mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-09-19 01:05:09 +02:00
rearrage some elements in launcher
This commit is contained in:
Concedo
+71
-63
@@ -5036,6 +5036,14 @@ def show_gui():
|
||||
def actually_resize(windowwidth,windowheight,lastpos,smallratio):
|
||||
root.geometry(str(windowwidth) + "x" + str(windowheight) + str(lastpos))
|
||||
ctk.set_widget_scaling(smallratio)
|
||||
changerunmode(1,1,1)
|
||||
togglerope(1,1,1)
|
||||
toggleflashattn(1,1,1)
|
||||
togglectxshift(1,1,1)
|
||||
togglehorde(1,1,1)
|
||||
toggletaesd(1,1,1)
|
||||
togglejinja(1,1,1)
|
||||
tabbuttonaction(tabnames[curr_tab_idx])
|
||||
pass
|
||||
def on_resize(event):
|
||||
nonlocal resizing, resizing_id1, resizing_id2
|
||||
@@ -5078,14 +5086,6 @@ def show_gui():
|
||||
root.after_cancel(resizing_id1)
|
||||
resizing_id1 = None
|
||||
resizing_id1 = root.after(5, clearesizing)
|
||||
changerunmode(1,1,1)
|
||||
togglerope(1,1,1)
|
||||
toggleflashattn(1,1,1)
|
||||
togglectxshift(1,1,1)
|
||||
togglehorde(1,1,1)
|
||||
toggletaesd(1,1,1)
|
||||
togglejinja(1,1,1)
|
||||
tabbuttonaction(tabnames[curr_tab_idx])
|
||||
|
||||
if sys.platform=="darwin":
|
||||
root.resizable(False,False)
|
||||
@@ -5566,12 +5566,16 @@ def show_gui():
|
||||
dict = json.load(f)
|
||||
import_vars(dict)
|
||||
|
||||
def setup_backend_tooltip(parent):
|
||||
def setup_backend_tooltip(parent, singlerow):
|
||||
# backend count label with the tooltip function
|
||||
nl = '\n'
|
||||
tooltxt = "Number of backends you have built and available." + (f"\n\nMissing Backends: \n\n{nl.join(antirunopts)}" if len(runopts) < 8 else "")
|
||||
num_backends_built = makelabel(parent, str(len(runopts)) + "/9", 5, 2,tooltxt)
|
||||
num_backends_built.grid(row=1, column=1, padx=205, pady=0)
|
||||
num_backends_built = None
|
||||
num_backends_built = makelabel(parent, str(len(runopts)) + "/9", 1, 1,tooltxt)
|
||||
if singlerow:
|
||||
num_backends_built.grid(row=1, column=0, padx=355, pady=0)
|
||||
else:
|
||||
num_backends_built.grid(row=1, column=1, padx=205, pady=0)
|
||||
num_backends_built.configure(text_color="#00ff00")
|
||||
|
||||
def gui_changed_modelfile(*args):
|
||||
@@ -5593,7 +5597,7 @@ def show_gui():
|
||||
max_gpu_layers = (f"/{modelfile_extracted_meta[1][0]+1}" if (modelfile_extracted_meta and modelfile_extracted_meta[1] and modelfile_extracted_meta[1][0]!=0) else "")
|
||||
index = runopts_var.get()
|
||||
gpu_be = (index == "Use Vulkan" or index == "Use Vulkan (Old CPU)" or index == "Use CLBlast" or index == "Use CLBlast (Old CPU)" or index == "Use CLBlast (Older CPU)" or index == "Use CUDA" or index == "Use hipBLAS (ROCm)")
|
||||
layercounter_label.grid(row=6, column=1, padx=75, sticky="W")
|
||||
layercounter_label.grid(row=6, column=0, padx=230, sticky="W")
|
||||
quick_layercounter_label.grid(row=6, column=1, padx=75, sticky="W")
|
||||
if sys.platform=="darwin" and gpulayers_var.get()=="-1":
|
||||
quick_layercounter_label.configure(text="(Auto: All Layers)")
|
||||
@@ -5685,11 +5689,11 @@ def show_gui():
|
||||
index = runopts_var.get()
|
||||
if index == "Use Vulkan" or index == "Use Vulkan (Old CPU)" or index == "Use CLBlast" or index == "Use CLBlast (Old CPU)" or index == "Use CLBlast (Older CPU)" or index == "Use CUDA" or index == "Use hipBLAS (ROCm)":
|
||||
quick_gpuname_label.grid(row=3, column=1, padx=75, sticky="W")
