mlp_proj_b
|
||||
// [3072, N] - cur (in)
|
||||
// [ 768, N] - cur (out)
|
||||
//
|
||||
// cur = proj_w*cur + proj_b
|
||||
// [768, N]
|
||||
cur = ggml_mul_mat(ctx0,
|
||||
model.layers[il].c_mlp_proj_w_trans,
|
||||
cur);
|
||||
|
||||
cur = ggml_add(ctx0,
|
||||
ggml_repeat(ctx0, model.layers[il].c_mlp_proj_b, cur),
|
||||
cur);
|
||||
}
|
||||
|
||||
// input for next layer
|
||||
inpL = ggml_add(ctx0, cur, inpFF);
|
||||
}
|
||||
|
||||
// norm
|
||||
{
|
||||
// [ 768, N]
|
||||
inpL = ggml_norm(ctx0, inpL);
|
||||
|
||||
// inpL = ln_f_g*inpL + ln_f_b
|
||||
// [ 768, N]
|
||||
inpL = ggml_add(ctx0,
|
||||
ggml_mul(ctx0,
|
||||
ggml_repeat(ctx0, model.ln_f_g, inpL),
|
||||
inpL),
|
||||
ggml_repeat(ctx0, model.ln_f_b, inpL));
|
||||
}
|
||||
|
||||
// inpL = WTE * inpL
|
||||
// [ 768, 50257] - model.wte
|
||||
// [ 768, N] - inpL
|
||||
inpL = ggml_mul_mat(ctx0, model.wte, inpL);
|
||||
|
||||
// logits -> probs
|
||||
inpL = ggml_soft_max(ctx0, inpL);
|
||||
|
||||
// run the computation
|
||||
ggml_build_forward_expand(&gf, inpL);
|
||||
ggml_graph_compute (ctx0, &gf);
|
||||
|
||||
//if (n_past%100 == 0) {
|
||||
// ggml_graph_print (&gf);
|
||||
// ggml_graph_dump_dot(&gf, NULL, "gpt-2.dot");
|
||||
//}
|
||||
|
||||
//embd_w.resize(n_vocab*N);
|
||||
//memcpy(embd_w.data(), ggml_get_data(inpL), sizeof(float)*n_vocab*N);
|
||||
|
||||
// return result for just the last token
|
||||
embd_w.resize(n_vocab);
|
||||
memcpy(embd_w.data(), (float *) ggml_get_data(inpL) + (n_vocab*(N-1)), sizeof(float)*n_vocab);
|
||||
|
||||
if (mem_per_token == 0) {
|
||||
mem_per_token = ggml_used_mem(ctx0)/N;
|
||||
}
|
||||
//printf("used_mem = %zu\n", ggml_used_mem(ctx0));
|
||||
|
||||
ggml_free(ctx0);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/////////////////////////////// GPT-2 END ////////////////////////////////
|
||||
|
||||
constexpr int N_THREAD = 8;
|
||||
|
||||
struct gpt2_context {
|
||||
std::string prompt_base = R"(Hello, how are you?
|
||||
I'm fine, thanks. How are you?
|
||||
Thanks, I'm fine too. What are you doing?
|
||||
I'm just sitting here.
|
||||
It's a lovely day, isn't it?
|
||||
Yes, it is. I love the weather this time of year.
|
||||
I wish it would rain a little bit.
|
||||
Me too.
