include llama.cpp since there were some edits
This commit is contained in:
zefie committed
1 parent
db7f28d087
commit
8c66990853
3918 files changed
+1640170
-1
No files matched your search
@@ -0,0 +1,213 @@
|
||||
#include "llama-impl.h"
|
||||
#include "llama-mmap.h"
|
||||
|
||||
#include "ggml-backend.h"
|
||||
#include "gguf.h"
|
||||
#include "llama.h"
|
||||
|
||||
#include <algorithm>
|
||||
#include <cinttypes>
|
||||
#include <climits>
|
||||
#include <cstdarg>
|
||||
#include <cstring>
|
||||
#include <vector>
|
||||
|
||||
struct llama_logger_state {
|
||||
ggml_log_callback log_callback = llama_log_callback_default;
|
||||
void * log_callback_user_data = nullptr;
|
||||
};
|
||||
|
||||
static llama_logger_state g_logger_state;
|
||||
|
||||
void llama_prefetch_rows(const ggml_tensor * tensor, const int32_t * rows, size_t n_rows) {
|
||||
if (!tensor || !tensor->data || !tensor->buffer || !ggml_backend_buffer_is_host(tensor->buffer) || n_rows == 0) {
|
||||
return;
|
||||
}
|
||||
|
||||
GGML_ASSERT(ggml_is_matrix(tensor));
|
||||
|
||||
const size_t row_bytes = ggml_row_size(tensor->type, tensor->ne[0]);
|
||||
const auto * base = (const char *) tensor->data;
|
||||
|
||||
std::vector<llama_memory_range> mr;
|
||||
mr.reserve(n_rows);
|
||||
for (size_t i = 0; i < n_rows; ++i) {
|
||||
GGML_ASSERT(rows[i] >= 0 && rows[i] < tensor->ne[1]);
|
||||
mr.push_back({ base + (size_t) rows[i] * tensor->nb[1], row_bytes });
|
||||
}
|
||||
|
||||
llama_prefetch(std::move(mr));
|
||||
}
|
||||
|
||||
time_meas::time_meas(int64_t & t_acc, bool disable) : t_start_us(disable ? -1 : ggml_time_us()), t_acc(t_acc) {}
|
||||
|
||||
time_meas::~time_meas() {
|
||||
if (t_start_us >= 0) {
|
||||
t_acc += ggml_time_us() - t_start_us;
|
||||
}
|
||||
}
|
||||
|
||||
void llama_log_get(ggml_log_callback * log_callback, void ** user_data) {
|
||||
ggml_log_get(log_callback, user_data);
|
||||
}
|
||||
|
||||
void llama_log_set(ggml_log_callback log_callback, void * user_data) {
|
||||
ggml_log_set(log_callback, user_data);
|
||||
g_logger_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
|
||||
g_logger_state.log_callback_user_data = user_data;
|
||||
}
|
||||
|
||||
static void llama_log_internal_v(ggml_log_level level, const char * format, va_list args) {
|
||||
va_list args_copy;
|
||||
va_copy(args_copy, args);
|
||||
char buffer[128];
|
||||
int len = vsnprintf(buffer, 128, format, args);
|
||||
if (len < 128) {
|
||||
g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
|
||||
} else {
|
||||
char * buffer2 = new char[len + 1];
|
||||
vsnprintf(buffer2, len + 1, format, args_copy);
|
||||
buffer2[len] = 0;
|
||||
g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
|
||||
delete[] buffer2;
|
||||
}
|
||||
va_end(args_copy);
|
||||
}
|
||||
|
||||
void llama_log_internal(ggml_log_level level, const char * format, ...) {
|
||||
va_list args;
|
||||
va_start(args, format);
|
||||
llama_log_internal_v(level, format, args);
|
||||
va_end(args);
|
||||
}
|
||||
|
||||
void llama_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
|
||||
(void) level;
|
||||
(void) user_data;
|
||||
fputs(text, stderr);
|
||||
fflush(stderr);
|
||||
}
|
||||
|
||||
void llama_clear_tensor_data(ggml_tensor * t, size_t offset, size_t size) {
|
||||
static const std::vector<uint8_t> zeros(1024*1024, 0);
|
||||
|
||||
// not all backend buffers implement ggml_backend_tensor_memset(), so write zeros instead
|
||||
// TODO: make this a generic fallback in `ggml_backend_tensor_memset` when `set_tensor` is available
|
||||
for (size_t ofs = 0; ofs < size; ofs += zeros.size()) {
|
||||
ggml_backend_tensor_set(t, zeros.data(), offset + ofs, std::min(size - ofs, zeros.size()));
|
||||
}
|
||||
}
|
||||
|
||||
void replace_all(std::string & s, const std::string & search, const std::string & replace) {
|
||||
if (search.empty()) {
|
||||
return;
|
||||
}
|
||||
std::string builder;
|
||||
builder.reserve(s.length());
|
||||
size_t pos = 0;
|
||||
size_t last_pos = 0;
|
||||
