Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-22 16:17:57 +01:00
Code Issues Packages Projects Releases Wiki Activity

Create get_max_memory_dict() function

Browse Source
This commit is contained in:
oobabooga 2023-05-15 19:38:27 -03:00
parent 071f0776ad
commit 4e66f68115
1 changed files with 27 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
modules/models.py
View File

@ -182,26 +182,7 @@ def load_model(model_name):
else: else:
params["torch_dtype"] = torch.float16 params["torch_dtype"] = torch.float16
if shared.args.gpu_memory: params['max_memory'] = get_max_memory_dict()
memory_map = list(map(lambda x: x.strip(), shared.args.gpu_memory))
max_cpu_memory = shared.args.cpu_memory.strip() if shared.args.cpu_memory is not None else '99GiB'
max_memory = {}
for i in range(len(memory_map)):
max_memory[i] = f'{memory_map[i]}GiB' if not re.match('.*ib$', memory_map[i].lower()) else memory_map[i]
max_memory['cpu'] = f'{max_cpu_memory}GiB' if not re.match('.*ib$', max_cpu_memory.lower()) else max_cpu_memory
params['max_memory'] = max_memory
elif shared.args.auto_devices:
total_mem = (torch.cuda.get_device_properties(0).total_memory / (1024 * 1024))
suggestion = round((total_mem - 1000) / 1000) * 1000
if total_mem - suggestion < 800:
suggestion -= 1000
suggestion = int(round(suggestion / 1000))
logging.warning(f"\033[1;32;1mAuto-assiging --gpu-memory {suggestion} for your GPU to try to prevent out-of-memory errors.\nYou can manually set other values.\033[0;37;0m")
max_memory = {0: f'{suggestion}GiB', 'cpu': f'{shared.args.cpu_memory or 99}GiB'}
params['max_memory'] = max_memory
if shared.args.disk: if shared.args.disk:
params["offload_folder"] = shared.args.disk_cache_dir params["offload_folder"] = shared.args.disk_cache_dir
@ -256,6 +237,32 @@ def load_model(model_name):
return model, tokenizer return model, tokenizer
def get_max_memory_dict():
max_memory = {}
if shared.args.gpu_memory:
memory_map = list(map(lambda x: x.strip(), shared.args.gpu_memory))
for i in range(len(memory_map)):
max_memory[i] = f'{memory_map[i]}GiB' if not re.match('.*ib$', memory_map[i].lower()) else memory_map[i]
max_cpu_memory = shared.args.cpu_memory.strip() if shared.args.cpu_memory is not None else '99GiB'
max_memory['cpu'] = f'{max_cpu_memory}GiB' if not re.match('.*ib$', max_cpu_memory.lower()) else max_cpu_memory
# If --auto-devices is provided standalone, try to get a reasonable value
# for the maximum memory of device :0
elif shared.args.auto_devices:
total_mem = (torch.cuda.get_device_properties(0).total_memory / (1024 * 1024))
suggestion = round((total_mem - 1000) / 1000) * 1000
if total_mem - suggestion < 800:
suggestion -= 1000
suggestion = int(round(suggestion / 1000))
logging.warning(f"Auto-assiging --gpu-memory {suggestion} for your GPU to try to prevent out-of-memory errors. You can manually set other values.")
max_memory = {0: f'{suggestion}GiB', 'cpu': f'{shared.args.cpu_memory or 99}GiB'}
return max_memory
def clear_torch_cache(): def clear_torch_cache():
gc.collect() gc.collect()
if not shared.args.cpu: if not shared.args.cpu: