deterministic_synthesis,
forbidden_tokens=None,
logit_biases=None,
- progress_bar_desc="autoregression",
+ progress_bar_desc=None,
device=torch.device("cpu"),
):
assert input.size() == ar_mask.size()
if result_dir is not None:
self.problem.save_quizzes(
- self.train_w_quizzes[:72], result_dir, f"culture_w_quizzes"
+ self.train_w_quizzes[:72], result_dir, "culture_w_quizzes"
)
def batches(self, split="train", desc=None):
def create_c_quizzes(
self,
+ nb,
+ model_for_generation,
+ models_for_validation,
+ min_ave_seq_logproba,
n_epoch,
result_dir,
logger,
- nb,
- model,
- other_models,
- min_ave_seq_logproba,
):
###############################################################
# Generate quizzes with model
seq_logproba[...] = 0
masked_inplace_autoregression(
- model=model,
+ model=model_for_generation,
batch_size=self.batch_size,
input=c_quizzes,
ar_mask=ar_mask,
seq_logproba=seq_logproba,
temperature=temperature,
deterministic_synthesis=False,
- progress_bar_desc="sampling c_quizzes",
+ # progress_bar_desc="sampling c_quizzes",
device=self.device,
)
else:
break
- logger(f"chaging temperature to {temperature}")
+ logger(f"changing temperature to {temperature}")
###############################################################
# Create the reverse quizzes
nb_correct = []
- for m in other_models:
+ for model in models_for_validation:
result = c_quizzes.clone()
masked_inplace_autoregression(
- model=m,
+ model=model,
batch_size=self.batch_size,
input=result,
ar_mask=ar_mask,
seq_logproba=seq_logproba,
temperature=1.0,
deterministic_synthesis=True,
- progress_bar_desc="solving c_quizzes",
+ # progress_bar_desc="solving c_quizzes",
device=self.device,
)
reverse_result = reverse_c_quizzes.clone()
masked_inplace_autoregression(
- model=m,
+ model=model,
batch_size=self.batch_size,
input=reverse_result,
ar_mask=ar_mask,
seq_logproba=seq_logproba,
temperature=1.0,
deterministic_synthesis=True,
- progress_bar_desc="solving reversed c_quizzes",
+ # progress_bar_desc="solving reversed c_quizzes",
device=self.device,
)