vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

batch size improvements

Browse source
This commit is contained in:
Concedo 2023-12-22 15:27:40 +08:00
parent 230a638512
commit 77463e0e9c
3 changed files with 7 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

1
expose.h
View file

@ -27,7 +27,6 @@ struct load_model_inputs
const int threads; const int threads;
const int blasthreads; const int blasthreads;
const int max_context_length; const int max_context_length;
const int batch_size;
const bool low_vram; const bool low_vram;
const bool use_mmq; const bool use_mmq;
const char * executable_path; const char * executable_path;

9
gpttype_adapter.cpp
View file

@ -82,6 +82,9 @@ static int n_batch = 8;
static bool useSmartContext = false; static bool useSmartContext = false;
static bool useContextShift = false; static bool useContextShift = false;
static int blasbatchsize = 512; static int blasbatchsize = 512;
static int dontblasbatchsize = 16;
static int normalbatchsize = 32;
static int smallbatchsize = 8;
static int debugmode = 0; //-1 = hide all, 0 = normal, 1 = showall static int debugmode = 0; //-1 = hide all, 0 = normal, 1 = showall
static std::string modelname; static std::string modelname;
static std::vector<gpt_vocab::id> last_n_tokens; static std::vector<gpt_vocab::id> last_n_tokens;
@ -671,7 +674,9 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
file_format = in_file_format; file_format = in_file_format;
n_threads = params.n_threads = inputs.threads; n_threads = params.n_threads = inputs.threads;
n_blasthreads = params.n_threads_batch = inputs.blasthreads; n_blasthreads = params.n_threads_batch = inputs.blasthreads;
n_batch = params.n_batch = inputs.batch_size; bool isGguf = (file_format == FileFormat::GGUF_LLAMA || file_format==FileFormat::GGUF_FALCON);
n_batch = params.n_batch = (isGguf?normalbatchsize:smallbatchsize);
modelname = params.model = inputs.model_filename; modelname = params.model = inputs.model_filename;
useSmartContext = inputs.use_smartcontext; useSmartContext = inputs.use_smartcontext;
useContextShift = inputs.use_contextshift; useContextShift = inputs.use_contextshift;
@ -679,7 +684,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
blasbatchsize = inputs.blasbatchsize; blasbatchsize = inputs.blasbatchsize;
if(blasbatchsize<=0) if(blasbatchsize<=0)
{ {
blasbatchsize = 8; blasbatchsize = (isGguf?dontblasbatchsize:smallbatchsize);
} }
auto clamped_max_context_length = inputs.max_context_length; auto clamped_max_context_length = inputs.max_context_length;

2
koboldcpp.py
View file

@ -23,7 +23,6 @@ class load_model_inputs(ctypes.Structure):
_fields_ = [("threads", ctypes.c_int), _fields_ = [("threads", ctypes.c_int),
("blasthreads", ctypes.c_int), ("blasthreads", ctypes.c_int),
("max_context_length", ctypes.c_int), ("max_context_length", ctypes.c_int),
("batch_size", ctypes.c_int),
("low_vram", ctypes.c_bool), ("low_vram", ctypes.c_bool),
("use_mmq", ctypes.c_bool), ("use_mmq", ctypes.c_bool),
("executable_path", ctypes.c_char_p), ("executable_path", ctypes.c_char_p),
@ -229,7 +228,6 @@ def load_model(model_filename):
global args global args
inputs = load_model_inputs() inputs = load_model_inputs()
inputs.model_filename = model_filename.encode("UTF-8") inputs.model_filename = model_filename.encode("UTF-8")
inputs.batch_size = 8
inputs.max_context_length = maxctx #initial value to use for ctx, can be overwritten inputs.max_context_length = maxctx #initial value to use for ctx, can be overwritten
inputs.threads = args.threads inputs.threads = args.threads
inputs.low_vram = (True if (args.usecublas and "lowvram" in args.usecublas) else False) inputs.low_vram = (True if (args.usecublas and "lowvram" in args.usecublas) else False)