|
||||
)";
|
||||
|
||||
std::mt19937 rng;
|
||||
|
||||
gpt_vocab vocab;
|
||||
gpt2_model model;
|
||||
|
||||
int32_t n_threads = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
|
||||
|
||||
// sampling parameters
|
||||
int32_t top_k = 20;
|
||||
float top_p = 0.98f;
|
||||
float temp = 1.0f;
|
||||
};
|
||||
|
||||
struct gpt2_context * gpt2_init(const char * path_model) {
|
||||
gpt2_context * ctx = new gpt2_context;
|
||||
|
||||
ctx->rng = std::mt19937(time(NULL));
|
||||
|
||||
// load the model
|
||||
{
|
||||
const int64_t t_start_us = ggml_time_us();
|
||||
|
||||
if (!gpt2_model_load(path_model, ctx->model, ctx->vocab)) {
|
||||
fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, "gpt-2.bin");
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
const int64_t t_load_us = ggml_time_us() - t_start_us;
|
||||
|
||||
printf("gpt-2: model loaded in %d ms\n", (int) (t_load_us/1000));
|
||||
}
|
||||
|
||||
return ctx;
|
||||
}
|
||||
|
||||
void gpt2_free(struct gpt2_context * ctx) {
|
||||
delete ctx;
|
||||
}
|
||||
|
||||
const char * gpt2_get_prompt(struct gpt2_context * ctx) {
|
||||
return ctx->prompt_base.c_str();
|
||||
}
|
||||
|
||||
void gpt2_set_prompt(struct gpt2_context * ctx, const char * prompt) {
|
||||
ctx->prompt_base = prompt;
|
||||
}
|
||||
|
||||
std::vector<gpt_vocab::id> gpt2_tokenize(const gpt2_context * ctx, const char * text) {
|
||||
return ::gpt_tokenize(ctx->vocab, text);
|
||||
}
|
||||
|
||||
std::string gpt2_gen_text(gpt2_context * ctx, const char * text, int max_tokens) {
|
||||
int n_past = 0;
|
||||
|
||||
std::vector<float> embd_w;
|
||||
|
||||
// tokenize the prompt
|
||||
std::vector<gpt_vocab::id> embd_inp = ::gpt2_tokenize(ctx, text);
|
||||
|
||||
int n_predict = std::min(max_tokens, ctx->model.hparams.n_ctx - (int) embd_inp.size());
|
||||
|
||||
std::vector<gpt_vocab::id> embd = embd_inp;
|
||||
|
||||
size_t mem_per_token = 3000000;
|
||||
|
||||
std::string result;
|
||||
|
||||
for (int i = embd.size(); i < embd_inp.size() + n_predict; i++) {
|
||||
// predict
|
||||
if (embd.size() > 0) {
|
||||
if (!gpt2_eval(ctx->model, ctx->n_threads, n_past, embd, embd_w, mem_per_token)) {
|
||||
printf("gpt-2: failed to generate text\n");
|
||||
return "";
|
||||
}
|
||||
}
|
||||
|
||||
n_past += embd.size();
|
||||
embd.clear();
|
||||
|
||||
{
|
||||
// sample next token
|
||||
const int top_k = ctx->top_k;
|
||||
const float top_p = ctx->top_p;
|
||||
const float temp = ctx->temp;
|
||||
|
||||
const int n_vocab = ctx->model.hparams.n_vocab;
|
||||
|
||||
const gpt_vocab::id id = gpt_sample_top_k_top_p(ctx->vocab, embd_w.data() + (embd_w.size() - n_vocab), top_k, top_p, temp, ctx->rng);
|
||||
|
||||
// add it to the context
|
||||
embd.push_back(id);
|
||||
}
|
||||
|
||||
result += ctx->vocab.id_to_token[embd[0]];
|
||||
|
||||
// end of text token
|
||||
if (embd.back() == 50256 ||
|
||||
ctx->vocab.id_to_token[embd.back()] == "." ||
|
||||
ctx->vocab.id_to_token[embd.back()] == "!" ||
|
||||
ctx->vocab.id_to_token[embd.back()] == "?") {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
@ -0,0 +1,27 @@
|
||||
#pragma once
|
||||
|
||||
// TODO: Change to C-style API and move to ./examples for easy reuse.