while ((pos = s.find(search, last_pos)) != std::string::npos) {
|
||||
builder.append(s, last_pos, pos - last_pos);
|
||||
builder.append(replace);
|
||||
last_pos = pos + search.length();
|
||||
}
|
||||
builder.append(s, last_pos, std::string::npos);
|
||||
s = std::move(builder);
|
||||
}
|
||||
|
||||
std::string format(const char * fmt, ...) {
|
||||
va_list ap;
|
||||
va_list ap2;
|
||||
va_start(ap, fmt);
|
||||
va_copy(ap2, ap);
|
||||
int size = vsnprintf(NULL, 0, fmt, ap);
|
||||
GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT
|
||||
std::vector<char> buf(size + 1);
|
||||
int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2);
|
||||
GGML_ASSERT(size2 == size);
|
||||
va_end(ap2);
|
||||
va_end(ap);
|
||||
return std::string(buf.data(), size);
|
||||
}
|
||||
|
||||
std::string llama_format_tensor_shape(const std::vector<int64_t> & ne) {
|
||||
char buf[256];
|
||||
snprintf(buf, sizeof(buf), "%6" PRId64, ne.at(0));
|
||||
for (size_t i = 1; i < ne.size(); i++) {
|
||||
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), ", %6" PRId64, ne.at(i));
|
||||
}
|
||||
return buf;
|
||||
}
|
||||
|
||||
std::string llama_format_tensor_shape(const struct ggml_tensor * t) {
|
||||
char buf[256];
|
||||
snprintf(buf, sizeof(buf), "%6" PRId64, t->ne[0]);
|
||||
for (int i = 1; i < GGML_MAX_DIMS; i++) {
|
||||
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), ", %6" PRId64, t->ne[i]);
|
||||
}
|
||||
return buf;
|
||||
}
|
||||
|
||||
static std::string gguf_data_to_str(enum gguf_type type, const void * data, size_t i) {
|
||||
switch (type) {
|
||||
case GGUF_TYPE_UINT8: return std::to_string(((const uint8_t *)data)[i]);
|
||||
case GGUF_TYPE_INT8: return std::to_string(((const int8_t *)data)[i]);
|
||||
case GGUF_TYPE_UINT16: return std::to_string(((const uint16_t *)data)[i]);
|
||||
case GGUF_TYPE_INT16: return std::to_string(((const int16_t *)data)[i]);
|
||||
case GGUF_TYPE_UINT32: return std::to_string(((const uint32_t *)data)[i]);
|
||||
case GGUF_TYPE_INT32: return std::to_string(((const int32_t *)data)[i]);
|
||||
case GGUF_TYPE_UINT64: return std::to_string(((const uint64_t *)data)[i]);
|
||||
case GGUF_TYPE_INT64: return std::to_string(((const int64_t *)data)[i]);
|
||||
case GGUF_TYPE_FLOAT32: return std::to_string(((const float *)data)[i]);
|
||||
case GGUF_TYPE_FLOAT64: return std::to_string(((const double *)data)[i]);
|
||||
case GGUF_TYPE_BOOL: return ((const int8_t *)data)[i] != 0 ? "true" : "false";
|
||||
default: return format("unknown type %d", type);
|
||||
}
|
||||
}
|
||||
|
||||
std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i, size_t max_len) {
|
||||
const enum gguf_type type = gguf_get_kv_type(ctx_gguf, i);
|
||||
|
||||
switch (type) {
|
||||
case GGUF_TYPE_STRING:
|
||||
return gguf_get_val_str(ctx_gguf, i);
|
||||
case GGUF_TYPE_ARRAY:
|
||||
{
|
||||
const enum gguf_type arr_type = gguf_get_arr_type(ctx_gguf, i);
|
||||
const size_t arr_n = gguf_get_arr_n(ctx_gguf, i);
|
||||
const void * data = arr_type == GGUF_TYPE_STRING ? nullptr : gguf_get_arr_data(ctx_gguf, i);
|
||||
std::string res = "[";
|
||||
size_t j = 0;
|
||||
for (; j < arr_n && res.size() <= max_len; j++) {
|
||||
if (j > 0) {
|
||||
res += ", ";
|
||||
}
|
||||
if (arr_type == GGUF_TYPE_STRING) {
|
||||
const char * s = gguf_get_arr_str(ctx_gguf, i, j);
|
||||
const size_t need = res.size() + 2 * strlen(s) + 2;
|
||||
if (res.capacity() < need) {
|
||||
res.reserve(std::max(need, 2 * res.capacity()));
|
||||
}
|
||||
res += '"';
|
||||
for (; *s; s++) {
|
||||
if (*s == '\\' || *s == '"') {
|
||||
res += '\\';
|
||||
}
|
||||
res += *s;
|
||||
}
|
||||
res += '"';
|
||||
} else if (arr_type == GGUF_TYPE_ARRAY) {
|
||||
res += "???";
|
||||
} else {
|
||||
res += gguf_data_to_str(arr_type, data, j);
|
||||
}
|
||||
}
|
||||
if (j == arr_n) {
|
||||
res += "]";
|
||||
}
|
||||
return res;
|
||||
}
|
||||
default:
|
||||
return gguf_data_to_str(type, gguf_get_val_data(ctx_gguf, i), 0);
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user