vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fixed savedatafile bug, try remove unneeded old clblast code path

Browse source
This commit is contained in:
Concedo 2025-04-03 19:11:27 +08:00
parent 103d60ed2c
commit 6e086bd309
2 changed files with 24 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

44
ggml/src/ggml-cpu/binary-ops.cpp
View file

@ -57,28 +57,28 @@ static void apply_binary_op(const ggml_compute_params * params, ggml_tensor * ds
GGML_ASSERT(ggml_can_repeat(src1, src0) && ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_can_repeat(src1, src0) && ggml_are_same_shape(src0, dst));
#if defined(GGML_USE_CLBLAST) // #if defined(GGML_USE_CLBLAST)
//do we even need this? it seems like its actually slower than just CPU // //do we even need this? it seems like its actually slower than just CPU
const int ith = params->ith; // const int ith = params->ith;
if (op == op_add && src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && src1->clblast_offload_gpu) { // if (op == op_add && src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && src1->clblast_offload_gpu) {
// TODO: OpenCL kernel support full broadcast // // TODO: OpenCL kernel support full broadcast
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); // static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
GGML_ASSERT((src1->ne[0] == src0->ne[0]) && ggml_can_repeat(src1, src0)); // GGML_ASSERT((src1->ne[0] == src0->ne[0]) && ggml_can_repeat(src1, src0));
if (ith == 0) { // if (ith == 0) {
ggml_cl_add(src0, src1, dst); // ggml_cl_add(src0, src1, dst);
} // }
return; // return;
} // }
if (op == op_mul && src0->type == GGML_TYPE_F32 && src1->clblast_offload_gpu) { // if (op == op_mul && src0->type == GGML_TYPE_F32 && src1->clblast_offload_gpu) {
// TODO: OpenCL kernel support full broadcast // // TODO: OpenCL kernel support full broadcast
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); // static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
GGML_ASSERT((src1->ne[0] == src0->ne[0]) && ggml_can_repeat(src1, src0)); // GGML_ASSERT((src1->ne[0] == src0->ne[0]) && ggml_can_repeat(src1, src0));
if (ith == 0) { // if (ith == 0) {
ggml_cl_mul(src0, src1, dst); // ggml_cl_mul(src0, src1, dst);
} // }
return; // return;
} // }
#endif // #endif
GGML_TENSOR_BINARY_OP_LOCALS GGML_TENSOR_BINARY_OP_LOCALS

3
koboldcpp.py
View file

@ -49,7 +49,7 @@ logit_bias_max = 512
dry_seq_break_max = 128 dry_seq_break_max = 128
# global vars # global vars
KcppVersion = "1.87.1" KcppVersion = "1.87.2"
showdebug = True showdebug = True
kcpp_instance = None #global running instance kcpp_instance = None #global running instance
global_memory = {"tunnel_url": "", "restart_target":"", "input_to_exit":False, "load_complete":False} global_memory = {"tunnel_url": "", "restart_target":"", "input_to_exit":False, "load_complete":False}
@ -5678,6 +5678,7 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
filepath = os.path.abspath(args.savedatafile) # Ensure it's an absolute path filepath = os.path.abspath(args.savedatafile) # Ensure it's an absolute path
if not filepath.endswith(".jsondb"): if not filepath.endswith(".jsondb"):
filepath += ".jsondb" filepath += ".jsondb"
args.savedatafile += ".jsondb"
try: try:
with open(filepath, 'r+', encoding='utf-8', errors='ignore') as f: with open(filepath, 'r+', encoding='utf-8', errors='ignore') as f:
loaded = json.load(f) loaded = json.load(f)