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

add option for 4th gpu, also fixed missing case in auto rope scaling

Browse source
This commit is contained in:
Concedo 2023-09-11 11:43:54 +08:00
parent 2dc96687eb
commit 6667fdcec8
2 changed files with 14 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

2
gpttype_adapter.cpp
View file

@ -443,7 +443,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
float factor = file_format_meta.n_ctx_train/2048; float factor = file_format_meta.n_ctx_train/2048;
effectivenctx = effectivenctx/factor; effectivenctx = effectivenctx/factor;
} }
rope_freq_base = (effectivenctx <= 3072 ? 26000.0f : (effectivenctx <= 4096 ? 32000.0f : (effectivenctx <= 6144 ? 54000.0f : (effectivenctx <= 8192 ? 82684.0f : (effectivenctx <= 12288 ? 140000.0f : 200000.0f))))); rope_freq_base = (effectivenctx <= 2048 ? 10000.0f : (effectivenctx <= 3072 ? 26000.0f : (effectivenctx <= 4096 ? 32000.0f : (effectivenctx <= 6144 ? 54000.0f : (effectivenctx <= 8192 ? 82684.0f : (effectivenctx <= 12288 ? 140000.0f : 200000.0f))))));
} }

21
koboldcpp.py
View file

@ -237,6 +237,9 @@ def load_model(model_filename):
elif (args.usecublas and "2" in args.usecublas): elif (args.usecublas and "2" in args.usecublas):
os.environ["CUDA_VISIBLE_DEVICES"] = "2" os.environ["CUDA_VISIBLE_DEVICES"] = "2"
os.environ["HIP_VISIBLE_DEVICES"] = "2" os.environ["HIP_VISIBLE_DEVICES"] = "2"
elif (args.usecublas and "3" in args.usecublas):
os.environ["CUDA_VISIBLE_DEVICES"] = "3"
os.environ["HIP_VISIBLE_DEVICES"] = "3"
else: else:
if (args.usecublas and "0" in args.usecublas): if (args.usecublas and "0" in args.usecublas):
inputs.cublas_info = 0 inputs.cublas_info = 0
@ -244,6 +247,8 @@ def load_model(model_filename):
inputs.cublas_info = 1 inputs.cublas_info = 1
elif (args.usecublas and "2" in args.usecublas): elif (args.usecublas and "2" in args.usecublas):
inputs.cublas_info = 2 inputs.cublas_info = 2
elif (args.usecublas and "3" in args.usecublas):
inputs.cublas_info = 3
inputs.executable_path = (getdirpath()+"/").encode("UTF-8") inputs.executable_path = (getdirpath()+"/").encode("UTF-8")
inputs.debugmode = args.debugmode inputs.debugmode = args.debugmode
@ -960,8 +965,8 @@ def show_new_gui():
# gpu options # gpu options
quick_gpu_selector_label = makelabel(quick_tab, "GPU ID:", 3) quick_gpu_selector_label = makelabel(quick_tab, "GPU ID:", 3)
quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=["1","2","3"], width=60, variable=gpu_choice_var, state="readonly") quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=["1","2","3","4"], width=60, variable=gpu_choice_var, state="readonly")
CUDA_quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=["1","2","3","All"], width=60, variable=gpu_choice_var, state="readonly") CUDA_quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=["1","2","3","4","All"], width=60, variable=gpu_choice_var, state="readonly")
quick_gpu_layers_entry,quick_gpu_layers_label = makelabelentry(quick_tab,"GPU Layers:", gpulayers_var, 5, 50) quick_gpu_layers_entry,quick_gpu_layers_label = makelabelentry(quick_tab,"GPU Layers:", gpulayers_var, 5, 50)
quick_lowvram_box = makecheckbox(quick_tab, "Low VRAM", lowvram_var, 4,0) quick_lowvram_box = makecheckbox(quick_tab, "Low VRAM", lowvram_var, 4,0)
quick_mmq_box = makecheckbox(quick_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1) quick_mmq_box = makecheckbox(quick_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1)
@ -996,8 +1001,8 @@ def show_new_gui():
# gpu options # gpu options
gpu_selector_label = makelabel(hardware_tab, "GPU ID:", 3) gpu_selector_label = makelabel(hardware_tab, "GPU ID:", 3)
gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=["1","2","3"], width=60, variable=gpu_choice_var, state="readonly") gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=["1","2","3","4"], width=60, variable=gpu_choice_var, state="readonly")
CUDA_gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=["1","2","3", "All"], width=60, variable=gpu_choice_var, state="readonly") CUDA_gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=["1","2","3","4", "All"], width=60, variable=gpu_choice_var, state="readonly")
gpu_layers_entry,gpu_layers_label = makelabelentry(hardware_tab,"GPU Layers:", gpulayers_var, 5, 50) gpu_layers_entry,gpu_layers_label = makelabelentry(hardware_tab,"GPU Layers:", gpulayers_var, 5, 50)
