-Wextra
-Wpedantic
-Wcast-qual
+ -Wmissing-declarations
-Wno-unused-function
-Wno-multichar
)
# warnings
MK_CFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wdouble-promotion -Wshadow -Wstrict-prototypes -Wpointer-arith \
-Wmissing-prototypes -Werror=implicit-int -Wno-unused-function
-MK_CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wno-multichar
+MK_CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wmissing-declarations -Wno-unused-function -Wno-multichar
-ifeq '' '$(findstring clang,$(shell $(CXX) --version))'
+# TODO(cebtenzzre): remove this once PR #2632 gets merged
+TTFS_CXXFLAGS = $(CXXFLAGS) -Wno-missing-declarations
+
+ifneq '' '$(findstring clang,$(shell $(CXX) --version))'
+ # clang++ only
+ MK_CXXFLAGS += -Wmissing-prototypes
+ TTFS_CXXFLAGS += -Wno-missing-prototypes
+else
# g++ only
MK_CXXFLAGS += -Wno-format-truncation -Wno-array-bounds
endif
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
train-text-from-scratch: examples/train-text-from-scratch/train-text-from-scratch.cpp ggml.o llama.o common.o $(OBJS)
- $(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
+ $(CXX) $(TTFS_CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
convert-llama2c-to-ggml: examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp ggml.o llama.o $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4;
}
-void process_escapes(std::string& input) {
+static void process_escapes(std::string& input) {
std::size_t input_len = input.length();
std::size_t output_idx = 0;
}
}
- char32_t getchar32() {
+ static char32_t getchar32() {
#if defined(_WIN32)
HANDLE hConsole = GetStdHandle(STD_INPUT_HANDLE);
wchar_t high_surrogate = 0;
#endif
}
- void pop_cursor() {
+ static void pop_cursor() {
#if defined(_WIN32)
if (hConsole != NULL) {
CONSOLE_SCREEN_BUFFER_INFO bufferInfo;
putc('\b', out);
}
- int estimateWidth(char32_t codepoint) {
+ static int estimateWidth(char32_t codepoint) {
#if defined(_WIN32)
(void)codepoint;
return 1;
#endif
}
- int put_codepoint(const char* utf8_codepoint, size_t length, int expectedWidth) {
+ static int put_codepoint(const char* utf8_codepoint, size_t length, int expectedWidth) {
#if defined(_WIN32)
CONSOLE_SCREEN_BUFFER_INFO bufferInfo;
if (!GetConsoleScreenBufferInfo(hConsole, &bufferInfo)) {
#endif
}
- void replace_last(char ch) {
+ static void replace_last(char ch) {
#if defined(_WIN32)
pop_cursor();
put_codepoint(&ch, 1, 1);
#endif
}
- void append_utf8(char32_t ch, std::string & out) {
+ static void append_utf8(char32_t ch, std::string & out) {
if (ch <= 0x7F) {
out.push_back(static_cast<unsigned char>(ch));
} else if (ch <= 0x7FF) {
}
// Helper function to remove the last UTF-8 character from a string
- void pop_back_utf8_char(std::string & line) {
+ static void pop_back_utf8_char(std::string & line) {
if (line.empty()) {
return;
}
line.erase(pos);
}
- bool readline_advanced(std::string & line, bool multiline_input) {
+ static bool readline_advanced(std::string & line, bool multiline_input) {
if (out != stdout) {
fflush(stdout);
}
return has_more;
}
- bool readline_simple(std::string & line, bool multiline_input) {
+ static bool readline_simple(std::string & line, bool multiline_input) {
#if defined(_WIN32)
std::wstring wline;
if (!std::getline(std::wcin, wline)) {
namespace grammar_parser {
// NOTE: assumes valid utf8 (but checks for overrun)
// copied from llama.cpp
- std::pair<uint32_t, const char *> decode_utf8(const char * src) {
+ static std::pair<uint32_t, const char *> decode_utf8(const char * src) {
static const int lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 };
uint8_t first_byte = static_cast<uint8_t>(*src);
uint8_t highbits = first_byte >> 4;
return std::make_pair(value, pos);
}
- uint32_t get_symbol_id(parse_state & state, const char * src, size_t len) {
+ static uint32_t get_symbol_id(parse_state & state, const char * src, size_t len) {
uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size());
auto result = state.symbol_ids.insert(std::make_pair(std::string(src, len), next_id));
return result.first->second;
}
- uint32_t generate_symbol_id(parse_state & state, const std::string & base_name) {
+ static uint32_t generate_symbol_id(parse_state & state, const std::string & base_name) {
uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size());
state.symbol_ids[base_name + '_' + std::to_string(next_id)] = next_id;
return next_id;
}
- void add_rule(
+ static void add_rule(
parse_state & state,
uint32_t rule_id,
const std::vector<llama_grammar_element> & rule) {
state.rules[rule_id] = rule;
}
- bool is_word_char(char c) {
+ static bool is_word_char(char c) {
return ('a' <= c && c <= 'z') || ('A' <= c && c <= 'Z') || c == '-' || ('0' <= c && c <= '9');
}
- std::pair<uint32_t, const char *> parse_hex(const char * src, int size) {
+ static std::pair<uint32_t, const char *> parse_hex(const char * src, int size) {
const char * pos = src;
const char * end = src + size;
uint32_t value = 0;
return std::make_pair(value, pos);
}
- const char * parse_space(const char * src, bool newline_ok) {
+ static const char * parse_space(const char * src, bool newline_ok) {
const char * pos = src;
while (*pos == ' ' || *pos == '\t' || *pos == '#' ||
(newline_ok && (*pos == '\r' || *pos == '\n'))) {
return pos;
}
- const char * parse_name(const char * src) {
+ static const char * parse_name(const char * src) {
const char * pos = src;
while (is_word_char(*pos)) {
pos++;
return pos;
}
- std::pair<uint32_t, const char *> parse_char(const char * src) {
+ static std::pair<uint32_t, const char *> parse_char(const char * src) {
if (*src == '\\') {
switch (src[1]) {
case 'x': return parse_hex(src + 2, 2);
uint32_t rule_id,
bool is_nested);
- const char * parse_sequence(
+ static const char * parse_sequence(
parse_state & state,
const char * src,
const std::string & rule_name,
return pos;
}
- const char * parse_rule(parse_state & state, const char * src) {
+ static const char * parse_rule(parse_state & state, const char * src) {
const char * name_end = parse_name(src);
const char * pos = parse_space(name_end, false);
size_t name_len = name_end - src;
}
}
- void print_grammar_char(FILE * file, uint32_t c) {
+ static void print_grammar_char(FILE * file, uint32_t c) {
if (0x20 <= c && c <= 0x7f) {
fprintf(file, "%c", static_cast<char>(c));
} else {
}
}
- bool is_char_element(llama_grammar_element elem) {
+ static bool is_char_element(llama_grammar_element elem) {
switch (elem.type) {
case LLAMA_GRETYPE_CHAR: return true;
case LLAMA_GRETYPE_CHAR_NOT: return true;
}
}
- void print_rule_binary(FILE * file, const std::vector<llama_grammar_element> & rule) {
+ static void print_rule_binary(FILE * file, const std::vector<llama_grammar_element> & rule) {
for (auto elem : rule) {
switch (elem.type) {
case LLAMA_GRETYPE_END: fprintf(file, "END"); break;
fprintf(file, "\n");
}
- void print_rule(
+ static void print_rule(
FILE * file,
uint32_t rule_id,
const std::vector<llama_grammar_element> & rule,
#endif
#ifdef LLAMA_DEFAULT_RMS_EPS
-static const float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS;
+constexpr float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS;
#else
-static const float rms_norm_eps = 5e-6f;
+constexpr float rms_norm_eps = 5e-6f;
#endif
-float frand() {
+static float frand() {
return (float)rand()/(float)RAND_MAX;
}
float max;
};
-void init_random_normal_distribution(struct random_normal_distribution * rnd, int seed, float mean, float std, float min, float max) {
+static void init_random_normal_distribution(
+ struct random_normal_distribution * rnd, int seed, float mean, float std, float min, float max
+) {
rnd->gen = std::mt19937(seed);
rnd->nd = std::normal_distribution<float>{mean, std};
rnd->min = min;
rnd->max = max;
}
-float frand_normal(struct random_normal_distribution * rnd) {
+static float frand_normal(struct random_normal_distribution * rnd) {
const float r = rnd->nd(rnd->gen);
return ((r < rnd->min) ? (rnd->min) : (r > rnd->max) ? (rnd->max) : r);
}
-void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph * graph, int n_threads) {
+static void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph * graph, int n_threads) {
struct ggml_cplan plan = ggml_graph_plan(graph, n_threads);
if (plan.work_size > 0) {
ggml_graph_compute(graph, &plan);
}
-struct ggml_tensor * randomize_tensor(
- struct ggml_tensor * tensor,
- int ndims,
- const int64_t ne[],
- float fmin,
- float fmax) {
-
+static struct ggml_tensor * randomize_tensor(
+ struct ggml_tensor * tensor, int ndims, const int64_t ne[], float fmin, float fmax
+) {
switch (ndims) {
case 1:
for (int i0 = 0; i0 < ne[0]; i0++) {
return tensor;
}
-struct ggml_tensor * randomize_tensor_normal(
- struct ggml_tensor * tensor,
- int ndims,
- const int64_t ne[],
- struct random_normal_distribution * rnd) {
+static struct ggml_tensor * randomize_tensor_normal(
+ struct ggml_tensor * tensor, int ndims, const int64_t ne[], struct random_normal_distribution * rnd
+) {
float scale = 1.0; // xavier
switch (ndims) {
case 1:
}
};
-uint32_t get_n_ff(const struct llama_hparams* hparams) {
+static uint32_t get_n_ff(const struct llama_hparams* hparams) {
const uint32_t n_ff = ((2*(4*hparams->n_embd)/3 + hparams->n_mult - 1)/hparams->n_mult)*hparams->n_mult;
return n_ff;
}
std::vector<llama_layer_lora> layers;
};
-void init_model(struct llama_model * model) {
+static void init_model(struct llama_model * model) {
const auto & hparams = model->hparams;
const uint32_t n_embd = hparams.n_embd;
}
-void init_model_lora(struct llama_model_lora * model) {
+static void init_model_lora(struct llama_model_lora * model) {
const auto & hparams = model->hparams;
const uint32_t n_embd = hparams.n_embd;
}
}
-void set_param_model(struct llama_model * model) {
+static void set_param_model(struct llama_model * model) {
const auto& hparams = model->hparams;
const uint32_t n_layer = hparams.n_layer;
}
}
-void set_param_model_lora(struct llama_model_lora * model) {
+static void set_param_model_lora(struct llama_model_lora * model) {
const auto& hparams = model->hparams;
const uint32_t n_layer = hparams.n_layer;
}
}
-void randomize_model(struct llama_model * model, int seed, float mean, float std, float min, float max) {
+static void randomize_model(struct llama_model * model, int seed, float mean, float std, float min, float max) {
const auto & hparams = model->hparams;
const uint32_t n_layer = hparams.n_layer;
}
-void randomize_model_lora(struct llama_model_lora * model, int seed, float mean, float std, float min, float max) {
+static void randomize_model_lora(
+ struct llama_model_lora * model, int seed, float mean, float std, float min, float max
+) {
const auto & hparams = model->hparams;
const uint32_t n_layer = hparams.n_layer;
}
}
-bool init_kv_cache(struct llama_kv_cache* cache, struct llama_model * model, int n_batch) {
+static bool init_kv_cache(struct llama_kv_cache* cache, struct llama_model * model, int n_batch) {
const auto & hparams = model->hparams;
const uint32_t n_ctx = hparams.n_ctx;
return true;
}
-bool init_kv_cache_lora(struct llama_kv_cache* cache, struct llama_model_lora * model, int n_batch) {
+static bool init_kv_cache_lora(struct llama_kv_cache* cache, struct llama_model_lora * model, int n_batch) {
const auto & hparams = model->hparams;
const uint32_t n_ctx = hparams.n_ctx;
return true;
}
-struct ggml_tensor * forward(
- struct llama_model * model,
- struct llama_kv_cache * cache,
- struct ggml_context * ctx0,
- struct ggml_cgraph * gf,
- struct ggml_tensor * tokens_input,
- const int n_tokens,
- const int n_past) {
-
+static struct ggml_tensor * forward(
+ struct llama_model * model,
+ struct llama_kv_cache * cache,
+ struct ggml_context * ctx0,
+ struct ggml_cgraph * gf,
+ struct ggml_tensor * tokens_input,
+ const int n_tokens,
+ const int n_past
+) {
const int N = n_tokens;
struct llama_kv_cache& kv_self = *cache;
return inpL;
}
-void assert_shape_1d(struct ggml_tensor * tensor, int64_t ne0) {
+static void assert_shape_1d(struct ggml_tensor * tensor, int64_t ne0) {
GGML_ASSERT(tensor->n_dims == 1);
GGML_ASSERT(tensor->ne[0] == ne0);
}
-void assert_shape_2d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1) {
+static void assert_shape_2d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1) {
GGML_ASSERT(tensor->n_dims == 2);
GGML_ASSERT(tensor->ne[0] == ne0);
GGML_ASSERT(tensor->ne[1] == ne1);
}
-void assert_shape_3d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1, int64_t ne2) {
+static void assert_shape_3d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1, int64_t ne2) {
GGML_ASSERT(tensor->n_dims == 3);
GGML_ASSERT(tensor->ne[0] == ne0);
GGML_ASSERT(tensor->ne[1] == ne1);
GGML_ASSERT(tensor->ne[2] == ne2);
}
-void assert_shape_4d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1, int64_t ne2, int64_t ne3) {
+static void assert_shape_4d(struct ggml_tensor * tensor, int64_t ne0, int64_t ne1, int64_t ne2, int64_t ne3) {
GGML_ASSERT(tensor->n_dims == 4);
GGML_ASSERT(tensor->ne[0] == ne0);
GGML_ASSERT(tensor->ne[1] == ne1);
GGML_ASSERT(tensor->ne[3] == ne3);
}
-struct ggml_tensor * forward_batch(
- struct llama_model * model,
- struct llama_kv_cache * cache,
- struct ggml_context * ctx0,
- struct ggml_cgraph * gf,
- struct ggml_tensor * tokens_input,
- const int n_tokens,
- const int n_past,
- const int n_batch) {
-
+static struct ggml_tensor * forward_batch(
+ struct llama_model * model,
+ struct llama_kv_cache * cache,
+ struct ggml_context * ctx0,
+ struct ggml_cgraph * gf,
+ struct ggml_tensor * tokens_input,
+ const int n_tokens,
+ const int n_past,
+ const int n_batch
+) {
const int N = n_tokens;
struct llama_kv_cache& kv_self = *cache;
return inpL;
}
-
-struct ggml_tensor * forward_lora(
- struct llama_model_lora * model,
- struct llama_kv_cache * cache,
- struct ggml_context * ctx0,
- struct ggml_cgraph * gf,
- struct ggml_tensor * tokens_input,
- const int n_tokens,
- const int n_past) {
-
+static struct ggml_tensor * forward_lora(
+ struct llama_model_lora * model,
+ struct llama_kv_cache * cache,
+ struct ggml_context * ctx0,
+ struct ggml_cgraph * gf,
+ struct ggml_tensor * tokens_input,
+ const int n_tokens,
+ const int n_past
+) {
const int N = n_tokens;
struct llama_kv_cache& kv_self = *cache;
return inpL;
}
-void sample_softmax(struct ggml_tensor * logits, struct ggml_tensor * probs, struct ggml_tensor * best_samples) {
+static void sample_softmax(struct ggml_tensor * logits, struct ggml_tensor * probs, struct ggml_tensor * best_samples) {
assert(logits->n_dims == 2);
assert(probs->n_dims == 2);
assert(best_samples->n_dims == 1);
}
}
-void sample_softmax_batch(struct ggml_context * ctx, struct ggml_tensor * logits, struct ggml_tensor * probs, struct ggml_tensor * best_samples) {
+static void sample_softmax_batch(
+ struct ggml_context * ctx, struct ggml_tensor * logits, struct ggml_tensor * probs,
+ struct ggml_tensor * best_samples
+) {
GGML_ASSERT(best_samples->n_dims == 2);
GGML_ASSERT(logits->n_dims == 3);
GGML_ASSERT(probs->n_dims == 3);
}
}
-void print_row(struct ggml_tensor * probs, int i) {
+static void print_row(struct ggml_tensor * probs, int i) {
for (int k = 0; k < probs->ne[0]; ++k) {
float p = ggml_get_f32_1d(probs, i*probs->ne[0] + k);
printf(" %.2f", p);
printf("\n");
}
-void print_matrix(struct ggml_tensor * probs) {
+static void print_matrix(struct ggml_tensor * probs) {
assert(probs->n_dims == 2);
for (int i = 0; i < probs->ne[1]; ++i) {
for (int k = 0; k < probs->ne[0]; ++k) {
}
}
-void print_token(int token, int n_vocab) {
+static void print_token(int token, int n_vocab) {
for (int k = 0; k < token; ++k) {
printf(" ");
}
printf("\n");
}
-void print_tokens(struct ggml_tensor * tokens, int n_vocab) {
+static void print_tokens(struct ggml_tensor * tokens, int n_vocab) {
for (int i=0; i<tokens->ne[0]; ++i) {
int token = ggml_get_i32_1d(tokens, i);
print_token(token, n_vocab);
}
}
-void get_example_targets(int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * targets) {
+static void get_example_targets(int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * targets) {
int n_tokens = tokens_input->ne[0];
int n_vocab = targets->ne[0];
float randomness = 0.0f;
}
}
-void get_example_targets_batch(struct ggml_context * ctx, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * targets) {
+static void get_example_targets_batch(
+ struct ggml_context * ctx, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * targets
+) {
GGML_ASSERT(tokens_input->n_dims == 2);
GGML_ASSERT( targets->n_dims == 3);
int n_tokens = tokens_input->ne[0];
}
}
-void lshift_examples(struct ggml_tensor * tokens_input, struct ggml_tensor * targets, int n_shift) {
+static void lshift_examples(struct ggml_tensor * tokens_input, struct ggml_tensor * targets, int n_shift) {
int n_tokens = tokens_input->ne[0];
int n_vocab = targets->ne[0];
for (int i=0; i<n_tokens-n_shift; ++i) {
}
}
-struct ggml_tensor * square_error_loss(struct ggml_context * ctx, struct ggml_tensor * a, struct ggml_tensor * b) {
+static struct ggml_tensor * square_error_loss(
+ struct ggml_context * ctx, struct ggml_tensor * a, struct ggml_tensor * b
+) {
// todo: instead of a-b: a[1:]-b[:-1]
return ggml_sum(ctx, ggml_sqr(ctx, ggml_sub(ctx, a, b)));
}
-struct ggml_tensor * cross_entropy_loss(struct ggml_context * ctx, struct ggml_tensor * a, struct ggml_tensor * b) {
+static struct ggml_tensor * cross_entropy_loss(
+ struct ggml_context * ctx, struct ggml_tensor * a, struct ggml_tensor * b
+) {
const float eps = 1e-3f;
return
ggml_sum(ctx,
llama_context * ctx;
llama_beam_view beam_view;
};
-std::ostream& operator<<(std::ostream& os, const ostream_beam_view & obv) {
+
+static std::ostream & operator<<(std::ostream & os, const ostream_beam_view & obv) {
os << "p(" << obv.beam_view.p << ") eob(" << std::boolalpha << obv.beam_view.eob << ") tokens(";
for (size_t i = 0 ; i < obv.beam_view.n_tokens ; ++i) {
os << llama_token_to_piece(obv.ctx, obv.beam_view.tokens[i]);
// In this case, end-of-beam (eob) is equivalent to end-of-sentence (eos) but this need not always be the same.
