vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 09:34:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add used memory checks, add gpulayers for metal

Browse source
This commit is contained in:
Concedo 2024-08-05 16:32:05 +08:00
parent 23caa63f94
commit bd4e55eb74
1 changed files with 27 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

36
koboldcpp.py
View file

@ -46,6 +46,7 @@ showdebug = True
guimode = False
showsamplerwarning = True
showmaxctxwarning = True
showusedmemwarning = True
session_kudos_earned = 0
session_jobs = 0
session_starttime = None
@ -80,6 +81,7 @@ CUDevicesNames = ["","","","",""]
VKDevicesNames = ["","","",""]
VKIsDGPU = [0,0,0,0]
MaxMemory = [0]
MaxFreeMemory = [0]
class logit_bias(ctypes.Structure):
_fields_ = [("token_id", ctypes.c_int32),
@ -622,8 +624,16 @@ def extract_modelfile_params(filepath,sdfilepath,whisperfilepath,mmprojfilepath)
except Exception as ex:
modelfile_extracted_meta = None
def autoset_gpu_layers(ctxsize,gpumem,sdquanted,bbs): #shitty algo to determine how many layers to use
global modelfile_extracted_meta # reference cached values instead
def autoset_gpu_layers(ctxsize,sdquanted,bbs): #shitty algo to determine how many layers to use
global showusedmemwarning, modelfile_extracted_meta # reference cached values instead
gpumem = MaxMemory[0]
usedmem = 0
if MaxFreeMemory[0]>0:
usedmem = MaxMemory[0]-MaxFreeMemory[0]
if showusedmemwarning and usedmem > (2.5*1024*1024*1024):
showusedmemwarning = False
print(f"Note: KoboldCpp has detected that a significant amount of GPU VRAM ({usedmem/1024/1024} MB) is currently used by another application.\nFor best results, you may wish to close that application and then restart KoboldCpp.\n***")
reservedmem = max(1.5*1024*1024*1024,(0.5*1024*1024*1024 + usedmem)) # determine vram overhead
try:
if not modelfile_extracted_meta:
return 0
@ -647,15 +657,14 @@ def autoset_gpu_layers(ctxsize,gpumem,sdquanted,bbs): #shitty algo to determine
ggufmeta = modelfile_extracted_meta[0]
if not ggufmeta or ggufmeta[0]==0: #fail to read or no layers
sizeperlayer = fsize*csmul*0.052
layerlimit = int(min(200,mem/sizeperlayer))
layerlimit = int(min(200,(mem-usedmem)/sizeperlayer))
else:
layers = ggufmeta[0]
headcount = ggufmeta[1]
headkvlen = (ggufmeta[2] if ggufmeta[2] > 0 else 128)
ratio = mem/(fsize*csmul*1.5)
computemem = layers*(4 if bbs <= 512 else (bbs/128))*headkvlen*cs*4*1.5 # For now the first 4 is the hardcoded result for a blasbatchsize of 512. Ideally we automatically calculate blasbatchsize / 4 but I couldn't easily grab the value yet - Henk
ratio = (mem-usedmem)/(fsize*csmul*1.55)
computemem = layers*(4 if bbs <= 512 else (bbs/128))*headkvlen*cs*4*1.5 # apply blasbatchsize calculations if over 512
contextmem = layers*headcount*headkvlen*cs*4*1.1
reservedmem = 1.5*1024*1024*1024 # Users often don't have their GPU's VRAM worth of memory, we assume 500MB to avoid driver swapping + 500MB for the OS + 500MB for background apps / browser - Henk
if headcount > 0:
ratio = max(ratio, (mem - reservedmem - computemem) / (fsize + contextmem))
layerlimit = min(int(ratio*layers), (layers + 3))
@ -698,11 +707,13 @@ def fetch_gpu_properties(testCL,testCU,testVK):
if testCU:
FetchedCUdevices = []
FetchedCUdeviceMem = []
FetchedCUfreeMem = []
AMDgpu = None
try: # Get NVIDIA GPU names
output = subprocess.run(['nvidia-smi','--query-gpu=name,memory.total','--format=csv,noheader'], capture_output=True, text=True, check=True, encoding='utf-8').stdout
output = subprocess.run(['nvidia-smi','--query-gpu=name,memory.total,memory.free','--format=csv,noheader'], capture_output=True, text=True, check=True, encoding='utf-8').stdout
FetchedCUdevices = [line.split(",")[0].strip() for line in output.splitlines()]
FetchedCUdeviceMem = [line.split(",")[1].strip().split(" ")[0].strip() for line in output.splitlines()]
FetchedCUfreeMem = [line.split(",")[2].strip().split(" ")[0].strip() for line in output.splitlines()]
except Exception as e:
pass
if len(FetchedCUdevices)==0:
@ -730,6 +741,8 @@ def fetch_gpu_properties(testCL,testCU,testVK):
MaxMemory[0] = max(int(FetchedCUdeviceMem[idx]),MaxMemory[0])
else:
MaxMemory[0] = max(int(FetchedCUdeviceMem[idx])*1024*1024,MaxMemory[0])
if len(FetchedCUfreeMem)>idx:
MaxFreeMemory[0] = max(int(FetchedCUfreeMem[idx])*1024*1024,MaxFreeMemory[0])
if testVK:
try: # Get Vulkan names
@ -2458,7 +2471,7 @@ def show_gui():
pass
def changed_gpulayers_estimate(*args):
predicted_gpu_layers = autoset_gpu_layers(int(contextsize_text[context_var.get()]),MaxMemory[0],(sd_quant_var.get()==1),int(blasbatchsize_values[int(blas_size_var.get())]))
predicted_gpu_layers = autoset_gpu_layers(int(contextsize_text[context_var.get()]),(sd_quant_var.get()==1),int(blasbatchsize_values[int(blas_size_var.get())]))
max_gpu_layers = (f"/{modelfile_extracted_meta[0][0]+3}" if (modelfile_extracted_meta and modelfile_extracted_meta[0] and modelfile_extracted_meta[0][0]!=0) else "")
index = runopts_var.get()
gpu_be = (index == "Use Vulkan" or index == "Vulkan NoAVX2 (Old CPU)" or index == "Use CLBlast" or index == "CLBlast NoAVX2 (Old CPU)" or index == "Use CuBLAS" or index == "Use hipBLAS (ROCm)")
@ -2590,6 +2603,11 @@ def show_gui():
gpu_layers_entry.grid(row=6, column=1, padx=8, 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")
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:
gpu_layers_label.grid_remove()
gpu_layers_entry.grid_remove()
@ -3925,7 +3943,7 @@ def main(launch_args,start_server=True):
pass
if MaxMemory[0] > 0:
extract_modelfile_params(args.model_param,args.sdmodel,args.whispermodel,args.mmproj)
layeramt = autoset_gpu_layers(args.contextsize, MaxMemory[0],args.sdquant,args.blasbatchsize)
layeramt = autoset_gpu_layers(args.contextsize,args.sdquant,args.blasbatchsize)
print(f"Auto Recommended Layers: {layeramt}")
args.gpulayers = layeramt
else: