2024-04-30 01:35:38 -04:00
from __future__ import annotations
2023-11-08 04:48:51 +01:00
import os
from ctypes import (
c_bool ,
c_char_p ,
c_int ,
c_uint8 ,
c_float ,
c_void_p ,
POINTER ,
_Pointer , # type: ignore
Structure ,
)
import pathlib
2024-04-30 01:35:38 -04:00
from typing import (
Union ,
NewType ,
Optional ,
TYPE_CHECKING ,
)
2023-11-08 04:48:51 +01:00
import llama_cpp . llama_cpp as llama_cpp
2024-09-26 11:23:00 -04:00
from llama_cpp . _ctypes_extensions import (
load_shared_library ,
ctypes_function_for_shared_library ,
)
2024-04-30 01:35:38 -04:00
2024-09-26 11:23:00 -04:00
if TYPE_CHECKING :
from llama_cpp . _ctypes_extensions import (
CtypesArray ,
2023-11-08 04:48:51 +01:00
)
# Specify the base name of the shared library to load
_libllava_base_name = " llava "
2024-09-26 11:23:00 -04:00
_libllava_override_path = os . environ . get ( " LLAVA_CPP_LIB " )
_libllava_base_path = pathlib . Path ( os . path . abspath ( os . path . dirname ( __file__ ) ) ) / " lib " if _libllava_override_path is None else pathlib . Path ( )
2023-11-08 04:48:51 +01:00
# Load the library
2024-09-26 11:23:00 -04:00
_libllava = load_shared_library ( _libllava_base_name , _libllava_base_path )
2024-02-26 11:07:33 -05:00
ctypes_function = ctypes_function_for_shared_library ( _libllava )
2023-11-08 04:48:51 +01:00
################################################
# llava.h
################################################
# struct clip_ctx;
2024-02-21 16:25:38 -05:00
clip_ctx_p = NewType ( " clip_ctx_p " , int )
clip_ctx_p_ctypes = c_void_p
2023-11-08 04:48:51 +01:00
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# struct llava_image_embed {
# float * embed;
# int n_image_pos;
# };
class llava_image_embed ( Structure ) :
_fields_ = [
( " embed " , POINTER ( c_float ) ) ,
( " n_image_pos " , c_int ) ,
]
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# /** sanity check for clip <-> llava embed size match */
# LLAVA_API bool llava_validate_embed_size(const llama_context * ctx_llama, const clip_ctx * ctx_clip);
2024-04-30 01:35:38 -04:00
@ctypes_function (
" llava_validate_embed_size " ,
[ llama_cpp . llama_context_p_ctypes , clip_ctx_p_ctypes ] ,
c_bool ,
)
def llava_validate_embed_size (
ctx_llama : llama_cpp . llama_context_p , ctx_clip : clip_ctx_p , /
2024-09-18 20:11:02 -04:00
) - > bool :
. . .
2023-11-08 04:48:51 +01:00
# /** build an image embed from image file bytes */
# LLAVA_API struct llava_image_embed * llava_image_embed_make_with_bytes(struct clip_ctx * ctx_clip, int n_threads, const unsigned char * image_bytes, int image_bytes_length);
2024-04-30 01:35:38 -04:00
@ctypes_function (
" llava_image_embed_make_with_bytes " ,
[ clip_ctx_p_ctypes , c_int , POINTER ( c_uint8 ) , c_int ] ,
POINTER ( llava_image_embed ) ,
)
def llava_image_embed_make_with_bytes (
ctx_clip : clip_ctx_p ,
n_threads : Union [ c_int , int ] ,
image_bytes : CtypesArray [ c_uint8 ] ,
image_bytes_length : Union [ c_int , int ] ,
/ ,
2024-09-18 20:11:02 -04:00
) - > " _Pointer[llava_image_embed] " :
. . .
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# /** build an image embed from a path to an image filename */
# LLAVA_API struct llava_image_embed * llava_image_embed_make_with_filename(struct clip_ctx * ctx_clip, int n_threads, const char * image_path);
2024-04-30 01:35:38 -04:00
@ctypes_function (
" llava_image_embed_make_with_filename " ,
[ clip_ctx_p_ctypes , c_int , c_char_p ] ,
POINTER ( llava_image_embed ) ,
)
def llava_image_embed_make_with_filename (
ctx_clip : clip_ctx_p , n_threads : Union [ c_int , int ] , image_path : bytes , /
2024-09-18 20:11:02 -04:00
) - > " _Pointer[llava_image_embed] " :
. . .
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# LLAVA_API void llava_image_embed_free(struct llava_image_embed * embed);
# /** free an embedding made with llava_image_embed_make_* */
2024-02-26 11:07:33 -05:00
@ctypes_function ( " llava_image_embed_free " , [ POINTER ( llava_image_embed ) ] , None )
2024-09-18 20:11:02 -04:00
def llava_image_embed_free ( embed : " _Pointer[llava_image_embed] " , / ) :
. . .
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# /** write the image represented by embed into the llama context with batch size n_batch, starting at context pos n_past. on completion, n_past points to the next position in the context after the image embed. */
# LLAVA_API bool llava_eval_image_embed(struct llama_context * ctx_llama, const struct llava_image_embed * embed, int n_batch, int * n_past);
2024-04-30 01:35:38 -04:00
@ctypes_function (
" llava_eval_image_embed " ,
[
llama_cpp . llama_context_p_ctypes ,
POINTER ( llava_image_embed ) ,
c_int ,
POINTER ( c_int ) ,
] ,
c_bool ,
)
def llava_eval_image_embed (
ctx_llama : llama_cpp . llama_context_p ,
embed : " _Pointer[llava_image_embed] " ,
n_batch : Union [ c_int , int ] ,
n_past : " _Pointer[c_int] " ,
/ ,
2024-09-18 20:11:02 -04:00
) - > bool :
. . .
2023-11-08 04:48:51 +01:00
################################################
# clip.h
################################################
# /** load mmproj model */
2024-02-14 03:47:21 -05:00
# CLIP_API struct clip_ctx * clip_model_load (const char * fname, int verbosity);
2024-02-26 11:07:33 -05:00
@ctypes_function ( " clip_model_load " , [ c_char_p , c_int ] , clip_ctx_p_ctypes )
2024-04-30 01:35:38 -04:00
def clip_model_load (
fname : bytes , verbosity : Union [ c_int , int ] , /
2024-09-18 20:11:02 -04:00
) - > Optional [ clip_ctx_p ] :
. . .
2024-04-30 01:35:38 -04:00
2023-11-08 04:48:51 +01:00
# /** free mmproj model */
# CLIP_API void clip_free(struct clip_ctx * ctx);
2024-02-26 11:07:33 -05:00
@ctypes_function ( " clip_free " , [ clip_ctx_p_ctypes ] , None )
2024-09-18 20:11:02 -04:00
def clip_free ( ctx : clip_ctx_p , / ) :
. . .
2024-09-26 11:23:00 -04:00