// For example, eob can be flagged due to maximum token length, stop words, etc.
-bool is_at_eob(const beam_search_callback_data & callback_data, const llama_token * tokens, const size_t n_tokens) {
+static bool is_at_eob(const beam_search_callback_data & callback_data, const llama_token * tokens, size_t n_tokens) {
return n_tokens && tokens[n_tokens-1] == llama_token_eos(callback_data.ctx);
}
// * When all beams converge to a common prefix, they are made available in beams_state.beams[0].
// This is also called when the stop condition is met.
// Collect tokens into std::vector<llama_token> response which is pointed to by callback_data.
-void beam_search_callback(void * callback_data_ptr, llama_beams_state beams_state) {
+static void beam_search_callback(void * callback_data_ptr, llama_beams_state beams_state) {
auto& callback_data = *static_cast<beam_search_callback_data*>(callback_data_ptr);
// Mark beams as EOS as needed.
for (size_t i = 0 ; i < beams_state.n_beams ; ++i) {
}
};
-void malloc_weights(TransformerWeights* w, Config* p, bool shared_weights) {
+static void malloc_weights(TransformerWeights* w, Config* p, bool shared_weights) {
// we calloc instead of malloc to keep valgrind happy
w->token_embedding_table = new float[p->vocab_size * p->dim]();
printf("[%s:AK] Allocating [%d] x [%d] = [%d] float space for w->token_embedding_table\n",__func__,p->vocab_size , p->dim, p->vocab_size * p->dim);
}
}
-int checkpoint_init_weights(TransformerWeights *w, Config* p, FILE* f, bool shared_weights) {
+static int checkpoint_init_weights(TransformerWeights *w, Config* p, FILE* f, bool shared_weights) {
if (fread(w->token_embedding_table, sizeof(float), p->vocab_size * p->dim, f) != static_cast<size_t>(p->vocab_size * p->dim)) return 1;
if (fread(w->rms_att_weight, sizeof(float), p->n_layers * p->dim, f) != static_cast<size_t>(p->n_layers * p->dim)) return 1;
if (fread(w->wq, sizeof(float), p->n_layers * p->dim * p->dim, f) != static_cast<size_t>(p->n_layers * p->dim * p->dim)) return 1;
return 0;
}
-void print_sample_weights(TransformerWeights *w){
+static void print_sample_weights(TransformerWeights *w){
printf("----- Quick print of first of the weight vales of all the variables\n");
printf("%f\n", w->token_embedding_table[0]);
printf("%f\n", w->rms_att_weight[0]);
int mem_compute1_gb;
};
-void print_params(struct my_llama_hparams * params) {
+static void print_params(struct my_llama_hparams * params) {
printf("%s: n_vocab: %d\n", __func__, params->n_vocab);
printf("%s: n_ctx: %d\n", __func__, params->n_ctx);
printf("%s: n_embd: %d\n", __func__, params->n_embd);
printf("%s: n_rot: %d\n", __func__, params->n_rot);
}
-void init_model(struct my_llama_model * model) {
+static void init_model(struct my_llama_model * model) {
const auto & hparams = model->hparams;
const uint32_t n_embd = hparams.n_embd;
}
}
-float get_f32_2d(struct ggml_tensor * tensor, int64_t i0, int64_t i1) {
+static float get_f32_2d(struct ggml_tensor * tensor, int64_t i0, int64_t i1) {
float * ptr = (float *) ((char *) tensor->data + i0*tensor->nb[0] + i1*tensor->nb[1]);
return *ptr;
}
-int32_t get_i32_2d(struct ggml_tensor * tensor, int64_t i0, int64_t i1) {
+static int32_t get_i32_2d(struct ggml_tensor * tensor, int64_t i0, int64_t i1) {
int32_t * ptr = (int32_t *) ((char *) tensor->data + i0*tensor->nb[0] + i1*tensor->nb[1]);
return *ptr;
}
-void print_row(struct ggml_tensor * probs, int i) {
+static void print_row(struct ggml_tensor * probs, int i) {
for (int k = 0; k < probs->ne[0]; ++k) {
float p = get_f32_2d(probs, k, i);
printf(" %f", p);
printf("\n");
}
-void print_matrix(struct ggml_tensor * probs) {
+static void print_matrix(struct ggml_tensor * probs) {
assert(probs->n_dims == 2);
for (int i = 0; i < probs->ne[1]; ++i) {
for (int k = 0; k < probs->ne[0]; ++k) {
}
};
-bool is_ggml_file(const char *filename) {
+static bool is_ggml_file(const char * filename) {
llama_file file(filename, "rb");
if (file.size < 4) {
return false;
return magic == GGUF_MAGIC;
}
-static std::string llama_escape_whitespaces(const std::string& text) {
+static std::string llama_escape_whitespaces(const std::string & text) {
std::ostringstream out;
for (char c : text) {
if (c == ' ') out << "\xe2\x96\x81";
return out.str();
}
-void load_vocab(const char *filename, Config *config, struct llama_vocab *vocab) {
+static void load_vocab(const char *filename, Config *config, struct llama_vocab *vocab) {
if (is_ggml_file(filename)) {
struct ggml_context * ctx_data = NULL;
}
}
-void convert_weights_ak_to_gg(struct ggml_tensor * gg_weights, const float * karpathy_weights) {
+static void convert_weights_ak_to_gg(struct ggml_tensor * gg_weights, const float * karpathy_weights) {
int ct;
switch (gg_weights->n_dims){
case 1:
}
}
-void save_as_llama_model(struct llama_vocab * vocab, struct my_llama_model * model, TransformerWeights* w, const char * filename) {
+static void save_as_llama_model(
+ struct llama_vocab * vocab, struct my_llama_model * model, TransformerWeights* w, const char * filename
+) {
// convert AK weights into GG weights one by one.
