2023-04-03 01:50:13 +00:00
|
|
|
import sys
|
|
|
|
import os
|
2023-03-23 09:33:06 +00:00
|
|
|
import ctypes
|
2023-04-02 17:33:49 +00:00
|
|
|
from ctypes import c_int, c_float, c_char_p, c_void_p, c_bool, POINTER, Structure, Array, c_uint8, c_size_t
|
2023-03-23 09:33:06 +00:00
|
|
|
import pathlib
|
|
|
|
|
|
|
|
# Load the library
|
2023-04-03 17:06:50 +00:00
|
|
|
def _load_shared_library(lib_base_name):
|
2023-04-03 01:50:13 +00:00
|
|
|
# Determine the file extension based on the platform
|
|
|
|
if sys.platform.startswith("linux"):
|
|
|
|
lib_ext = ".so"
|
|
|
|
elif sys.platform == "darwin":
|
2023-04-08 06:45:21 +00:00
|
|
|
lib_ext = ".so"
|
2023-04-03 01:50:13 +00:00
|
|
|
elif sys.platform == "win32":
|
|
|
|
lib_ext = ".dll"
|
|
|
|
else:
|
|
|
|
raise RuntimeError("Unsupported platform")
|
|
|
|
|
|
|
|
# Construct the paths to the possible shared library names
|
|
|
|
_base_path = pathlib.Path(__file__).parent.resolve()
|
|
|
|
# Searching for the library in the current directory under the name "libllama" (default name
|
|
|
|
# for llamacpp) and "llama" (default name for this repo)
|
|
|
|
_lib_paths = [
|
|
|
|
_base_path / f"lib{lib_base_name}{lib_ext}",
|
|
|
|
_base_path / f"{lib_base_name}{lib_ext}"
|
|
|
|
]
|
|
|
|
|
2023-04-10 15:00:35 +00:00
|
|
|
if ("LLAMA_LIB" in os.environ):
|
2023-04-10 15:06:58 +00:00
|
|
|
lib_base_name = os.environ["LLAMA_LIB"]
|
2023-04-10 15:12:25 +00:00
|
|
|
_lib = pathlib.Path(lib_base_name)
|
|
|
|
_base_path = _lib.parent.resolve()
|
|
|
|
_lib_paths = [_lib.resolve()]
|
2023-04-10 15:00:35 +00:00
|
|
|
|
2023-04-03 01:50:13 +00:00
|
|
|
# Add the library directory to the DLL search path on Windows (if needed)
|
|
|
|
if sys.platform == "win32" and sys.version_info >= (3, 8):
|
|
|
|
os.add_dll_directory(str(_base_path))
|
|
|
|
|
|
|
|
# Try to load the shared library, handling potential errors
|
|
|
|
for _lib_path in _lib_paths:
|
|
|
|
if _lib_path.exists():
|
|
|
|
try:
|
|
|
|
return ctypes.CDLL(str(_lib_path))
|
|
|
|
except Exception as e:
|
|
|
|
raise RuntimeError(f"Failed to load shared library '{_lib_path}': {e}")
|
|
|
|
|
|
|
|
raise FileNotFoundError(f"Shared library with base name '{lib_base_name}' not found")
|
|
|
|
|
|
|
|
# Specify the base name of the shared library to load
|
2023-04-03 17:06:50 +00:00
|
|
|
_lib_base_name = "llama"
|
2023-04-03 01:50:13 +00:00
|
|
|
|
|
|
|
# Load the library
|
2023-04-03 17:06:50 +00:00
|
|
|
_lib = _load_shared_library(_lib_base_name)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
|
|
|
# C types
|
2023-03-24 18:58:42 +00:00
|
|
|
llama_context_p = c_void_p
|
|
|
|
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
llama_token = c_int
|
|
|
|
llama_token_p = POINTER(llama_token)
|
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
class llama_token_data(Structure):
|
|
|
|
_fields_ = [
|
2023-03-24 18:35:41 +00:00
|
|
|
("id", llama_token), # token id
|
|
|
|
("p", c_float), # probability of the token
|
|
|
|
("plog", c_float), # log probability of the token
|
2023-03-23 09:33:06 +00:00
|
|
|
]
|
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
llama_token_data_p = POINTER(llama_token_data)
|
|
|
|
|
2023-03-29 01:10:23 +00:00
|
|
|
llama_progress_callback = ctypes.CFUNCTYPE(None, c_float, c_void_p)
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-25 20:26:03 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
class llama_context_params(Structure):
|
|
|
|
_fields_ = [
|
2023-03-24 18:35:41 +00:00
|
|
|
("n_ctx", c_int), # text context
|
|
|
|
("n_parts", c_int), # -1 for default
|
|
|
|
("seed", c_int), # RNG seed, 0 for random
|
|
|
|
("f16_kv", c_bool), # use fp16 for KV cache
|
|
|
|
(
|
|
|
|
"logits_all",
|
|
|
|
c_bool,
|
|
|
|
), # the llama_eval() call computes all logits, not just the last one
|
|
|
|
("vocab_only", c_bool), # only load the vocabulary, no weights
|
2023-04-10 02:01:33 +00:00
|
|
|
("use_mmap", c_bool), # use mmap if possible
|
2023-03-24 18:58:42 +00:00
|
|
|
("use_mlock", c_bool), # force system to keep model in RAM
|
|
|
|
("embedding", c_bool), # embedding mode only
|
2023-03-25 16:12:09 +00:00
|
|
|
# called with a progress value between 0 and 1, pass NULL to disable
|
|
|
|
("progress_callback", llama_progress_callback),
|
|
|
|
# context pointer passed to the progress callback
|
|
|
|
("progress_callback_user_data", c_void_p),
|
2023-03-23 09:33:06 +00:00
|
|
|
]
|
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
llama_context_params_p = POINTER(llama_context_params)
|
|
|
|
|
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Functions
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_context_default_params() -> llama_context_params:
|
2023-03-29 01:10:23 +00:00
|
|
|
return _lib.llama_context_default_params()
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_context_default_params.argtypes = []
|
|
|
|
_lib.llama_context_default_params.restype = llama_context_params
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-04-10 02:01:33 +00:00
|
|
|
def llama_mmap_supported() -> c_bool:
|
|
|
|
return _lib.llama_mmap_supported()
|
|
|
|
|
|
|
|
_lib.llama_mmap_supported.argtypes = []
|
|
|
|
_lib.llama_mmap_supported.restype = c_bool
|
|
|
|
|
|
|
|
def llama_mlock_supported() -> c_bool:
|
|
|
|
return _lib.llama_mlock_supported()
|
|
|
|
|
|
|
|
_lib.llama_mlock_supported.argtypes = []
|
|
|
|
_lib.llama_mlock_supported.restype = c_bool
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Various functions for loading a ggml llama model.
|
|
|
|
# Allocate (almost) all memory needed for the model.
|
|
|
|
# Return NULL on failure
|
2023-03-24 18:35:41 +00:00
|
|
|
def llama_init_from_file(
|
|
|
|
path_model: bytes, params: llama_context_params
|
|
|
|
) -> llama_context_p:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_init_from_file(path_model, params)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_init_from_file.argtypes = [c_char_p, llama_context_params]
|
|
|
|
_lib.llama_init_from_file.restype = llama_context_p
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Frees all allocated memory
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_free(ctx: llama_context_p):
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_free(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_free.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_free.restype = None
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# TODO: not great API - very likely to change
|
|
|
|
# Returns 0 on success
|
2023-03-24 18:35:41 +00:00
|
|
|
def llama_model_quantize(
|
2023-04-05 02:36:59 +00:00
|
|
|
fname_inp: bytes, fname_out: bytes, itype: c_int
|
2023-03-24 18:35:41 +00:00
|
|
|
) -> c_int:
|
2023-04-05 02:36:59 +00:00
|
|
|
return _lib.llama_model_quantize(fname_inp, fname_out, itype)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-04-05 02:36:59 +00:00
|
|
|
_lib.llama_model_quantize.argtypes = [c_char_p, c_char_p, c_int]
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_model_quantize.restype = c_int
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-04-02 17:33:49 +00:00
|
|
|
# Returns the KV cache that will contain the context for the
|
|
|
|
# ongoing prediction with the model.
|
|
|
|
def llama_get_kv_cache(ctx: llama_context_p):
|
|
|
|
return _lib.llama_get_kv_cache(ctx)
|
|
|
|
|
|
|
|
_lib.llama_get_kv_cache.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_get_kv_cache.restype = POINTER(c_uint8)
|
|
|
|
|
|
|
|
# Returns the size of the KV cache
|
|
|
|
def llama_get_kv_cache_size(ctx: llama_context_p) -> c_size_t:
|
|
|
|
return _lib.llama_get_kv_cache_size(ctx)
|
|
|
|
|
|
|
|
_lib.llama_get_kv_cache_size.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_get_kv_cache_size.restype = c_size_t
|
|
|
|
|
|
|
|
# Returns the number of tokens in the KV cache
|
|
|
|
def llama_get_kv_cache_token_count(ctx: llama_context_p) -> c_int:
|
|
|
|
return _lib.llama_get_kv_cache_token_count(ctx)
|
|
|
|
|
|
|
|
_lib.llama_get_kv_cache_token_count.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_get_kv_cache_token_count.restype = c_int
|
|
|
|
|
|
|
|
|
|
|
|
# Sets the KV cache containing the current context for the model
|
|
|
|
def llama_set_kv_cache(ctx: llama_context_p, kv_cache, n_size: c_size_t, n_token_count: c_int):
|
|
|
|
return _lib.llama_set_kv_cache(ctx, kv_cache, n_size, n_token_count)
|
|
|
|
|
|
|
|
_lib.llama_set_kv_cache.argtypes = [llama_context_p, POINTER(c_uint8), c_size_t, c_int]
|
|
|
|
_lib.llama_set_kv_cache.restype = None
|
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Run the llama inference to obtain the logits and probabilities for the next token.
|
|
|
|
# tokens + n_tokens is the provided batch of new tokens to process
|
|
|
|
# n_past is the number of tokens to use from previous eval calls
|
|
|
|
# Returns 0 on success
|
2023-03-24 18:35:41 +00:00
|
|
|
def llama_eval(
|
|
|
|
ctx: llama_context_p,
|
2023-03-31 07:20:15 +00:00
|
|
|
tokens, # type: Array[llama_token]
|
2023-03-24 18:35:41 +00:00
|
|
|
n_tokens: c_int,
|
|
|
|
n_past: c_int,
|
|
|
|
n_threads: c_int,
|
|
|
|
) -> c_int:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_eval(ctx, tokens, n_tokens, n_past, n_threads)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_eval.argtypes = [llama_context_p, llama_token_p, c_int, c_int, c_int]
|
|
|
|
_lib.llama_eval.restype = c_int
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
# Convert the provided text into tokens.
|
|
|
|
# The tokens pointer must be large enough to hold the resulting tokens.
|
|
|
|
# Returns the number of tokens on success, no more than n_max_tokens
|
|
|
|
# Returns a negative number on failure - the number of tokens that would have been returned
|
|
|
|
# TODO: not sure if correct
|
2023-03-24 18:35:41 +00:00
|
|
|
def llama_tokenize(
|
|
|
|
ctx: llama_context_p,
|
|
|
|
text: bytes,
|
2023-03-31 07:20:15 +00:00
|
|
|
tokens, # type: Array[llama_token]
|
2023-03-24 18:35:41 +00:00
|
|
|
n_max_tokens: c_int,
|
|
|
|
add_bos: c_bool,
|
|
|
|
) -> c_int:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_tokenize(ctx, text, tokens, n_max_tokens, add_bos)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_tokenize.argtypes = [llama_context_p, c_char_p, llama_token_p, c_int, c_bool]
|
|
|
|
_lib.llama_tokenize.restype = c_int
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_n_vocab(ctx: llama_context_p) -> c_int:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_n_vocab(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_n_vocab.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_n_vocab.restype = c_int
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_n_ctx(ctx: llama_context_p) -> c_int:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_n_ctx(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_n_ctx.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_n_ctx.restype = c_int
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-25 20:26:03 +00:00
|
|
|
def llama_n_embd(ctx: llama_context_p) -> c_int:
|
2023-04-08 19:05:33 +00:00
|
|
|
return _lib.llama_n_embd(ctx)
|
2023-03-25 20:26:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
_lib.llama_n_embd.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_n_embd.restype = c_int
|
|
|
|
|
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Token logits obtained from the last call to llama_eval()
|
|
|
|
# The logits for the last token are stored in the last row
|
|
|
|
# Can be mutated in order to change the probabilities of the next token
|
|
|
|
# Rows: n_tokens
|
|
|
|
# Cols: n_vocab
|
2023-03-31 07:20:15 +00:00
|
|
|
def llama_get_logits(ctx: llama_context_p):
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_get_logits(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_get_logits.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_get_logits.restype = POINTER(c_float)
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Get the embeddings for the input
|
|
|
|
# shape: [n_embd] (1-dimensional)
|
2023-03-31 07:20:15 +00:00
|
|
|
def llama_get_embeddings(ctx: llama_context_p):
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_get_embeddings(ctx)
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_get_embeddings.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_get_embeddings.restype = POINTER(c_float)
|
2023-03-24 18:58:42 +00:00
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-24 18:58:42 +00:00
|
|
|
# Token Id -> String. Uses the vocabulary in the provided context
|
2023-03-31 07:25:12 +00:00
|
|
|
def llama_token_to_str(ctx: llama_context_p, token: llama_token) -> bytes:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_token_to_str(ctx, token)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_token_to_str.argtypes = [llama_context_p, llama_token]
|
|
|
|
_lib.llama_token_to_str.restype = c_char_p
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
# Special tokens
|
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_token_bos() -> llama_token:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_token_bos()
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_token_bos.argtypes = []
|
|
|
|
_lib.llama_token_bos.restype = llama_token
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_token_eos() -> llama_token:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_token_eos()
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_token_eos.argtypes = []
|
|
|
|
_lib.llama_token_eos.restype = llama_token
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
# TODO: improve the last_n_tokens interface ?
|
2023-03-24 18:35:41 +00:00
|
|
|
def llama_sample_top_p_top_k(
|
|
|
|
ctx: llama_context_p,
|
2023-03-31 07:20:15 +00:00
|
|
|
last_n_tokens_data, # type: Array[llama_token]
|
2023-03-24 18:35:41 +00:00
|
|
|
last_n_tokens_size: c_int,
|
|
|
|
top_k: c_int,
|
2023-03-29 01:10:23 +00:00
|
|
|
top_p: c_float,
|
|
|
|
temp: c_float,
|
|
|
|
repeat_penalty: c_float,
|
2023-03-24 18:35:41 +00:00
|
|
|
) -> llama_token:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_sample_top_p_top_k(
|
2023-03-24 18:35:41 +00:00
|
|
|
ctx, last_n_tokens_data, last_n_tokens_size, top_k, top_p, temp, repeat_penalty
|
|
|
|
)
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_sample_top_p_top_k.argtypes = [
|
2023-03-24 18:58:42 +00:00
|
|
|
llama_context_p,
|
|
|
|
llama_token_p,
|
|
|
|
c_int,
|
|
|
|
c_int,
|
2023-03-29 01:10:23 +00:00
|
|
|
c_float,
|
|
|
|
c_float,
|
|
|
|
c_float,
|
2023-03-24 18:58:42 +00:00
|
|
|
]
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_sample_top_p_top_k.restype = llama_token
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
# Performance information
|
|
|
|
|
2023-03-24 18:59:29 +00:00
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_print_timings(ctx: llama_context_p):
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_print_timings(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_print_timings.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_print_timings.restype = None
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_reset_timings(ctx: llama_context_p):
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_reset_timings(ctx)
|
2023-03-23 09:33:06 +00:00
|
|
|
|
2023-03-24 18:35:41 +00:00
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_reset_timings.argtypes = [llama_context_p]
|
|
|
|
_lib.llama_reset_timings.restype = None
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
# Print system information
|
2023-03-23 09:33:06 +00:00
|
|
|
def llama_print_system_info() -> bytes:
|
2023-03-24 22:43:29 +00:00
|
|
|
return _lib.llama_print_system_info()
|
2023-03-24 18:58:42 +00:00
|
|
|
|
|
|
|
|
2023-03-24 22:43:29 +00:00
|
|
|
_lib.llama_print_system_info.argtypes = []
|
|
|
|
_lib.llama_print_system_info.restype = c_char_p
|