def create_model(name, model_spec, checkpoint_folder, write_spec=True):
checkpoint_path = os.path.join(checkpoint_folder, "{}.state.pth.tar".format(name))
print (checkpoint_path)
if os.path.isfile(checkpoint_path):
#my_p.load_state_dict(torch.load(checkpoint_path))
a = artificery.Artificery(checkpoint_init=False)
else:
a = artificery.Artificery(checkpoint_init=True)
spec_path = os.path.expanduser(model_spec["spec_path"])
spec_dir = os.path.dirname(spec_path)
my_p = a.parse(model_spec["spec_path"])
my_p.name = name
if write_spec:
model_spec_dst = os.path.join(checkpoint_folder, "model_spec.json")
subprocess.Popen("ls {}".format(spec_path), shell=True)
subprocess.Popen("cp {} {}".format(spec_path, model_spec_dst), shell=True)
for sf in a.used_specfiles:
sf = os.path.expanduser(sf)
rel_folder = os.path.dirname(os.path.relpath(sf, spec_dir))
dst_folder = os.path.join(checkpoint_folder, rel_folder)
if not os.path.exists(dst_folder):
os.makedirs(dst_folder)
subprocess.Popen("cp {} {}".format(sf, dst_folder), shell=True)
if os.path.isfile(checkpoint_path):
#my_p.load_state_dict(torch.load(checkpoint_path))
state_dict = torch.load(checkpoint_path)
own_state = my_p.state_dict()
reinit_downmodules = [7]
reinit_upmodules = []
for layer_name, param in state_dict.items():
load_weights = True
if "level_downmodules" in layer_name:
level = layer_name[len("level_downmodules.")]
level = int(level[:1])
if level in reinit_downmodules:
load_weights = False
elif "level_upmodules" in layer_name:
level = layer_name[len("level_upmodules.")]
level = int(level[:1])
if level in reinit_upmodules:
load_weights = False
if layer_name not in own_state:
load_weights = False
if load_weights:
if isinstance(param, Parameter):
# backwards compatibility for serialized parameters
param = param.data
own_state[layer_name].copy_(param)
#print ("Overwriting '{}'".format(layer_name))
return my_p
def open_model(name, checkpoint_folder):
a = artificery.Artificery()
spec_path = os.path.join(checkpoint_folder, "model_spec.json")
my_p = a.parse(spec_path)
checkpoint_path = os.path.join(checkpoint_folder, "{}.state.pth.tar".format(name))
if os.path.isfile(checkpoint_path):
my_p.load_state_dict(torch.load(checkpoint_path))
my_p.name = name
return my_p.cuda()
def create_model(name, checkpoint_folder, checkpoint_init=False):
a = artificery.Artificery(checkpoint_init=checkpoint_init)
spec_path = os.path.join(checkpoint_folder, "model_spec.json")
my_p = a.parse(spec_path)
checkpoint_path = os.path.join(checkpoint_folder,
"{}.state.pth.tar".format(name))
if os.path.isfile(checkpoint_path):
load_my_state_dict(my_p, torch.load(checkpoint_path))
#my_p.load_state_dict(torch.load(checkpoint_path))
my_p.name = name
return my_p.cuda()
def create_model(checkpoint_folder, device):
a = artificery.Artificery()
spec_path = os.path.join(checkpoint_folder, "model_spec.json")
my_p = a.parse(spec_path)
checkpoint_path = os.path.join(checkpoint_folder,
"checkpoint.state.pth.tar")
if not os.path.isfile(checkpoint_path):
raise Exception(f"Checkput path {checkpoint_path} no found!")
load_my_state_dict(
my_p, torch.load(checkpoint_path, map_location=torch.device(device)))
return my_p
def create_model(checkpoint_folder,
device='cpu',
checkpoint_name="checkpoint"):
a = artificery.Artificery()
spec_path = os.path.join(checkpoint_folder, "model_spec.json")
my_p = a.parse(spec_path)
checkpoint_path = os.path.join(checkpoint_folder,
f"{checkpoint_name}.state.pth.tar")
if not os.path.isfile(checkpoint_path):
print("creating new checkpiont...")
return my_p
load_my_state_dict(
my_p, torch.load(checkpoint_path, map_location=torch.device(device)))
my_p.name = checkpoint_name
return my_p
import artificery
a = artificery.Artificery()
net = a.parse("../params/pyramid_m5m6m7_fm24.json")
import pdb
pdb.set_trace()
