tmp finish naive ar sampling

generation/__init__.py

generation/autoregressive_sampling.py

generation/cuda2d_sampling.py

generation/local_sampling.py

generation/sampling_strategies/__init__.py

generation/sampling_strategies/base_strategy.py

model/cached_autoregressive_model.py

model/mixins.py

mpu/transformer.py

pretrain_cogview2.py

scripts/finetune_into_cogview2.sh

training/model_io.py