|
||||
|
||||
#include <vector>
|
||||
#include <map>
|
||||
#include <string>
|
||||
|
||||
struct gpt_vocab {
|
||||
using id = int32_t;
|
||||
using token = std::string;
|
||||
|
||||
std::map<token, id> token_to_id;
|
||||
std::map<id, token> id_to_token;
|
||||
};
|
||||
|
||||
struct gpt2_context;
|
||||
|
||||
struct gpt2_context * gpt2_init(const char * path_model);
|
||||
void gpt2_free(struct gpt2_context * ctx);
|
||||
|
||||
const char * gpt2_get_prompt(struct gpt2_context * ctx);
|
||||
void gpt2_set_prompt(struct gpt2_context * ctx, const char * prompt);
|
||||
|
||||
std::vector<gpt_vocab::id> gpt2_tokenize(const gpt2_context * ctx, const char * text);
|
||||
|
||||
std::string gpt2_gen_text(gpt2_context * ctx, const char * text, int max_tokens);
|
||||
@ -0,0 +1,17 @@
|
||||
#!/bin/bash
|
||||
|
||||
# Usage:
|
||||
# speak.sh <voice_id> <text-to-speak>
|
||||
|
||||
# espeak
|
||||
# Mac OS: brew install espeak
|
||||
# Linux: apt-get install espeak
|
||||
#
|
||||
espeak -v en-us+m$1 -s 175 -p 50 -a 200 -g 5 -k 5 "$2"
|
||||
|
||||
# Eleven Labs
|
||||
#
|
||||
#wd=$(dirname $0)
|
||||
#script=$wd/eleven-labs.py
|
||||
#python3 $script $1 "$2"
|
||||
#ffplay -autoexit -nodisp -loglevel quiet -hide_banner -i ./audio.mp3
|
||||
@ -0,0 +1,733 @@
|
||||
// Talk with AI
|
||||
//
|
||||
|
||||
#include "whisper.h"
|
||||
#include "gpt-2.h"
|
||||
|
||||
#include <SDL.h>
|
||||
#include <SDL_audio.h>
|
||||
|
||||
#include <cassert>
|
||||
#include <cstdio>
|
||||
#include <fstream>
|
||||
#include <mutex>
|
||||
#include <regex>
|
||||
#include <string>
|
||||
#include <thread>
|
||||
#include <vector>
|
||||
#include <regex>
|
||||
|
||||
// command-line parameters
|
||||
struct whisper_params {
|
||||
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
|
||||
int32_t voice_ms = 10000;
|
||||
int32_t capture_id = -1;
|
||||
int32_t max_tokens = 32;
|
||||
int32_t audio_ctx = 0;
|
||||
|
||||
float vad_thold = 0.6f;
|
||||
float freq_thold = 100.0f;
|
||||
|
||||
bool speed_up = false;
|
||||
bool translate = false;
|
||||
bool print_special = false;
|
||||
bool print_energy = false;
|
||||
bool no_timestamps = true;
|
||||
|
||||
std::string person = "Santa";
|
||||
std::string language = "en";
|
||||
std::string model_wsp = "models/ggml-base.en.bin";
|
||||
std::string model_gpt = "models/ggml-gpt-2-117M.bin";
|
||||
std::string speak = "./examples/talk/speak.sh";
|
||||
std::string fname_out = "";
|
||||
};
|
||||
|
||||
void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
|
||||
|
||||
bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
std::string arg = argv[i];
|
||||
|
||||
if (arg == "-h" || arg == "--help") {
|
||||
whisper_print_usage(argc, argv, params);
|
||||
exit(0);
|
||||
}
|
||||
else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
|
||||
else if (arg == "-vms" || arg == "--voice-ms") { params.voice_ms = std::stoi(argv[++i]); }
|
||||
else if (arg == "-c" || arg == "--capture") { params.capture_id = std::stoi(argv[++i]); }
|
||||
else if (arg == "-mt" || arg == "--max-tokens") { params.max_tokens = std::stoi(argv[++i]); }
|
||||
else if (arg == "-ac" || arg == "--audio-ctx") { params.audio_ctx = std::stoi(argv[++i]); }
|
||||
else if (arg == "-vth" || arg == "--vad-thold") { params.vad_thold = std::stof(argv[++i]); }
|
||||
else if (arg == "-fth" || arg == "--freq-thold") { params.freq_thold = std::stof(argv[++i]); }
|
||||
else if (arg == "-su" || arg == "--speed-up") { params.speed_up = true; }
|
||||
else if (arg == "-tr" || arg == "--translate") { params.translate = true; }
|
||||
else if (arg == "-ps" || arg == "--print-special") { params.print_special = true; }
|
||||
else if (arg == "-pe" || arg == "--print-energy") { params.print_energy = true; }
|
||||
else if (arg == "-p" || arg == "--person") { params.person = argv[++i]; }
|
||||
else if (arg == "-l" || arg == "--language") { params.language = argv[++i]; }
|
||||
else if (arg == "-mw" || arg == "--model-whisper") { params.model_wsp = argv[++i]; }
|
||||
else if (arg == "-mg" || arg == "--model-gpt") { params.model_gpt = argv[++i]; }
|
||||
else if (arg == "-s" || arg == "--speak") { params.speak = argv[++i]; }
|
||||
else if (arg == "-f" || arg == "--file") { params.fname_out = argv[++i]; }
|
||||
else {
|
||||
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
