zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-10-31 16:21:27 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Remove unnecessary string copies (#891)

Browse Source 
1. Use string_view instead of string when there is no need for copy.
2. Otherwise move string when possible.
This commit is contained in:
Cheng
2024-03-29 05:14:59 +09:00
committed by GitHub
parent 45f636e759
commit 46caf0bef0
11 changed files with 36 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
mlx/io/gguf.cpp
View File

@@ -230,8 +230,8 @@ std::unordered_map<std::string, array> load_arrays(gguf_ctx* ctx) {
return array_map;
}
GGUFLoad load_gguf(const std::string& file, StreamOrDevice s) {
gguf_ctx* ctx = gguf_open(file.c_str());
GGUFLoad load_gguf(std::string_view file, StreamOrDevice s) {
gguf_ctx* ctx = gguf_open(file.data());
if (!ctx) {
throw std::runtime_error("[load_gguf] gguf_init failed");
}

12
mlx/io/gguf_quants.cpp
View File

@@ -105,7 +105,7 @@ void gguf_load_quantized(
weights_per_byte = 1;
}
std::string name = std::string(tensor.name, tensor.namelen);
std::string name(tensor.name, tensor.namelen);
std::vector<int> shape = get_shape(tensor);
const uint64_t weights_per_block = 32;
if (shape[shape.size() - 1] % weights_per_block != 0) {
@@ -136,9 +136,9 @@ void gguf_load_quantized(
extract_q8_0_data(tensor, weights, scales, biases);
}
a.insert({name, weights});
a.emplace(std::move(name), std::move(weights));
auto check_insert = [](auto inserted) {
auto check_insert = [](const auto& inserted) {
if (!inserted.second) {
std::ostringstream msg;
msg << "[load_gguf] Duplicate parameter name " << inserted.first->second
@@ -147,11 +147,11 @@ void gguf_load_quantized(
}
};
const std::string weight_suffix = ".weight";
constexpr std::string_view weight_suffix = ".weight";
const std::string name_prefix =
name.substr(0, name.length() - weight_suffix.length());
check_insert(a.insert({name_prefix + ".scales", scales}));
check_insert(a.insert({name_prefix + ".biases", biases}));
check_insert(a.emplace(name_prefix + ".scales", std::move(scales)));
check_insert(a.emplace(name_prefix + ".biases", std::move(biases)));
}
} // namespace mlx::core

11
mlx/io/load.cpp
View File

@@ -114,16 +114,13 @@ void save(std::shared_ptr<io::Writer> out_stream, array a) {
}
/** Save array to file in .npy format */
void save(const std::string& file_, array a) {
// Open and check file
std::string file = file_;
void save(std::string file, array a) {
// Add .npy to file name if it is not there
if (file.length() < 4 || file.substr(file.length() - 4, 4) != ".npy")
file += ".npy";
// Serialize array
save(std::make_shared<io::FileWriter>(file), a);
save(std::make_shared<io::FileWriter>(std::move(file)), a);
}
/** Load array from reader in .npy format */
@@ -227,8 +224,8 @@ array load(std::shared_ptr<io::Reader> in_stream, StreamOrDevice s) {
}
/** Load array from file in .npy format */
array load(const std::string& file, StreamOrDevice s) {
return load(std::make_shared<io::FileReader>(file), s);
array load(std::string file, StreamOrDevice s) {
return load(std::make_shared<io::FileReader>(std::move(file)), s);
}
} // namespace mlx::core

19
mlx/io/safetensor.cpp
View File

@@ -60,7 +60,7 @@ std::string dtype_to_safetensor_str(Dtype t) {
}
}
Dtype dtype_from_safetensor_str(std::string str) {
Dtype dtype_from_safetensor_str(std::string_view str) {
if (str == ST_F32) {
return float32;
} else if (str == ST_F16) {
@@ -88,7 +88,8 @@ Dtype dtype_from_safetensor_str(std::string str) {
} else if (str == ST_C64) {
return complex64;
} else {
throw std::runtime_error("[safetensor] unsupported dtype " + str);
throw std::runtime_error(
"[safetensor] unsupported dtype " + std::string(str));
}
}
@@ -129,9 +130,9 @@ SafetensorsLoad load_safetensors(
}
continue;
}
std::string dtype = item.value().at("dtype");
std::vector<int> shape = item.value().at("shape");
std::vector<size_t> data_offsets = item.value().at("data_offsets");
const std::string& dtype = item.value().at("dtype");
const std::vector<int>& shape = item.value().at("shape");
const std::vector<size_t>& data_offsets = item.value().at("data_offsets");
Dtype type = dtype_from_safetensor_str(dtype);
auto loaded_array = array(
shape,
@@ -207,19 +208,17 @@ void save_safetensors(
}
void save_safetensors(
const std::string& file_,
std::string file,
std::unordered_map<std::string, array> a,
std::unordered_map<std::string, std::string> metadata /* = {} */) {
// Open and check file
std::string file = file_;
// Add .safetensors to file name if it is not there
if (file.length() < 12 ||
file.substr(file.length() - 12, 12) != ".safetensors")
file += ".safetensors";
// Serialize array
save_safetensors(std::make_shared<io::FileWriter>(file), a, metadata);
save_safetensors(
std::make_shared<io::FileWriter>(std::move(file)), a, metadata);
}
} // namespace mlx::core