Update.

diff --git a/main.py b/main.py
index d63398c..7b8b642 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -362,6 +362,10 @@ def run_tests(model, quizz_machine, deterministic_synthesis):
 
             nb_test_samples += input.size(0)
 
+        test_perplexity = math.exp(min(100, acc_test_loss / nb_test_samples))
+
+        log_string(f"test_perplexity {n_epoch} {test_perplexity}")
+
         model.main_test_accuracy = quizz_machine.produce_results(
             n_epoch=n_epoch,
             model=model,
@@ -369,10 +373,6 @@ def run_tests(model, quizz_machine, deterministic_synthesis):
             deterministic_synthesis=deterministic_synthesis,
         )
 
-        test_perplexity = math.exp(min(100, acc_test_loss / nb_test_samples))
-
-        log_string(f"test_perplexity {n_epoch} {test_perplexity}")
-
 
 ######################################################################
 
@@ -401,33 +401,41 @@ def create_c_quizzes(
         nb_correct >= args.min_to_validate, nb_correct <= args.max_to_validate
     )
 
-    while valid_c_quizzes(recorded, standard_validity).size(0) < nb_to_create:
-        model_for_generation = models[torch.randint(len(models), (1,))]
+    file_name = os.path.join(args.result_dir, f"culture_c_quiz_{n_epoch:04d}_logp.dat")
+    with open(file_name, "w") as logp_file:
+        while valid_c_quizzes(recorded, standard_validity).size(0) < nb_to_create:
+            # Select a model at random to generate the new quizzes
 
-        c_quizzes, ave_seq_logproba = quizz_machine.generate_quizzes(
-            nb_to_create,
-            model_for_generation=model_for_generation,
-        )
+            model_for_generation = models[torch.randint(len(models), (1,))]
 
-        nb_correct = quizz_machine.compute_correctness(
-            c_quizzes, models, both_directions=args.both_directions
-        )
+            c_quizzes = quizz_machine.generate_quizzes(
+                nb_to_create,
+                model_for_generation=model_for_generation,
+            )
 
-        if args.dirty_debug:
-            nb_correct = torch.randint(
-                len(models) + 1, nb_correct.size(), device=c_quizzes.device
+            nb_correct, seq_logproba = quizz_machine.compute_correctness(
+                c_quizzes, models, both_directions=args.both_directions
             )
 
-        recorded.append((c_quizzes, nb_correct))
+            for n, l in zip(nb_correct, seq_logproba):
+                s = " ".join([str(x.item()) for x in l])
+                logp_file.write(f"{n} {s}\n")
 
-        nv = F.one_hot(nb_correct, num_classes=len(models) + 1).sum(0)
-        nv = " ".join([str(x.item()) for x in nv])
+            if args.dirty_debug:
+                nb_correct = torch.randint(
+                    len(models) + 1, nb_correct.size(), device=c_quizzes.device
+                )
 
-        nb_validated = valid_c_quizzes(recorded, standard_validity).size(0)
+            recorded.append((c_quizzes, nb_correct))
 
-        log_string(
-            f"keep c_quizzes model {model_for_generation.id} kept {nv} nb_accumulated {nb_validated} / {nb_to_create}"
-        )
+            nv = F.one_hot(nb_correct, num_classes=len(models) + 1).sum(0)
+            nv = " ".join([str(x.item()) for x in nv])
+
+            nb_validated = valid_c_quizzes(recorded, standard_validity).size(0)
+
+            log_string(
+                f"keep c_quizzes model {model_for_generation.id} kept {nv} nb_accumulated {nb_validated} / {nb_to_create}"
+            )
 
     # store the new c_quizzes which have been validated
 

diff --git a/quizz_machine.py b/quizz_machine.py
index 0d6d8f5..470b095 100755 (executable)
--- a/quizz_machine.py
+++ b/quizz_machine.py
@@ -29,8 +29,6 @@ def one_batch_masked_inplace_autoregression(
     seq_logproba,
     temperature=1.0,
     deterministic_synthesis=False,
-    forbidden_tokens=None,
-    forced_biases=None,
 ):
     to_generate = (ar_mask.sum(0) > 0).nonzero()
 