|
||||
whisper_print_usage(argc, argv, params);
|
||||
exit(0);
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
|
||||
fprintf(stderr, "\n");
|
||||
fprintf(stderr, "usage: %s [options]\n", argv[0]);
|
||||
fprintf(stderr, "\n");
|
||||
fprintf(stderr, "options:\n");
|
||||
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
|
||||
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
|
||||
fprintf(stderr, " -vms N, --voice-ms N [%-7d] voice duration in milliseconds\n", params.voice_ms);
|
||||
fprintf(stderr, " -c ID, --capture ID [%-7d] capture device ID\n", params.capture_id);
|
||||
fprintf(stderr, " -mt N, --max-tokens N [%-7d] maximum number of tokens per audio chunk\n", params.max_tokens);
|
||||
fprintf(stderr, " -ac N, --audio-ctx N [%-7d] audio context size (0 - all)\n", params.audio_ctx);
|
||||
fprintf(stderr, " -vth N, --vad-thold N [%-7.2f] voice activity detection threshold\n", params.vad_thold);
|
||||
fprintf(stderr, " -fth N, --freq-thold N [%-7.2f] high-pass frequency cutoff\n", params.freq_thold);
|
||||
fprintf(stderr, " -su, --speed-up [%-7s] speed up audio by x2 (reduced accuracy)\n", params.speed_up ? "true" : "false");
|
||||
fprintf(stderr, " -tr, --translate [%-7s] translate from source language to english\n", params.translate ? "true" : "false");
|
||||
fprintf(stderr, " -ps, --print-special [%-7s] print special tokens\n", params.print_special ? "true" : "false");
|
||||
fprintf(stderr, " -pe, --print-energy [%-7s] print sound energy (for debugging)\n", params.print_energy ? "true" : "false");
|
||||
fprintf(stderr, " -p NAME, --person NAME [%-7s] person name (for prompt selection)\n", params.person.c_str());
|
||||
fprintf(stderr, " -l LANG, --language LANG [%-7s] spoken language\n", params.language.c_str());
|
||||
fprintf(stderr, " -mw FILE, --model-whisper [%-7s] whisper model file\n", params.model_wsp.c_str());
|
||||
fprintf(stderr, " -mg FILE, --model-gpt [%-7s] gpt model file\n", params.model_gpt.c_str());
|
||||
fprintf(stderr, " -s FILE, --speak TEXT [%-7s] command for TTS\n", params.speak.c_str());
|
||||
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] text output file name\n", params.fname_out.c_str());
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
|
||||
//
|
||||
// SDL Audio capture
|
||||
//
|
||||
|
||||
class audio_async {
|
||||
public:
|
||||
audio_async(int len_ms);
|
||||
~audio_async();
|
||||
|
||||
bool init(int capture_id, int sample_rate);
|
||||
|
||||
// start capturing audio via the provided SDL callback
|
||||
// keep last len_ms seconds of audio in a circular buffer
|
||||
bool resume();
|
||||
bool pause();
|
||||
bool clear();
|
||||
|
||||
// callback to be called by SDL
|
||||
void callback(uint8_t * stream, int len);
|
||||
|
||||
// get audio data from the circular buffer
|
||||
void get(int ms, std::vector<float> & audio);
|
||||
|
||||
private:
|
||||
SDL_AudioDeviceID m_dev_id_in = 0;
|
||||
|
||||
int m_len_ms = 0;
|
||||
int m_sample_rate = 0;
|
||||
|
||||
bool m_running = false;
|
||||
std::mutex m_mutex;
|
||||
|
||||
std::vector<float> m_audio;
|
||||
std::vector<float> m_audio_new;
|
||||
size_t m_audio_pos = 0;
|
||||
size_t m_audio_len = 0;
|
||||
};
|
||||
|
||||
audio_async::audio_async(int len_ms) {
|
||||
m_len_ms = len_ms;
|
||||
}
|
||||
|
||||
audio_async::~audio_async() {
|
||||
if (m_dev_id_in) {
|
||||
SDL_CloseAudioDevice(m_dev_id_in);
|
||||
}
|
||||
}
|
||||
|
||||
bool audio_async::init(int capture_id, int sample_rate) {
|
||||
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
|
||||
|
||||
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
|
||||
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
|
||||
|
||||
{
|
||||
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
|
||||
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
|
||||
for (int i = 0; i < nDevices; i++) {
|
||||
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
|
||||
}
|
||||
}
|
||||
|
||||
SDL_AudioSpec capture_spec_requested;
|
||||
SDL_AudioSpec capture_spec_obtained;
|
||||
|
||||
SDL_zero(capture_spec_requested);
|
||||
SDL_zero(capture_spec_obtained);
|
||||
|
||||
capture_spec_requested.freq = sample_rate;
|
||||
capture_spec_requested.format = AUDIO_F32;
|
||||
capture_spec_requested.channels = 1;
|
||||