// w->token_embedding_table -> model->tok_embeddings
// float* -> struct ggml_tensor
gguf_free(ctx);
}
-struct train_params get_default_train_params() {
+static struct train_params get_default_train_params() {
struct train_params params;
params.fn_vocab_model = "models/7B/ggml-model-f16.gguf";
params.fn_llama2c_output_model = "ak_llama_model.bin";
return params;
}
-void print_usage(int /*argc*/, char ** argv, const struct train_params * params) {
+static void print_usage(int /*argc*/, char ** argv, const struct train_params * params) {
fprintf(stderr, "usage: %s [options]\n", argv[0]);
fprintf(stderr, "\n");
fprintf(stderr, "options:\n");
fprintf(stderr, "\n");
}
-bool params_parse(int argc, char ** argv, struct train_params * params) {
+static bool params_parse(int argc, char ** argv, struct train_params * params) {
bool invalid_param = false;
bool reqd_param_found = false;
std::string arg;
return true;
}
-std::string basename(const std::string &path) {
+static std::string basename(const std::string &path) {
size_t pos = path.find_last_of("/\\");
if (pos == std::string::npos) {
return path;
#define MIN(a, b) ((a) < (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))
-template<typename T>
+template <typename T>
static std::string to_string(const T & val) {
std::stringstream ss;
ss << val;
return ss.str();
}
-bool gguf_ex_write(const std::string & fname) {
+static bool gguf_ex_write(const std::string & fname) {
struct gguf_context * ctx = gguf_init_empty();
gguf_set_val_u8 (ctx, "some.parameter.uint8", 0x12);
}
// just read tensor info
-bool gguf_ex_read_0(const std::string & fname) {
+static bool gguf_ex_read_0(const std::string & fname) {
struct gguf_init_params params = {
/*.no_alloc = */ false,
/*.ctx = */ NULL,
}
// read and create ggml_context containing the tensors and their data
-bool gguf_ex_read_1(const std::string & fname) {
+static bool gguf_ex_read_1(const std::string & fname) {
struct ggml_context * ctx_data = NULL;
struct gguf_init_params params = {
static std::vector<llama_token> * g_output_tokens;
static bool is_interacting = false;
-void write_logfile(
+
+static void write_logfile(
const llama_context * ctx, const gpt_params & params, const llama_model * model,
const std::vector<llama_token> & input_tokens, const std::string & output,
const std::vector<llama_token> & output_tokens
}
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
-void sigint_handler(int signo) {
+static void sigint_handler(int signo) {
if (signo == SIGINT) {
if (!is_interacting) {
is_interacting = true;
float prob;
};
-void write_logfile(const llama_context * ctx, const gpt_params & params,
- const llama_model * model, const struct results_perplexity & results) {
-
+static void write_logfile(
+ const llama_context * ctx, const gpt_params & params, const llama_model * model,
+ const struct results_perplexity & results
+) {
if (params.logdir.empty()) {
return;
}
fclose(logfile);
}
-std::vector<float> softmax(const std::vector<float>& logits) {
+static std::vector<float> softmax(const std::vector<float>& logits) {
std::vector<float> probs(logits.size());
float max_logit = logits[0];
for (float v : logits) max_logit = std::max(max_logit, v);
return probs;
}
-results_log_softmax log_softmax(int n_vocab, const float * logits, int tok) {
+static results_log_softmax log_softmax(int n_vocab, const float * logits, int tok) {
float max_logit = logits[0];
for (int i = 1; i < n_vocab; ++i) max_logit = std::max(max_logit, logits[i]);
double sum_exp = 0.0;
return {logits[tok] - max_logit - log(sum_exp), logits[tok], expf(logits[tok] - max_logit) / (float) sum_exp};
}
-void process_logits(int n_vocab, const float * logits, const int * tokens, int n_token, std::vector<std::thread> & workers,
- double & nll, double & nll2, float * logit_history, float * prob_history) {
-
+static void process_logits(
+ int n_vocab, const float * logits, const int * tokens, int n_token, std::vector<std::thread> & workers,
+ double & nll, double & nll2, float * logit_history, float * prob_history
+) {
std::mutex mutex;
int counter = 0;
auto compute = [&mutex, &counter, &nll, &nll2, logit_history, prob_history, n_vocab, logits, tokens, n_token] () {
}
-results_perplexity perplexity_v2(llama_context * ctx, const gpt_params & params) {
+static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params & params) {
// Download: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip?ref=salesforce-research
// Run `./perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
// Output: `perplexity: 13.5106 [114/114]`
return {tokens, std::exp(nll / count), logit_history, prob_history};
}
-results_perplexity perplexity(llama_context * ctx, const gpt_params & params) {
-
+static results_perplexity perplexity(llama_context * ctx, const gpt_params & params) {
if (params.ppl_stride > 0) {
return perplexity_v2(ctx, params);
}
return {tokens, ppl, logit_history, prob_history};
}
-std::vector<float> hellaswag_evaluate_tokens(llama_context * ctx, const std::vector<int>& tokens, int n_past, int n_batch,
- int n_vocab, int n_thread) {
+static std::vector<float> hellaswag_evaluate_tokens(
+ llama_context * ctx, const std::vector<int>& tokens, int n_past, int n_batch, int n_vocab, int n_thread
+) {
std::vector<float> result;
result.reserve(tokens.size() * n_vocab);
size_t n_chunk = (tokens.size() + n_batch - 1)/n_batch;
return result;
}
-void hellaswag_score(llama_context * ctx, const gpt_params & params) {
+static void hellaswag_score(llama_context * ctx, const gpt_params & params) {
// Calculates hellaswag score (acc_norm) from prompt
//
// Data extracted from the HellaSwag validation dataset (MIT license) https://github.com/rowanz/hellaswag/blob/master/data/hellaswag_val.jsonl
std::vector<enum ggml_type> include_types;
};
-const size_t HISTOGRAM_BUCKETS = 150;
