def get_model_and_data_loaders(
config: ConfigParser,
logger: logging.Logger,
ckpt_path: Path,
) -> Tuple[torch.nn.Module, module_data.ExpertDataLoader]:
expert_dims, raw_input_dims, text_dim = compute_dims(config)
data_loaders = config.init(
name='data_loader',
module=module_data,
logger=logger,
raw_input_dims=raw_input_dims,
challenge_mode=config.get("challenge_mode", False),
text_dim=text_dim,
text_feat=config["experts"]["text_feat"],
text_agg=config["experts"]["text_agg"],
use_zeros_for_missing=config["experts"].get("use_zeros_for_missing",
False),
task=config.get("task", "retrieval"),
eval_only=True,
distil_params=config.get("distil_params", None),
training_file=config.get("training_file", None),
caption_masks=config.get("caption_masks", None),
ce_shared_dim=config["experts"].get("ce_shared_dim", None),
)
trn_config = compute_trn_config(config)
model = config.init(
name='arch',
module=module_arch,
trn_config=trn_config,
expert_dims=expert_dims,
text_dim=text_dim,
disable_nan_checks=config["disable_nan_checks"],
task=config.get("task", "retrieval"),
ce_shared_dim=config["experts"].get("ce_shared_dim", None),
feat_aggregation=config["data_loader"]["args"]["feat_aggregation"],
trn_cat=config["data_loader"]["args"].get("trn_cat", 0),
)
ckpt_path = config._args.resume
logger.info(f"Loading checkpoint: {ckpt_path} ...")
checkpoint = torch.load(ckpt_path)
state_dict = checkpoint['state_dict']
if config['n_gpu'] > 1:
model = torch.nn.DataParallel(model)
# support backwards compatibility
deprecated = ["ce.moe_fc_bottleneck1", "ce.moe_cg", "ce.moe_fc_proj"]
for mod in deprecated:
for suffix in ("weight", "bias"):
key = f"{mod}.{suffix}"
if key in state_dict:
print(f"WARNING: Removing deprecated key {key} from model")
state_dict.pop(key)
model.load_state_dict(state_dict)
return model, data_loaders
def get_model_and_data_loaders(
config: ConfigParser,
logger: logging.Logger,
ckpt_path: Path,
) -> Tuple[torch.nn.Module, module_data.ExpertDataLoader]:
expert_dims, raw_input_dims = compute_dims(config)
trn_config = compute_trn_config(config)
data_loaders = config.init(
name='data_loader',
module=module_data,
logger=logger,
raw_input_dims=raw_input_dims,
challenge_mode=config.get("challenge_mode", False),
text_feat=config["experts"]["text_feat"],
text_dim=config["experts"]["text_dim"],
text_agg=config["experts"]["text_agg"],
use_zeros_for_missing=config["experts"].get("use_zeros_for_missing", False),
task=config.get("task", "retrieval"),
eval_only=True,
)
model = config.init(
name='arch',
module=module_arch,
trn_config=trn_config,
expert_dims=expert_dims,
text_dim=config["experts"]["text_dim"],
disable_nan_checks=config["disable_nan_checks"],
task=config.get("task", "retrieval"),
ce_shared_dim=config["experts"].get("ce_shared_dim", None),
feat_aggregation=config["data_loader"]["args"]["feat_aggregation"],
trn_cat=config["data_loader"]["args"].get("trn_cat", 0),
)
ckpt_path = config._args.resume
logger.info(f"Loading checkpoint: {ckpt_path} ...")
checkpoint = torch.load(ckpt_path)
state_dict = checkpoint['state_dict']
if config['n_gpu'] > 1:
model = torch.nn.DataParallel(model)
model.load_state_dict(state_dict)
return model, data_loaders