|
||||
gpuname_label.grid(row=3, column=1, padx=75, sticky="W")
|
||||
gpuname_label.grid(row=3, column=0, padx=230, sticky="W")
|
||||
gpu_selector_label.grid(row=3, column=0, padx = 8, pady=1, stick="nw")
|
||||
quick_gpu_selector_label.grid(row=3, column=0, padx = 8, pady=1, stick="nw")
|
||||
if index == "Use CLBlast" or index == "Use CLBlast (Old CPU)" or index == "Use CLBlast (Older CPU)":
|
||||
gpu_selector_box.grid(row=3, column=1, padx=8, pady=1, stick="nw")
|
||||
gpu_selector_box.grid(row=3, column=0, padx=160, pady=1, stick="nw")
|
||||
quick_gpu_selector_box.grid(row=3, column=1, padx=8, pady=1, stick="nw")
|
||||
CUDA_gpu_selector_box.grid_remove()
|
||||
CUDA_quick_gpu_selector_box.grid_remove()
|
||||
@@ -5701,10 +5705,10 @@ def show_gui():
|
||||
elif index == "Use Vulkan" or index == "Use Vulkan (Old CPU)" or index == "Use CUDA" or index == "Use hipBLAS (ROCm)":
|
||||
gpu_selector_box.grid_remove()
|
||||
quick_gpu_selector_box.grid_remove()
|
||||
CUDA_gpu_selector_box.grid(row=3, column=1, padx=8, pady=1, stick="nw")
|
||||
CUDA_gpu_selector_box.grid(row=3, column=0, padx=160, pady=1, stick="nw")
|
||||
CUDA_quick_gpu_selector_box.grid(row=3, column=1, padx=8, pady=1, stick="nw")
|
||||
maingpu_label.grid(row=10, column=0, padx = 8, pady=1, stick="nw")
|
||||
maingpu_entry.grid(row=10, column=1, padx = 8, pady=1, stick="nw")
|
||||
maingpu_label.grid(row=8, column=0, padx = 270, pady=1, stick="nw")
|
||||
maingpu_entry.grid(row=8, column=0, padx = 340, pady=1, stick="nw")
|
||||
lowvram_box.grid(row=4, column=0, padx=8, pady=1, stick="nw")
|
||||
else:
|
||||
quick_gpuname_label.grid_remove()
|
||||
@@ -5720,11 +5724,11 @@ def show_gui():
|
||||
lowvram_box.grid_remove()
|
||||
|
||||
if index == "Use CUDA" or index == "Use hipBLAS (ROCm)":
|
||||
mmq_box.grid(row=4, column=1, padx=8, pady=1, stick="nw")
|
||||
mmq_box.grid(row=4, column=0, padx=160, pady=1, stick="nw")
|
||||
quick_mmq_box.grid(row=4, column=1, padx=8, pady=1, stick="nw")
|
||||
splitmode_box.grid(row=5, column=1, padx=8, pady=1, stick="nw")
|
||||
splitmode_box.grid(row=4, column=0, padx=300, pady=1, stick="nw")
|
||||
tensor_split_label.grid(row=8, column=0, padx = 8, pady=1, stick="nw")
|
||||
tensor_split_entry.grid(row=8, column=1, padx=8, pady=1, stick="nw")
|
||||
tensor_split_entry.grid(row=8, column=0, padx = 160, pady=1, stick="nw")
|
||||
else:
|
||||
mmq_box.grid_remove()
|
||||
quick_mmq_box.grid_remove()
|
||||
@@ -5734,16 +5738,16 @@ def show_gui():
|
||||
|
||||
if index == "Use Vulkan" or index == "Use Vulkan (Old CPU)":
|
||||
tensor_split_label.grid(row=8, column=0, padx = 8, pady=1, stick="nw")
|
||||
tensor_split_entry.grid(row=8, column=1, padx=8, pady=1, stick="nw")
|
||||
tensor_split_entry.grid(row=8, column=0, padx = 160, pady=1, stick="nw")
|
||||
|
||||
if index == "Use Vulkan" or index == "Use Vulkan (Old CPU)" or index == "Use CLBlast" or index == "Use CLBlast (Old CPU)" or index == "Use CLBlast (Older CPU)" or index == "Use CUDA" or index == "Use hipBLAS (ROCm)":
|
||||
gpu_layers_label.grid(row=6, column=0, padx = 8, pady=1, stick="nw")
|
||||
gpu_layers_entry.grid(row=6, column=1, padx=8, pady=1, stick="nw")
|
||||
gpu_layers_label.grid(row=6, column=0, padx=8, pady=1, stick="nw")
|
||||
gpu_layers_entry.grid(row=6, column=0, padx=160, pady=1, stick="nw")
|
||||
