self.train_c_quizzes = []
self.test_c_quizzes = []
- def save_quizzes(
+ def save_quiz_illustrations(
self,
result_dir,
filename_prefix,
predicted_prompts *= 2
predicted_answers *= 2
- self.problem.save_quizzes(
+ self.problem.save_quiz_illustrations(
result_dir,
filename_prefix,
quizzes[:, 1 : 1 + self.prompt_len],
backward_nb_total = correct[n_backward].size(0)
self.logger(
- f"{log_prefix}_forward_accuracy {n_epoch} model {model.id} nb_correct {forward_nb_correct} / {forward_nb_total} ({forward_nb_correct*100/forward_nb_total} %)"
- )
-
- self.logger(
- f"{log_prefix}_backward_accuracy {n_epoch} model {model.id} nb_correct {backward_nb_correct} / {backward_nb_total} ({backward_nb_correct*100/backward_nb_total} %)"
+ f"{log_prefix}_accuracy {n_epoch} model {model.id} forward {forward_nb_correct} / {forward_nb_total} backward {backward_nb_correct} / {backward_nb_total}"
)
return result, correct
- compute_accuracy(model.train_w_quizzes[:nmax], log_prefix="train")
+ # compute_accuracy(model.train_w_quizzes[:nmax], log_prefix="train")
test_result, test_correct = compute_accuracy(
model.test_w_quizzes[:nmax], log_prefix="test"
##############################
- self.save_quizzes(
+ self.save_quiz_illustrations(
result_dir,
f"culture_prediction_{n_epoch:04d}_{model.id:02d}",
quizzes=test_result[:72],
else:
self.test_c_quizzes.append(new_c_quizzes.to("cpu"))
+ def save_c_quizzes(self, filename):
+ torch.save((self.train_c_quizzes, self.test_c_quizzes), filename)
+
+ def load_c_quizzes(self, filename):
+ self.train_c_quizzes, self.test_c_quizzes = torch.load(filename)
+
######################################################################
def logproba_of_solutions(self, models, c_quizzes):
logproba = c_quizzes.new_zeros(
- c_quizzes.size(0), len(models), device=self.device
+ c_quizzes.size(0), len(models), device=self.device, dtype=torch.float32
)
for model in models:
- for input, l in zip(
- c_quizzes.split(self.batch_size), logproba.split(self.batch_size)
- ):
- input = input.to(self.device)
- ar_mask = self.make_ar_mask(input)
- output = model(mygpt.BracketedSequence(input)).x
- ce = (
- F.cross_entropy(output.transpose(1, 2), input, reduction="none")
- * ar_mask
- )
- l[:, model.id] = -ce.sum(dim=-1)
+ with torch.autograd.no_grad():
+ t = model.training
+ model.eval()
+
+ for input, l in zip(
+ c_quizzes.split(self.batch_size), logproba.split(self.batch_size)
+ ):
+ input = input.to(self.device)
+ ar_mask = self.make_ar_mask(input)
+ output = model(mygpt.BracketedSequence(input)).x
+ ce = (
+ F.cross_entropy(output.transpose(1, 2), input, reduction="none")
+ * ar_mask
+ )
+ l[:, model.id] = -ce.sum(dim=-1)
+
+ model.train(t)
return logproba.to("cpu")