-const double HISTOGRAM_RANGE = 0.03;
+constexpr size_t HISTOGRAM_BUCKETS = 150;
+constexpr double HISTOGRAM_RANGE = 0.03;
struct error_stats {
size_t num_samples;
uint64_t error_histogram[HISTOGRAM_BUCKETS];
};
-
-void quantize_stats_print_usage(int /*argc*/, char ** argv) {
+static void quantize_stats_print_usage(int /*argc*/, char ** argv) {
quantize_stats_params params;
fprintf(stderr, "usage: %s [options]\n", argv[0]);
fprintf(stderr, "\n");
}
// Check if a layer is included/excluded by command line
-bool layer_included(const quantize_stats_params & params, const std::string & layer) {
+static bool layer_included(const quantize_stats_params & params, const std::string & layer) {
for (const auto& excluded : params.exclude_layers) {
if (std::regex_search(layer, std::regex(excluded))) {
return false;
}
// Update error statistics given vectors with the before/after result of quantization
-void update_error_stats(int64_t nelements, const float * input, const float * output, error_stats & stats) {
+static void update_error_stats(int64_t nelements, const float * input, const float * output, error_stats & stats) {
for (int64_t i = 0; i < nelements; i++) {
double diff = input[i] - output[i];
stats.total_error += diff * diff;
stats.num_samples += nelements;
}
-void combine_error_stats(error_stats & into, const error_stats & from) {
+static void combine_error_stats(error_stats & into, const error_stats & from) {
into.num_samples += from.num_samples;
into.total_error += from.total_error;
if (from.max_error > into.max_error) into.max_error = from.max_error;
for (size_t i=0; i<HISTOGRAM_BUCKETS; ++i) into.error_histogram[i] += from.error_histogram[i];
}
-double find_quantile(const error_stats & stats, double quantile) {
+static double find_quantile(const error_stats & stats, double quantile) {
double sum = std::accumulate(std::begin(stats.error_histogram), std::end(stats.error_histogram), 0.0);
double accum = 0;
return INFINITY;
}
-void print_error_stats(const std::string & name, const error_stats & stats, bool print_histogram) {
+static void print_error_stats(const std::string & name, const error_stats & stats, bool print_histogram) {
double rmse = sqrt(stats.total_error / (double) stats.num_samples);
double median = find_quantile(stats, .5);
double pct95 = find_quantile(stats, .95);
tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
}
-void test_roundtrip_on_chunk(
- const ggml_tensor * layer,
- int64_t offset,
- int64_t chunk_size,
- const ggml_type_traits_t & qfns,
- bool use_reference,
- float * input_scratch,
- char * quantized_scratch,
- float * output_scratch,
- error_stats & stats) {
-
+static void test_roundtrip_on_chunk(
+ const ggml_tensor * layer, int64_t offset, int64_t chunk_size, const ggml_type_traits_t & qfns, bool use_reference,
+ float * input_scratch, char * quantized_scratch, float * output_scratch, error_stats & stats
+) {
if (layer->type == GGML_TYPE_F16) {
for (int i = 0; i < chunk_size; i++) {
input_scratch[i] = ggml_get_f32_1d(layer, i + offset);
// Run quantization function for a single layer and update error stats
-void test_roundtrip_on_layer(
- std::string & name,
- bool print_layer_stats,
- const ggml_type_traits_t & qfns,
- bool use_reference,
- const ggml_tensor * layer,
- std::vector<float> & input_scratch,
- std::vector<char> & quantized_scratch,
- std::vector<float> & output_scratch,
- error_stats & total_error,
- int max_thread = 0) {
-
+static void test_roundtrip_on_layer(
+ std::string & name, bool print_layer_stats, const ggml_type_traits_t & qfns, bool use_reference,
+ const ggml_tensor * layer, std::vector<float> & input_scratch, std::vector<char> & quantized_scratch,
+ std::vector<float> & output_scratch, error_stats & total_error, int max_thread = 0
+) {
assert(tensor_is_contiguous(layer));
error_stats layer_error {};
uint64_t nelements = ggml_nelements(layer);
};
-bool try_parse_ftype(const std::string & ftype_str_in, llama_ftype & ftype, std::string & ftype_str_out) {
+static bool try_parse_ftype(const std::string & ftype_str_in, llama_ftype & ftype, std::string & ftype_str_out) {
std::string ftype_str;
for (auto ch : ftype_str_in) {
// usage:
// ./quantize [--allow-requantize] [--leave-output-tensor] models/llama/ggml-model.gguf [models/llama/ggml-model-quant.gguf] type [nthreads]
//
-void usage(const char * executable) {
+static void usage(const char * executable) {
printf("usage: %s [--help] [--allow-requantize] [--leave-output-tensor] model-f32.gguf [model-quant.gguf] type [nthreads]\n\n", executable);
printf(" --allow-requantize: Allows requantizing tensors that have already been quantized. Warning: This can severely reduce quality compared to quantizing from 16bit or 32bit\n");
printf(" --leave-output-tensor: Will leave output.weight un(re)quantized. Increases model size but may also increase quality, especially when requantizing\n");
return res;
}
-static json format_partial_response(llama_server_context &llama, const std::string &content, const std::vector<completion_token_output> &probs)
-{
+static json format_partial_response(
+ llama_server_context &llama, const std::string &content, const std::vector<completion_token_output> &probs
+) {
json res = json{
{"content", content},
{"stop", false},
});
}
-bool is_at_eob(llama_server_context & server_context, const llama_token * tokens, const size_t n_tokens) {
+static bool is_at_eob(llama_server_context &server_context, const llama_token *tokens, const size_t n_tokens) {
return n_tokens && tokens[n_tokens-1] == llama_token_eos(server_context.ctx);
}
// * When all beams converge to a common prefix, they are made available in beams_state.beams[0].
// This is also called when the stop condition is met.
// Collect tokens into std::vector<llama_token> response which is pointed to by callback_data.