quick_gpu_layers_label.grid(row=6, column=0, padx = 8, pady=1, stick="nw")
|
||||
quick_gpu_layers_entry.grid(row=6, column=1, padx=8, pady=1, stick="nw")
|
||||
elif sys.platform=="darwin":
|
||||
gpu_layers_label.grid(row=6, column=0, padx = 8, pady=1, stick="nw")
|
||||
gpu_layers_entry.grid(row=6, column=1, padx=8, pady=1, stick="nw")
|
||||
gpu_layers_label.grid(row=6, column=0, padx=8, pady=1, stick="nw")
|
||||
gpu_layers_entry.grid(row=6, column=0, padx=160, pady=1, stick="nw")
|
||||
quick_gpu_layers_label.grid(row=6, column=0, padx = 8, pady=1, stick="nw")
|
||||
quick_gpu_layers_entry.grid(row=6, column=1, padx=8, pady=1, stick="nw")
|
||||
else:
|
||||
@@ -5762,20 +5766,23 @@ def show_gui():
|
||||
runoptbox.set(runopts[0]) # Set to first available option
|
||||
|
||||
# Tell user how many backends are available
|
||||
setup_backend_tooltip(quick_tab)
|
||||
setup_backend_tooltip(quick_tab,False)
|
||||
|
||||
# gpu options
|
||||
quick_gpu_selector_label = makelabel(quick_tab, "GPU ID:", 3,0,"Which GPU ID to load the model with.\nNormally your main GPU is #1, but it can vary for multi GPU setups.")
|
||||
quick_gpu_selector_label = makelabel(quick_tab, "GPU ID:", 3,0,"Which GPU ID to load the model with.\nNormally your main GPU is #1, but it can vary for multi GPU setups.",padx=8)
|
||||
quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=CLDevices, width=60, variable=gpu_choice_var, state="readonly")
|
||||
CUDA_quick_gpu_selector_box = ctk.CTkComboBox(quick_tab, values=CUDevices, width=60, variable=gpu_choice_var, state="readonly")
|
||||
CUDA_quick_gpu_selector_box.grid(row=3, column=1, padx=8, pady=1, stick="nw")
|
||||
quick_gpuname_label = ctk.CTkLabel(quick_tab, text="")
|
||||
quick_gpuname_label.grid(row=3, column=1, padx=75, sticky="W")
|
||||
quick_gpuname_label.configure(text_color="#ffff00")
|
||||
quick_gpu_layers_entry,quick_gpu_layers_label = makelabelentry(quick_tab,"GPU Layers:", gpulayers_var, 6, 50,tooltip="How many layers to offload onto the GPU.\nUsage varies based on model type and increases with model and context size.\nRequires some trial and error to find the best fit value.\n\nNote: The auto estimation is often inaccurate! Please set layers yourself for best results!")
|
||||
quick_gpu_layers_label.grid(row=6, column=0, padx = 8, pady=1, stick="nw")
|
||||
quick_gpu_layers_entry.grid(row=6, column=1, padx=8, pady=1, stick="nw")
|
||||
quick_layercounter_label = ctk.CTkLabel(quick_tab, text="")
|
||||
quick_layercounter_label.grid(row=6, column=1, padx=75, sticky="W")
|
||||
quick_layercounter_label.configure(text_color="#ffff00")
|
||||
quick_mmq_box = makecheckbox(quick_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1,tooltiptxt="Enable MMQ mode instead of CuBLAS for prompt processing. Read the wiki. Speed may vary.")
|
||||
quick_mmq_box = makecheckbox(quick_tab, "Use MMQ", mmq_var, 4,1,tooltiptxt="Enable MMQ mode instead of CuBLAS for prompt processing. Read the wiki. Speed may vary.")
|
||||
|
||||
# quick boxes
|
||||
quick_boxes = {
|
||||
@@ -5805,32 +5812,35 @@ def show_gui():
|
||||
# presets selector
|
||||
makelabel(hardware_tab, "Backend:", 1,0,"Select a backend to use.\nCUDA runs on Nvidia GPUs, and is much faster.\nVulkan and CLBlast works on all GPUs but is somewhat slower.\nOtherwise, runs on CPU only.\nNoAVX2 and Failsafe modes support older PCs.")
|
||||
runoptbox = ctk.CTkComboBox(hardware_tab, values=runopts, width=180,variable=runopts_var, state="readonly")
|
||||
runoptbox.grid(row=1, column=1,padx=8, stick="nw")
|
||||
runoptbox.grid(row=1, column=0,padx=160, stick="nw")
|
||||
runoptbox.set(runopts[0]) # Set to first available option
|
||||
|
||||
# Tell user how many backends are available
|
||||
setup_backend_tooltip(hardware_tab)
|
||||
setup_backend_tooltip(hardware_tab,True)
|
||||
|
||||
# gpu options
|
||||
gpu_selector_label = makelabel(hardware_tab, "GPU ID:", 3,0,"Which GPU ID to load the model with.\nNormally your main GPU is #1, but it can vary for multi GPU setups.")
|
||||
gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=CLDevices, width=60, variable=gpu_choice_var, state="readonly")
|
||||
gpu_selector_box.grid(row=3, column=0, padx=160, pady=1, stick="nw")
|
||||
CUDA_gpu_selector_box = ctk.CTkComboBox(hardware_tab, values=CUDevices, width=60, variable=gpu_choice_var, state="readonly")
|
||||
CUDA_gpu_selector_box.grid(row=3, column=0, padx=160, pady=1, stick="nw")
|
||||
gpuname_label = ctk.CTkLabel(hardware_tab, text="")
|
||||
gpuname_label.grid(row=3, column=1, padx=75, sticky="W")
|
||||
gpuname_label.grid(row=3, column=0, padx=230, sticky="W")
|
||||
gpuname_label.configure(text_color="#ffff00")
|
||||
gpu_layers_entry,gpu_layers_label = makelabelentry(hardware_tab,"GPU Layers:", gpulayers_var, 6, 50,tooltip="How many layers to offload onto the GPU.\nVRAM intensive, usage increases with model and context size.\nRequires some trial and error to find the best fit value.\n\nCommon values for total layers, accuracy not guaranteed.\n\nLlama/Mistral 7b/8b: 33\nSolar 10.7b/11b: 49\nLlama 13b: 41\nLlama 20b(stack): 63\nLlama/Yi 34b: 61\nMixtral 8x7b: 33\nLlama 70b: 81")
|
||||
lowvram_box = makecheckbox(hardware_tab, "No KV offload", lowvram_var, 4,0, tooltiptxt='Avoid offloading KV Cache or scratch buffers to VRAM.\nAllows more layers to fit, but may result in a large speed loss.')
|
||||
mmq_box = makecheckbox(hardware_tab, "Use MMQ", mmq_var, 4,0,padx=160, tooltiptxt="Enable MMQ mode to use finetuned kernels instead of default CuBLAS/HipBLAS for prompt processing.\nRead the wiki. Speed may vary.")
|
||||
splitmode_box = makecheckbox(hardware_tab, "Row-Split", rowsplit_var, 4,0,padx=300, tooltiptxt="Split rows across GPUs instead of splitting layers and KV across GPUs.\nUses the main GPU for small tensors and intermediate results. Speed may vary.")
|
||||
gpu_layers_entry,gpu_layers_label = makelabelentry(hardware_tab,"GPU Layers:", gpulayers_var, 6, 50, padx=160,singleline=True,tooltip="How many layers to offload onto the GPU.\nVRAM intensive, usage increases with model and context size.\nRequires some trial and error to find the best fit value.\n\nCommon values for total layers, accuracy not guaranteed.\n\nLlama/Mistral 7b/8b: 33\nSolar 10.7b/11b: 49\nLlama 13b: 41\nLlama 20b(stack): 63\nLlama/Yi 34b: 61\nMixtral 8x7b: 33\nLlama 70b: 81")
|
||||
layercounter_label = ctk.CTkLabel(hardware_tab, text="")
|
||||
layercounter_label.grid(row=6, column=1, padx=75, sticky="W")
|
||||
layercounter_label.grid(row=6, column=0, padx=230, sticky="W")
|
||||
layercounter_label.configure(text_color="#ffff00")
|
||||
tensor_split_entry,tensor_split_label = makelabelentry(hardware_tab, "Tensor Split:", tensor_split_str_vars, 8, 80, tooltip='When using multiple GPUs this option controls how large tensors should be split across all GPUs.\nUses a comma-separated list of non-negative values that assigns the proportion of data that each GPU should get in order.\nFor example, "3,2" will assign 60% of the data to GPU 0 and 40% to GPU 1.')