lowvram_box = makecheckbox(hardware_tab, "Low VRAM", lowvram_var, 4,0) lowvram_box = makecheckbox(hardware_tab, "Low VRAM", lowvram_var, 4,0)
mmq_box = makecheckbox(hardware_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1) mmq_box = makecheckbox(hardware_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1)
@ -1131,7 +1136,7 @@ def show_new_gui():
if gpu_choice_var.get()!="All": if gpu_choice_var.get()!="All":
gpuchoiceidx = int(gpu_choice_var.get())-1 gpuchoiceidx = int(gpu_choice_var.get())-1
if runopts_var.get() == "Use CLBlast": if runopts_var.get() == "Use CLBlast":
args.useclblast = [[0,0], [1,0], [0,1]][gpuchoiceidx] args.useclblast = [[0,0], [1,0], [0,1], [1,1]][gpuchoiceidx]
if runopts_var.get() == "Use CuBLAS/hipBLAS": if runopts_var.get() == "Use CuBLAS/hipBLAS":
if gpu_choice_var.get()=="All": if gpu_choice_var.get()=="All":
args.usecublas = ["lowvram"] if lowvram_var.get() == 1 else ["normal"] args.usecublas = ["lowvram"] if lowvram_var.get() == 1 else ["normal"]
@ -1187,14 +1192,14 @@ def show_new_gui():
if "useclblast" in dict and dict["useclblast"]: if "useclblast" in dict and dict["useclblast"]:
if clblast_option is not None: if clblast_option is not None:
runopts_var.set(clblast_option) runopts_var.set(clblast_option)
gpu_choice_var.set(str(["0 0", "1 0", "0 1"].index(str(dict["useclblast"][0]) + " " + str(dict["useclblast"][1])) + 1)) gpu_choice_var.set(str(["0 0", "1 0", "0 1", "1 1"].index(str(dict["useclblast"][0]) + " " + str(dict["useclblast"][1])) + 1))
elif "usecublas" in dict and dict["usecublas"]: elif "usecublas" in dict and dict["usecublas"]:
if cublas_option is not None: if cublas_option is not None:
runopts_var.set(cublas_option) runopts_var.set(cublas_option)
lowvram_var.set(1 if "lowvram" in dict["usecublas"] else 0) lowvram_var.set(1 if "lowvram" in dict["usecublas"] else 0)
mmq_var.set(1 if "mmq" in dict["usecublas"] else 0) mmq_var.set(1 if "mmq" in dict["usecublas"] else 0)
gpu_choice_var.set("All") gpu_choice_var.set("All")
for g in range(3): for g in range(4):
if str(g) in dict["usecublas"]: if str(g) in dict["usecublas"]:
gpu_choice_var.set(str(g+1)) gpu_choice_var.set(str(g+1))
break break
@ -1800,7 +1805,7 @@ if __name__ == '__main__':
compatgroup = parser.add_mutually_exclusive_group() compatgroup = parser.add_mutually_exclusive_group()
compatgroup.add_argument("--noblas", help="Do not use OpenBLAS for accelerated prompt ingestion", action='store_true') compatgroup.add_argument("--noblas", help="Do not use OpenBLAS for accelerated prompt ingestion", action='store_true')
compatgroup.add_argument("--useclblast", help="Use CLBlast for GPU Acceleration. Must specify exactly 2 arguments, platform ID and device ID (e.g. --useclblast 1 0).", type=int, choices=range(0,9), nargs=2) compatgroup.add_argument("--useclblast", help="Use CLBlast for GPU Acceleration. Must specify exactly 2 arguments, platform ID and device ID (e.g. --useclblast 1 0).", type=int, choices=range(0,9), nargs=2)
compatgroup.add_argument("--usecublas", help="Use CuBLAS/hipBLAS for GPU Acceleration. Requires CUDA. Select lowvram to not allocate VRAM scratch buffer. Enter a number afterwards to select and use 1 GPU. Leaving no number will use all GPUs.", nargs='*',metavar=('[lowvram|normal] [main GPU ID] [mmq]'), choices=['normal', 'lowvram', '0', '1', '2', 'mmq']) compatgroup.add_argument("--usecublas", help="Use CuBLAS/hipBLAS for GPU Acceleration. Requires CUDA. Select lowvram to not allocate VRAM scratch buffer. Enter a number afterwards to select and use 1 GPU. Leaving no number will use all GPUs.", nargs='*',metavar=('[lowvram|normal] [main GPU ID] [mmq]'), choices=['normal', 'lowvram', '0', '1', '2', '3', 'mmq'])
parser.add_argument("--gpulayers", help="Set number of layers to offload to GPU when using GPU. Requires GPU.",metavar=('[GPU layers]'), type=int, default=0) parser.add_argument("--gpulayers", help="Set number of layers to offload to GPU when using GPU. Requires GPU.",metavar=('[GPU layers]'), type=int, default=0)
parser.add_argument("--tensor_split", help="For CUDA with ALL GPU set only, ratio to split tensors across multiple GPUs, space-separated list of proportions, e.g. 7 3", metavar=('[Ratios]'), type=float, nargs='+') parser.add_argument("--tensor_split", help="For CUDA with ALL GPU set only, ratio to split tensors across multiple GPUs, space-separated list of proportions, e.g. 7 3", metavar=('[Ratios]'), type=float, nargs='+')