-void beam_search_callback(void * callback_data, llama_beams_state beams_state) {
+static void beam_search_callback(void *callback_data, llama_beams_state beams_state) {
auto & llama = *static_cast<llama_server_context*>(callback_data);
// Mark beams as EOS as needed.
for (size_t i = 0 ; i < beams_state.n_beams ; ++i) {
std::string operator()(const completion_token_output & cto) const { return (*this)(cto.tok); }
};
-void append_to_generated_text_from_generated_token_probs(llama_server_context & llama) {
+static void append_to_generated_text_from_generated_token_probs(llama_server_context &llama)
+{
auto & gtps = llama.generated_token_probs;
auto translator = token_translator{llama.ctx};
auto add_strlen = [=](size_t sum, const completion_token_output & cto) { return sum + translator(cto).size(); };
+#define LLAMA_API_INTERNAL
#include "llama.h"
#include "ggml.h"
return lookup[highbits];
}
-void replace_all(std::string & s, const std::string & search, const std::string & replace) {
+static void replace_all(std::string & s, const std::string & search, const std::string & replace) {
std::string result;
for (size_t pos = 0; ; pos += search.length()) {
auto new_pos = s.find(search, pos);
// load LLaMA models
//
-std::string llama_model_ftype_name(enum llama_ftype ftype) {
+static std::string llama_model_ftype_name(enum llama_ftype ftype) {
if (ftype & LLAMA_FTYPE_GUESSED) {
return llama_model_ftype_name((enum llama_ftype) (ftype & ~LLAMA_FTYPE_GUESSED)) + " (guessed)";
}
// Decodes a UTF-8 string which may end in an incomplete sequence. Adds a terminating 0 for use as
// pointer. If an invalid sequence is encountered, returns `llama_partial_utf8.n_remain == -1`.
-std::pair<std::vector<uint32_t>, llama_partial_utf8> decode_utf8(
+static std::pair<std::vector<uint32_t>, llama_partial_utf8> decode_utf8(
const char * src,
llama_partial_utf8 partial_start) {
static const int lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 2, 2, 3, 4 };
}
// TODO: after the GGUF PR, this likely won't work and needs to be updated
-int llama_apply_lora_from_file_internal(const struct llama_model & model, const char * path_lora, const char * path_base_model, int n_threads) {
+static int llama_apply_lora_from_file_internal(
+ const struct llama_model & model, const char * path_lora, const char * path_base_model, int n_threads
+) {
LLAMA_LOG_INFO("%s: applying lora adapter from '%s' - please wait ...\n", __func__, path_lora);
const int64_t t_start_lora_us = ggml_time_us();
return ctx;
}
-struct llama_context * llama_init_from_file(
+static struct llama_context * llama_init_from_file(
const char * path_model,
struct llama_context_params params) {
struct llama_model * model = llama_load_model_from_file(path_model, params);
* llama_copy_state_data(ctx, &data_ctx);
*
*/
-void llama_copy_state_data_internal(struct llama_context * ctx, llama_data_context * data_ctx) {
+static void llama_copy_state_data_internal(struct llama_context * ctx, llama_data_context * data_ctx) {
// copy rng
{
std::stringstream rng_ss;
}
// For internal test use
-const std::vector<std::pair<std::string, struct ggml_tensor *>>& llama_internal_get_tensor_map(struct llama_context * ctx) {
+const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal_get_tensor_map(
+ struct llama_context * ctx
+) {
return ctx->model.tensors_by_name;
}
struct ggml_tensor;
-const std::vector<std::pair<std::string, struct ggml_tensor *>>& llama_internal_get_tensor_map(struct llama_context * ctx);
+const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal_get_tensor_map(
+ struct llama_context * ctx
+);
#endif // LLAMA_API_INTERNAL
constexpr int kVecSize = 1 << 18;
-float drawFromGaussianPdf(std::mt19937& rndm) {
+static float drawFromGaussianPdf(std::mt19937& rndm) {
constexpr double kScale = 1./(1. + std::mt19937::max());
constexpr double kTwoPiTimesScale = 6.28318530717958647692*kScale;
static float lastX;
haveX = true;
return r*cos(phi);
}
-void fillRandomGaussianFloats(std::vector<float>& values, std::mt19937& rndm, float mean = 0) {
+
+static void fillRandomGaussianFloats(std::vector<float>& values, std::mt19937& rndm, float mean = 0) {
for (auto& v : values) v = mean + drawFromGaussianPdf(rndm);
}
#define GGML_PRINT(...) printf(__VA_ARGS__)
-float frand(void) {
+static float frand(void) {
return (float)rand()/(float)RAND_MAX;
}
-int irand(int n) {
+static int irand(int n) {
return rand()%n;
}
-void get_random_dims(int64_t * dims, int ndims) {
+static void get_random_dims(int64_t * dims, int ndims) {
dims[0] = dims[1] = dims[2] = dims[3] = 1;
for (int i = 0; i < ndims; i++) {
}
}
-void get_random_dims_minmax(int64_t * dims, int ndims, int min, int max) {
+static void get_random_dims_minmax(int64_t * dims, int ndims, int min, int max) {
dims[0] = dims[1] = dims[2] = dims[3] = 1;
for (int i = 0; i < ndims; i++) {
}
-struct ggml_tensor * get_random_tensor(
- struct ggml_context * ctx0,
- int ndims,
- int64_t ne[],
- float fmin,
- float fmax) {
+static struct ggml_tensor * get_random_tensor(
+ struct ggml_context * ctx0, int ndims, int64_t ne[], float fmin, float fmax
+) {
struct ggml_tensor * result = ggml_new_tensor(ctx0, GGML_TYPE_F32, ndims, ne);
switch (ndims) {
return result;
}
-float get_element(const struct ggml_tensor * t, int idx) {
+static float get_element(const struct ggml_tensor * t, int idx) {
return ((float *)t->data)[idx];
}
-void set_element(struct ggml_tensor * t, int idx, float value) {
+static void set_element(struct ggml_tensor * t, int idx, float value) {
((float *)t->data)[idx] = value;
}
#pragma warning(disable: 4244 4267) // possible loss of data
#endif
-const float MAX_QUANTIZATION_REFERENCE_ERROR = 0.0001f;
-const float MAX_QUANTIZATION_TOTAL_ERROR = 0.002f;
-const float MAX_QUANTIZATION_TOTAL_ERROR_2BITS = 0.0075f;
-const float MAX_QUANTIZATION_TOTAL_ERROR_3BITS = 0.0040f;
-const float MAX_DOT_PRODUCT_ERROR = 0.02f;
+constexpr float MAX_QUANTIZATION_REFERENCE_ERROR = 0.0001f;
+constexpr float MAX_QUANTIZATION_TOTAL_ERROR = 0.002f;
+constexpr float MAX_QUANTIZATION_TOTAL_ERROR_2BITS = 0.0075f;
+constexpr float MAX_QUANTIZATION_TOTAL_ERROR_3BITS = 0.0040f;
+constexpr float MAX_DOT_PRODUCT_ERROR = 0.02f;