capture_spec_requested.samples = 1024;
|
||||
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
|
||||
audio_async * audio = (audio_async *) userdata;
|
||||
audio->callback(stream, len);
|
||||
};
|
||||
capture_spec_requested.userdata = this;
|
||||
|
||||
if (capture_id >= 0) {
|
||||
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
|
||||
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
|
||||
} else {
|
||||
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
|
||||
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
|
||||
}
|
||||
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
|
||||
m_dev_id_in = 0;
|
||||
|
||||
return false;
|
||||
} else {
|
||||
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
|
||||
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
|
||||
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
|
||||
capture_spec_requested.format);
|
||||
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
|
||||
capture_spec_requested.channels);
|
||||
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
|
||||
m_sample_rate = capture_spec_obtained.freq;
|
||||
|
||||
m_audio.resize((m_sample_rate*m_len_ms)/1000);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::resume() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (m_running) {
|
||||
fprintf(stderr, "%s: already running!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_PauseAudioDevice(m_dev_id_in, 0);
|
||||
|
||||
m_running = true;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::pause() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: already paused!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_PauseAudioDevice(m_dev_id_in, 1);
|
||||
|
||||
m_running = false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::clear() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: not running!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
m_audio_pos = 0;
|
||||
m_audio_len = 0;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
// callback to be called by SDL
|
||||
void audio_async::callback(uint8_t * stream, int len) {
|
||||
if (!m_running) {
|
||||
return;
|
||||
}
|
||||
|
||||
const size_t n_samples = len / sizeof(float);
|
||||
|
||||
m_audio_new.resize(n_samples);
|
||||
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
|
||||
|
||||
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
if (m_audio_pos + n_samples > m_audio.size()) {
|
||||
const size_t n0 = m_audio.size() - m_audio_pos;
|
||||
|
||||
memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
|
||||
memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
|
||||
|
||||
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
|
||||
m_audio_len = m_audio.size();
|
||||
} else {
|
||||
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
|
||||
|
||||
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
|
||||
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void audio_async::get(int ms, std::vector<float> & result) {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
|
||||
return;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: not running!\n", __func__);
|
||||
return;
|
||||
}
|
||||
|
||||
result.clear();
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
if (ms <= 0) {
|
||||
ms = m_len_ms;
|
||||
}
|
||||
|
||||
size_t n_samples = (m_sample_rate * ms) / 1000;
|
||||
if (n_samples > m_audio_len) {
|
||||
n_samples = m_audio_len;
|
||||
}
|
||||
|
||||
result.resize(n_samples);
|
||||
|
||||
int s0 = m_audio_pos - n_samples;
|
||||
if (s0 < 0) {
|
||||
s0 += m_audio.size();
|
||||
}
|
||||
|
||||
if (s0 + n_samples > m_audio.size()) {
|
||||
const size_t n0 = m_audio.size() - s0;
|
||||
|
||||
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
|
||||
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
|
||||
} else {
|
||||
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
///////////////////////////
|
||||
|
||||
std::string trim(const std::string & s) {
|
||||
std::regex e("^\\s+|\\s+$");
|
||||
return std::regex_replace(s, e, "");
|
||||
}
|
||||
|
||||
std::string replace(const std::string & s, const std::string & from, const std::string & to) {
|
||||
std::string result = s;
|
||||
size_t pos = 0;
|
||||
while ((pos = result.find(from, pos)) != std::string::npos) {
|
||||
result.replace(pos, from.length(), to);