|
||||
lowvram_box = makecheckbox(hardware_tab, "Low VRAM (No KV offload)", lowvram_var, 4,0, tooltiptxt='Avoid offloading KV Cache or scratch buffers to VRAM.\nAllows more layers to fit, but may result in a large speed loss.')
|
||||
mmq_box = makecheckbox(hardware_tab, "Use QuantMatMul (mmq)", mmq_var, 4,1, tooltiptxt="Enable MMQ mode to use finetuned kernels instead of default CuBLAS/HipBLAS for prompt processing.\nRead the wiki. Speed may vary.")
|
||||
splitmode_box = makecheckbox(hardware_tab, "Row-Split", rowsplit_var, 5,0, tooltiptxt="Split rows across GPUs instead of splitting layers and KV across GPUs.\nUses the main GPU for small tensors and intermediate results. Speed may vary.")
|
||||
|
||||
maingpu_entry,maingpu_label = makelabelentry(hardware_tab, "Main GPU:" , maingpu_var, 10, 50,tooltip="Only for multi-gpu, which GPU to set as main?\nIf left blank, uses default value.")
|
||||
tensor_split_entry,tensor_split_label = makelabelentry(hardware_tab, "Tensor Split:", tensor_split_str_vars, 8, 80, padx=160, singleline=True, tooltip='When using multiple GPUs this option controls how large tensors should be split across all GPUs.\nUses a comma-separated list of non-negative values that assigns the proportion of data that each GPU should get in order.\nFor example, "3,2" will assign 60% of the data to GPU 0 and 40% to GPU 1.')
|
||||
maingpu_entry,maingpu_label = makelabelentry(hardware_tab, "Main GPU:" , maingpu_var, 8, 50,padx=340,singleline=True,tooltip="Only for multi-gpu, which GPU to set as main?\nIf left blank, uses default value.",labelpadx=270)
|
||||
|
||||
# threads
|
||||
makelabelentry(hardware_tab, "Threads:" , threads_var, 11, 50,tooltip="How many threads to use.\nRecommended value is your CPU core count, defaults are usually OK.")
|
||||
makelabelentry(hardware_tab, "Threads:" , threads_var, 11, 50, padx=160, singleline=True,tooltip="How many threads to use.\nRecommended value is your CPU core count, defaults are usually OK.")
|
||||
# blas thread specifier
|
||||
makelabelentry(hardware_tab, "Batch Threads:" , blas_threads_var, 11, 50,padx=340, singleline=True,tooltip="How many threads to use during batched processing.\nIf left blank, uses same value as regular thread count.",labelpadx=240)
|
||||
|
||||
# hardware checkboxes
|
||||
hardware_boxes = {
|
||||
@@ -5845,17 +5855,15 @@ def show_gui():
|
||||
}
|
||||
|
||||
for idx, (name, properties) in enumerate(hardware_boxes.items()):
|
||||
makecheckbox(hardware_tab, name, properties[0], int(idx/2) + 30, idx % 2, tooltiptxt=properties[1])
|
||||
makecheckbox(hardware_tab, name, properties[0], int(idx/2) + 30, 0, padx=(160 if idx % 2 else 8), tooltiptxt=properties[1])
|
||||
|
||||
# blas thread specifier
|
||||
makelabelentry(hardware_tab, "Batch Threads:" , blas_threads_var, 14, 50,tooltip="How many threads to use during batched processing.\nIf left blank, uses same value as regular thread count.")
|
||||
# blas batch size
|
||||
makeslider(hardware_tab, "Batch Size:", batchsize_text, blas_size_var, 0, len(batchsize_values)-1, 16,width=200, set=6,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
|
||||
blas_size_var.trace_add("write", changed_gpulayers_estimate)
|
||||
|
||||
# force version
|
||||
makelabelentry(hardware_tab, "Force Version:" , version_var, 100, 50,tooltip="If the autodetected version is wrong, you can change it here.\nLeave as 0 for default.")