@@ -45,12 +43,6 @@ def one_batch_masked_inplace_autoregression(
 
         logits = (logits / temperature).log_softmax(dim=-1)
 
-        if forbidden_tokens is not None:
-            logits = logits.masked_fill(forbidden_tokens, float("-inf"))
-
-        if forced_biases is not None:
-            logits = logits + forced_biases[None, :]
-
         if deterministic_synthesis:
             t_next = logits.argmax(-1)
         else:
@@ -104,8 +96,6 @@ def masked_inplace_autoregression(
                 seq_logproba=seq_logproba,
                 temperature=temperature,
                 deterministic_synthesis=deterministic_synthesis,
-                forbidden_tokens=forbidden_tokens,
-                forced_biases=logit_biases,
             )
 
         model.train(t)
@@ -170,7 +160,6 @@ class QuizzMachine:
             )
 
     def save_quizzes(self, result_dir, filename_prefix, quizzes, prediction=False):
-        print(f"DEBUG {quizzes.size()=}")
         l = (quizzes.size(1) - 1) // 2
         forward = (quizzes[:, 0] == self.token_forward).long()
         backward = (quizzes[:, 0] == self.token_backward).long()
@@ -338,7 +327,11 @@ class QuizzMachine:
         reversed_c_quizzes = self.reverse_time(c_quizzes)
 
         ar_mask = self.make_ar_mask(c_quizzes)
-        seq_logproba = torch.empty(ar_mask.size(0), device=self.device)
+        seq_logproba = torch.zeros(
+            c_quizzes.size(0),
+            max([m.id for m in models_for_validation]) + 1,
+            device=self.device,
+        )
 
         # Check how many of models can solve the quizzes in both directions
 
@@ -347,12 +340,14 @@ class QuizzMachine:
         for model in models_for_validation:
             result = c_quizzes.clone()
 
+            seq_logproba[...] = 0.0
+
             masked_inplace_autoregression(
                 model=model,
                 batch_size=self.batch_size,
                 input=result,
                 ar_mask=ar_mask,
-                seq_logproba=seq_logproba,
+                seq_logproba=seq_logproba[:, model.id],
                 temperature=1.0,
                 deterministic_synthesis=True,
                 # progress_bar_desc="solving c_quizzes",
@@ -369,7 +364,7 @@ class QuizzMachine:
                     batch_size=self.batch_size,
                     input=reversed_result,
                     ar_mask=ar_mask,
-                    seq_logproba=seq_logproba,
+                    seq_logproba=seq_logproba[:, model.id],
                     temperature=1.0,
                     deterministic_synthesis=True,
                     # progress_bar_desc="solving reversed c_quizzes",
@@ -386,7 +381,7 @@ class QuizzMachine:
 
             nb_correct += correct
 
-        return nb_correct
+        return nb_correct, seq_logproba
 
     ###############################################################
 
@@ -401,7 +396,7 @@ class QuizzMachine:
         ar_mask_first[:, 0] = 0
         ar_mask_second[:, 0] = 0
 
-        seq_logproba = torch.empty(ar_mask_first.size(0), device=self.device)
+        seq_logproba = torch.zeros(ar_mask_first.size(0), device=self.device)
 
         temperature = 10.0
 
@@ -420,8 +415,6 @@ class QuizzMachine:
             device=self.device,
         )
 
-        ave_seq_logproba = seq_logproba.mean()
-
         # Then, we generate the prompt deterministically
 
         masked_inplace_autoregression(
@@ -451,4 +444,4 @@ class QuizzMachine:
             device=self.device,
         )
 
-        return c_quizzes, seq_logproba.mean()
+        return c_quizzes