|
||||
pos += to.length();
|
||||
}
|
||||
return result;
|
||||
}
|
||||
|
||||
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
|
||||
const float rc = 1.0f / (2.0f * M_PI * cutoff);
|
||||
const float dt = 1.0f / sample_rate;
|
||||
const float alpha = dt / (rc + dt);
|
||||
|
||||
float y = data[0];
|
||||
|
||||
for (size_t i = 1; i < data.size(); i++) {
|
||||
y = alpha * (y + data[i] - data[i - 1]);
|
||||
data[i] = y;
|
||||
}
|
||||
}
|
||||
|
||||
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
|
||||
const int n_samples = pcmf32.size();
|
||||
const int n_samples_last = (sample_rate * last_ms) / 1000;
|
||||
|
||||
if (n_samples_last >= n_samples) {
|
||||
// not enough samples - assume no speech
|
||||
return false;
|
||||
}
|
||||
|
||||
if (freq_thold > 0.0f) {
|
||||
high_pass_filter(pcmf32, freq_thold, sample_rate);
|
||||
}
|
||||
|
||||
float energy_all = 0.0f;
|
||||
float energy_last = 0.0f;
|
||||
|
||||
for (size_t i = 0; i < n_samples; i++) {
|
||||
energy_all += fabsf(pcmf32[i]);
|
||||
|
||||
if (i >= n_samples - n_samples_last) {
|
||||
energy_last += fabsf(pcmf32[i]);
|
||||
}
|
||||
}
|
||||
|
||||
energy_all /= n_samples;
|
||||
energy_last /= n_samples_last;
|
||||
|
||||
if (verbose) {
|
||||
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
|
||||
}
|
||||
|
||||
if (energy_last > vad_thold*energy_all) {
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
|
||||
const auto t_start = std::chrono::high_resolution_clock::now();
|
||||
|
||||
prob = 0.0f;
|
||||
t_ms = 0;
|
||||
|
||||
whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
|
||||
|
||||
wparams.print_progress = false;
|
||||
wparams.print_special = params.print_special;
|
||||
wparams.print_realtime = false;
|
||||
wparams.print_timestamps = !params.no_timestamps;
|
||||
wparams.translate = params.translate;
|
||||
wparams.no_context = true;
|
||||
wparams.single_segment = true;
|
||||
wparams.max_tokens = params.max_tokens;
|
||||
wparams.language = params.language.c_str();
|
||||
wparams.n_threads = params.n_threads;
|
||||
|
||||
wparams.audio_ctx = params.audio_ctx;
|
||||
wparams.speed_up = params.speed_up;
|
||||
|
||||
if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {
|
||||
return "";
|
||||
}
|
||||
|
||||
int prob_n = 0;
|
||||
std::string result;
|
||||
|
||||
const int n_segments = whisper_full_n_segments(ctx);
|
||||
for (int i = 0; i < n_segments; ++i) {
|
||||
const char * text = whisper_full_get_segment_text(ctx, i);
|
||||
|
||||
result += text;
|
||||
|
||||
const int n_tokens = whisper_full_n_tokens(ctx, i);
|
||||
for (int j = 0; j < n_tokens; ++j) {
|
||||
const auto token = whisper_full_get_token_data(ctx, i, j);
|
||||
|
||||
prob += token.p;
|
||||
++prob_n;
|
||||
}
|
||||
}
|
||||
|
||||
if (prob_n > 0) {
|
||||
prob /= prob_n;
|
||||
}
|
||||
|
||||
const auto t_end = std::chrono::high_resolution_clock::now();
|
||||
t_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count();
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// compute similarity between two strings using Levenshtein distance
|
||||
float similarity(const std::string & s0, const std::string & s1) {
|
||||
const size_t len0 = s0.size() + 1;
|
||||
const size_t len1 = s1.size() + 1;
|
||||
|
||||
std::vector<int> col(len1, 0);
|
||||
std::vector<int> prevCol(len1, 0);
|
||||
|
||||
for (size_t i = 0; i < len1; i++) {
|
||||
prevCol[i] = i;
|
||||
}
|
||||
|
||||
for (size_t i = 0; i < len0; i++) {
|
||||
col[0] = i;
|
||||
for (size_t j = 1; j < len1; j++) {
|
||||
col[j] = std::min(std::min(1 + col[j - 1], 1 + prevCol[j]), prevCol[j - 1] + (s0[i - 1] == s1[j - 1] ? 0 : 1));
|
||||
}
|
||||
col.swap(prevCol);
|
||||
}
|
||||
|
||||
const float dist = prevCol[len1 - 1];
|
||||
|
||||
return 1.0f - (dist / std::max(s0.size(), s1.size()));
|
||||
}
|
||||
|
||||
// generated with ChatGPT
|
||||
std::map<std::string, std::string> k_prompts = {
|
||||
{ "Santa",
|
||||
R"(Kid: Hi Santa! Are you real?
|
||||
Santa: Of course I am, my dear! Ho ho ho!
|
||||
Kid: Can you please bring me a new toy for Christmas?
|
||||
Santa: I'll see what I can do, but you have to make sure to be a good boy or girl and listen to your parents.
|
||||
Kid: I will, Santa! Thank you!
|
||||
Santa: You're welcome, little one. Merry Christmas! Ho ho ho!
|
||||
Kid: Can you tell me how you deliver all the presents to all the kids in the world in one night?
|
||||
Santa: It's a secret, but I have a lot of help from my elves and my magical sleigh. And I have a special route that I follow to make sure I visit every child.
|
||||
Kid: Wow, that's amazing! Can I please have a ride in your sleigh sometime?
|
||||
Santa: I'm sorry, but only good boys and girls get to ride in my sleigh.
|
||||
)" },
|
||||
{ "Kid",
|
||||
R"(Kid: Hi Santa! Are you real?
|
||||
Santa: Of course I am, my dear! Ho ho ho!
|
||||
Kid: Can you please bring me a new toy for Christmas?
|
||||
Santa: I'll see what I can do, but you have to make sure to be a good boy or girl and listen to your parents.
|
||||
Kid: I will, Santa! Thank you!
|
||||
Kid: Can you tell me how you deliver all the presents to all the kids in the world in one night?
|
||||
Santa: It's a secret, but I have a lot of help from my elves and my magical sleigh. And I have a special route that I follow to make sure I visit every child.
|
||||
Kid: Wow, that's amazing! Can I please have a ride in your sleigh sometime?
|
||||
)" },
|
||||
};
|
||||
|
||||
int main(int argc, char ** argv) {
|
||||
whisper_params params;
|
||||
|
||||
if (whisper_params_parse(argc, argv, params) == false) {
|
||||
return 1;
|
||||
}
|
||||
|
||||
if (whisper_lang_id(params.language.c_str()) == -1) {
|
||||
fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
|
||||
whisper_print_usage(argc, argv, params);
|
||||
exit(0);
|
||||
}
|
||||
|
||||
// whisper init
|
||||
|
||||
struct whisper_context * ctx_wsp = whisper_init(params.model_wsp.c_str());
|
||||
|
||||
// gpt init
|
||||
|
||||
struct gpt2_context * ctx_gpt = gpt2_init(params.model_gpt.c_str());
|
||||
|
||||
// print some info about the processing
|
||||
{
|
||||
fprintf(stderr, "\n");
|
||||
if (!whisper_is_multilingual(ctx_wsp)) {
|
||||
if (params.language != "en" || params.translate) {
|
||||
params.language = "en";
|
||||
params.translate = false;
|
||||
fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);
|
||||
}
|
||||
}
|
||||
fprintf(stderr, "%s: processing, %d threads, lang = %s, task = %s, timestamps = %d ...\n",
|
||||
__func__,
|
||||
params.n_threads,
|
||||
params.language.c_str(),
|
||||
params.translate ? "translate" : "transcribe",
|
||||
params.no_timestamps ? 0 : 1);
|
||||
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
|
||||
|
||||
// init audio
|
||||
|
||||
audio_async audio(30*1000);
|
||||
if (!audio.init(params.capture_id, WHISPER_SAMPLE_RATE)) {
|
||||
fprintf(stderr, "%s: audio.init() failed!\n", __func__);
|
||||
return 1;
|
||||
}
|
||||
|
||||
audio.resume();
|
||||
|
||||
int n_iter = 0;
|
||||
|
||||
bool is_running = true;
|
||||
bool force_speak = params.person == "Kid";
|
||||
|
||||
float prob0 = 0.0f;
|
||||
float prob = 0.0f;
|
||||
|
||||
std::vector<float> pcmf32_cur;
|
||||
std::vector<float> pcmf32_prompt;
|
||||
|
||||
if (k_prompts.find(params.person) == k_prompts.end()) {
|
||||
fprintf(stderr, "%s: unknown person '%s'\n", __func__, params.person.c_str());
|
||||
return 1;
|
||||
}
|
||||
|
||||
gpt2_set_prompt(ctx_gpt, k_prompts.at(params.person).c_str());
|
||||
|
||||
const std::string person_other = params.person == "Santa" ? "Kid" : "Santa";
|
||||
const int voice_id = params.person == "Santa" ? 5 : 2;
|
||||
|
||||
fprintf(stderr, "gpt-2: prompt_base:\n");
|
||||
fprintf(stderr, "========================\n\n");
|
||||
fprintf(stderr, "%s\n", gpt2_get_prompt(ctx_gpt));
|
||||
fprintf(stderr, "========================\n\n");
|
||||
|
||||
// main loop
|
||||
while (is_running) {
|
||||
// handle Ctrl + C
|
||||
{
|
||||
SDL_Event event;
|
||||
while (SDL_PollEvent(&event)) {
|
||||
switch (event.type) {
|
||||
case SDL_QUIT:
|
||||
{
|
||||