-const char* RESULT_STR[] = {"ok", "FAILED"};
+static const char* RESULT_STR[] = {"ok", "FAILED"};
// Generate synthetic data
-void generate_data(float offset, size_t n, float * dst) {
+static void generate_data(float offset, size_t n, float * dst) {
for (size_t i = 0; i < n; i++) {
dst[i] = 0.1 + 2*cosf(i + offset);
}
}
// Calculate RMSE between two float arrays
-float array_rmse(const float * a1, const float * a2, size_t n) {
+static float array_rmse(const float * a1, const float * a2, size_t n) {
double sum = 0;
for (size_t i = 0; i < n; i++) {
double diff = a1[i] - a2[i];
}
// Total quantization error on test data
-float total_quantization_error(ggml_type_traits_t & qfns, size_t test_size, const float * test_data) {
+static float total_quantization_error(ggml_type_traits_t & qfns, size_t test_size, const float * test_data) {
std::vector<uint8_t> tmp_q(2*test_size);
std::vector<float> tmp_out(test_size);
}
// Total quantization error on test data
-float reference_quantization_error(ggml_type_traits_t & qfns, size_t test_size, const float * test_data) {
+static float reference_quantization_error(ggml_type_traits_t & qfns, size_t test_size, const float * test_data) {
std::vector<uint8_t> tmp_q(2*test_size);
std::vector<float> tmp_out(test_size);
std::vector<float> tmp_out_ref(test_size);
return array_rmse(tmp_out.data(), tmp_out_ref.data(), test_size);
}
-float dot_product(const float * a1, const float * a2, size_t test_size) {
+static float dot_product(const float * a1, const float * a2, size_t test_size) {
double sum = 0;
for (size_t i = 0; i < test_size; i++) {
sum += a1[i] * a2[i];
}
// Total dot product error
-float dot_product_error(ggml_type_traits_t & qfns, size_t test_size, const float * test_data1, const float *test_data2) {
+static float dot_product_error(
+ ggml_type_traits_t & qfns, size_t test_size, const float * test_data1, const float *test_data2
+) {
std::vector<uint8_t> tmp_q1(2*test_size);
std::vector<uint8_t> tmp_q2(2*test_size);
// Generate synthetic data
-void generate_data(float offset, size_t n, float * dst) {
+static void generate_data(float offset, size_t n, float * dst) {
for (size_t i = 0; i < n; i++) {
dst[i] = 0.1 + 2*cosf(i + offset);
}
}
-float gigabytes_per_second(size_t bytes, int64_t usecs) {
+static float gigabytes_per_second(size_t bytes, int64_t usecs) {
return bytes / (float) usecs * 1000000 / (1024*1024*1024);
}
-void * align_with_offset(void * ptr, int offset) {
+static void * align_with_offset(void * ptr, int offset) {
size_t dummy_size = MAX_ALIGNMENT * 4;
return (char *) std::align(MAX_ALIGNMENT, MAX_ALIGNMENT, ptr, dummy_size) + offset;
}
-void benchmark_function(size_t size, size_t q_size, int64_t iterations, const std::function<size_t(void)> & function) {
+static void benchmark_function(size_t size, size_t q_size, int64_t iterations, const std::function<size_t(void)> & function) {
int64_t min_time_us = INT64_MAX;
int64_t total_time_us = 0;
int64_t min_time_cycles = INT64_MAX;
printf(" quantized throughput : %9.2f GB/s\n", gigabytes_per_second(q_size * iterations, total_time_us));
}
-void usage(char * argv[]) {
+static void usage(char * argv[]) {
printf("Benchmark quantization specific functions on synthetic data\n");
printf("\n");
printf("usage: %s [options]\n", argv[0]);
#include <vector>
#include <algorithm>
-void dump(const llama_token_data_array * candidates) {
+
+static void dump(const llama_token_data_array * candidates) {
for (size_t i = 0; i < candidates->size; i++) {
printf("%d: %f (%f)\n", candidates->data[i].id, candidates->data[i].p, candidates->data[i].logit);
}
#define DUMP(__candidates) do { printf("%s:%d (%s)\n", __FILE__, __LINE__, __func__); dump((__candidates)); printf("-\n"); } while(0)
-void test_top_k(const std::vector<float> & probs,
- const std::vector<float> & expected_probs,
- int k) {
+static void test_top_k(const std::vector<float> & probs, const std::vector<float> & expected_probs, int k) {
size_t n_vocab = probs.size();
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
}
-void test_top_p(const std::vector<float> & probs,
- const std::vector<float> & expected_probs,
- float p) {
-
+static void test_top_p(const std::vector<float> & probs, const std::vector<float> & expected_probs, float p) {
size_t n_vocab = probs.size();
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
}
-void test_tfs(const std::vector<float> & probs,
- const std::vector<float> & expected_probs,
- float z) {
+static void test_tfs(const std::vector<float> & probs, const std::vector<float> & expected_probs, float z) {
size_t n_vocab = probs.size();
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
}
-void test_typical(const std::vector<float> & probs,
- const std::vector<float> & expected_probs,
- float p) {
+static void test_typical(const std::vector<float> & probs, const std::vector<float> & expected_probs, float p) {
size_t n_vocab = probs.size();
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
}
-void test_repetition_penalty(
- const std::vector<float> & probs,
- const std::vector<llama_token> & last_tokens,
- const std::vector<float> & expected_probs,
- float penalty) {
+static void test_repetition_penalty(
+ const std::vector<float> & probs, const std::vector<llama_token> & last_tokens,
+ const std::vector<float> & expected_probs, float penalty
+) {
assert(probs.size() == expected_probs.size());
size_t n_vocab = probs.size();
}
-void test_frequency_presence_penalty(
- const std::vector<float> & probs,
- const std::vector<llama_token> & last_tokens,
- const std::vector<float> & expected_probs,
- float alpha_frequency, float alpha_presence) {
+static void test_frequency_presence_penalty(
+ const std::vector<float> & probs, const std::vector<llama_token> & last_tokens,
+ const std::vector<float> & expected_probs, float alpha_frequency, float alpha_presence
+) {
assert(probs.size() == expected_probs.size());
size_t n_vocab = probs.size();
typedef int codepoint;
-std::string codepoint_to_utf8(codepoint cp) {
+static std::string codepoint_to_utf8(codepoint cp) {
std::string result;
if (0x00 <= cp && cp <= 0x7f) {
result.push_back(cp);
overview / pkg / ggml / sources / llama.cpp / commitdiff