|
||||
ctk.CTkButton(hardware_tab , text = "Run Benchmark", command = guibench ).grid(row=110,column=0, stick="se", padx= 0, pady=2)
|
||||
makelabelentry(hardware_tab, "Force Version:" , version_var, 100, 50,padx=160,singleline=True,tooltip="If the autodetected version is wrong, you can change it here.\nLeave as 0 for default.")
|
||||
ctk.CTkButton(hardware_tab , text = "Run Benchmark", command = guibench ).grid(row=110,column=0, stick="nw", padx= 8, pady=2)
|
||||
|
||||
|
||||
# Tokens Tab
|
||||
@@ -5877,25 +5885,25 @@ def show_gui():
|
||||
customrope_scale_entry, customrope_scale_label = makelabelentry(tokens_tab, "RoPE Scale:", customrope_scale, row=23, padx=(160 if corrupt_scaler else 100), singleline=True, tooltip="For Linear RoPE scaling. RoPE frequency scale.")
|
||||
customrope_base_entry, customrope_base_label = makelabelentry(tokens_tab, "Base:", customrope_base, row=23, padx=(420 if corrupt_scaler else 220), singleline=True, tooltip="For NTK Aware Scaling. RoPE frequency base.",labelpadx=(280 if corrupt_scaler else 180))
|
||||
def togglerope(a,b,c):
|
||||
manualropebox.grid_remove()
|
||||
nativectx_label.grid_remove()
|
||||
nativectx_entry.grid_remove()
|
||||
customrope_scale_label.grid_remove()
|
||||
customrope_scale_entry.grid_remove()
|
||||
customrope_base_label.grid_remove()
|
||||
customrope_base_entry.grid_remove()
|
||||
if customrope_var.get() == 1:
|
||||
manualropebox.grid()
|
||||
enabled_items = [customrope_scale_label, customrope_scale_entry,customrope_base_label, customrope_base_entry]
|
||||
disabled_items = [nativectx_entry,nativectx_label]
|
||||
for idx, item in enumerate(enabled_items):
|
||||
if manualrope_var.get() == 1:
|
||||
item.grid()
|
||||
else:
|
||||
item.grid_remove()
|
||||
for idx, item in enumerate(disabled_items):
|
||||
if manualrope_var.get() == 0:
|
||||
item.grid()
|
||||
else:
|
||||
item.grid_remove()
|
||||
else:
|
||||
disabled_items = [manualropebox, nativectx_entry,nativectx_label, customrope_scale_label, customrope_scale_entry, customrope_base_label, customrope_base_entry]
|
||||
for idx, item in enumerate(disabled_items):
|
||||
item.grid_remove()
|
||||
manualropebox = makecheckbox(tokens_tab, "Manual Rope Scale", variable=manualrope_var, row=22, command=togglerope, padx=(276 if corrupt_scaler else 166), tooltiptxt="Set RoPE base and scale manually.")
|
||||
manualropebox.grid(row=22, column=0,padx=(300 if corrupt_scaler else 200), pady=1, stick="nw")
|
||||
if manualrope_var.get() == 1:
|
||||
customrope_scale_label.grid(row=23, column=0, padx=8, pady=1, stick="nw")
|
||||
customrope_scale_entry.grid(row=23, column=0, padx=(160 if corrupt_scaler else 100), pady=1, stick="nw")
|
||||
customrope_base_label.grid(row=23, column=0, padx=(280 if corrupt_scaler else 180), pady=1, stick="nw")
|
||||
customrope_base_entry.grid(row=23, column=0, padx=(420 if corrupt_scaler else 220), pady=1, stick="nw")
|
||||
else:
|
||||
nativectx_label.grid(row=23, column=0, padx=8, pady=1, stick="nw")
|
||||
nativectx_entry.grid(row=23, column=0, padx=(246 if corrupt_scaler else 146), pady=1, stick="nw")
|
||||
|
||||
manualropebox = makecheckbox(tokens_tab, "Manual Rope Scale", variable=manualrope_var, row=22, command=togglerope, padx=(300 if corrupt_scaler else 200), tooltiptxt="Set RoPE base and scale manually.")
|
||||
|
||||
makecheckbox(tokens_tab, "Custom RoPE Config", variable=customrope_var, row=22, command=togglerope,tooltiptxt="Override the default RoPE configuration with custom RoPE scaling.")
|
||||
makecheckbox(tokens_tab, "Use FlashAttention", flashattention_var, 28, command=toggleflashattn, tooltiptxt="Enable flash attention for GGUF models.")
|
||||
|
||||
Reference in New Issue
Block a user