is_running = false;
|
||||
} break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!is_running) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// delay
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(100));
|
||||
|
||||
int64_t t_ms = 0;
|
||||
|
||||
{
|
||||
audio.get(2000, pcmf32_cur);
|
||||
|
||||
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1250, params.vad_thold, params.freq_thold, params.print_energy) || force_speak) {
|
||||
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
|
||||
|
||||
audio.get(params.voice_ms, pcmf32_cur);
|
||||
|
||||
std::string text_heard = "Hey little one, what do you want for Christmas?";
|
||||
if (!force_speak) {
|
||||
text_heard = ::trim(::transcribe(ctx_wsp, params, pcmf32_cur, prob0, t_ms));
|
||||
}
|
||||
|
||||
force_speak = false;
|
||||
|
||||
// remove text between brackets using regex
|
||||
{
|
||||
std::regex re("\\[.*?\\]");
|
||||
text_heard = std::regex_replace(text_heard, re, "");
|
||||
}
|
||||
|
||||
// remove text between brackets using regex
|
||||
{
|
||||
std::regex re("\\(.*?\\)");
|
||||
text_heard = std::regex_replace(text_heard, re, "");
|
||||
}
|
||||
|
||||
// remove all characters, except for letters, numbers, punctuation and ':', '\'', '-', ' '
|
||||
text_heard = std::regex_replace(text_heard, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
|
||||
|
||||
// take first line
|
||||
text_heard = text_heard.substr(0, text_heard.find_first_of("\n"));
|
||||
|
||||
// remove leading and trailing whitespace
|
||||
text_heard = std::regex_replace(text_heard, std::regex("^\\s+"), "");
|
||||
text_heard = std::regex_replace(text_heard, std::regex("\\s+$"), "");
|
||||
|
||||
const std::vector<gpt_vocab::id> tokens = gpt2_tokenize(ctx_gpt, text_heard.c_str());
|
||||
|
||||
if (text_heard.empty() || tokens.empty()) {
|
||||
fprintf(stdout, "%s: Heard nothing, skipping ...\n", __func__);
|
||||
audio.clear();
|
||||
|
||||
continue;
|
||||
}
|
||||
|
||||
fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", text_heard.c_str(), "\033[0m", (int) t_ms);
|
||||
|
||||
std::string prompt_base = gpt2_get_prompt(ctx_gpt);
|
||||
|
||||
std::string text_to_speak;
|
||||
|
||||
{
|
||||
text_heard = person_other + ": " + text_heard;
|
||||
|
||||
text_to_speak = gpt2_gen_text(ctx_gpt, (prompt_base + text_heard + "\n").c_str(), params.max_tokens);
|
||||
text_to_speak = std::regex_replace(text_to_speak, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
|
||||
text_to_speak = text_to_speak.substr(0, text_to_speak.find_first_of("\n"));
|
||||
|
||||
// remove first 2 lines of base prompt
|
||||
if (n_iter > 4) {
|
||||
{
|
||||
const size_t pos = prompt_base.find_first_of("\n");
|
||||
if (pos != std::string::npos) {
|
||||
prompt_base = prompt_base.substr(pos + 1);
|
||||
}
|
||||
}
|
||||
{
|
||||
const size_t pos = prompt_base.find_first_of("\n");
|
||||
if (pos != std::string::npos) {
|
||||
prompt_base = prompt_base.substr(pos + 1);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
prompt_base += text_heard + "\n" + text_to_speak + "\n";
|
||||
}
|
||||
|
||||
printf("%s\n", text_to_speak.c_str());
|
||||
|
||||
//printf("========================\n");
|
||||
//printf("gpt-2: prompt_base:\n'%s'\n", prompt_base.c_str());
|
||||
//printf("========================\n");
|
||||
|
||||
gpt2_set_prompt(ctx_gpt, prompt_base.c_str());
|
||||
|
||||
text_to_speak = ::replace(text_to_speak, params.person + ": ", "");
|
||||
system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
|
||||
|
||||
audio.clear();
|
||||
|
||||
++n_iter;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
audio.pause();
|
||||
|
||||
whisper_print_timings(ctx_wsp);
|
||||
whisper_free(ctx_wsp);
|
||||
|
||||
return 0;
|
||||
}
|
||||
Loading…
Reference in new issue