Update.
authorFrançois Fleuret <francois@fleuret.org>
Tue, 2 Jul 2024 06:55:03 +0000 (09:55 +0300)
committerFrançois Fleuret <francois@fleuret.org>
Tue, 2 Jul 2024 06:55:03 +0000 (09:55 +0300)
main.py
quizz_machine.py
sky.py

diff --git a/main.py b/main.py
index 714327d..d412e6c 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -79,7 +79,7 @@ parser.add_argument("--dropout", type=float, default=0.1)
 
 parser.add_argument("--deterministic_synthesis", action="store_true", default=False)
 
-parser.add_argument("--reverse_cleanup", action="store_true", default=False)
+parser.add_argument("--reverse_cleanup", action="store_true", default=True)
 
 parser.add_argument("--validation_forward_only", action="store_true", default=False)
 
@@ -364,7 +364,7 @@ def run_tests(model, quizz_machine, deterministic_synthesis):
 
             nb_test_samples += input.size(0)
 
-        main_test_accuracy = quizz_machine.produce_results(
+        model.main_test_accuracy = quizz_machine.produce_results(
             n_epoch=n_epoch,
             model=model,
             result_dir=args.result_dir,
@@ -375,8 +375,6 @@ def run_tests(model, quizz_machine, deterministic_synthesis):
 
         log_string(f"test_perplexity {n_epoch} {test_perplexity}")
 
-    model.main_test_accuracy = main_test_accuracy
-
 
 ######################################################################
 
@@ -397,8 +395,6 @@ def create_c_quizzes(
 ):
     recorded = []
 
-    sum_logits, sum_nb_c_quizzes = 0, 0
-
     nb_to_create = nb_for_train + nb_for_test
 
     # ------------------------------------------------------------
@@ -416,9 +412,6 @@ def create_c_quizzes(
             reverse_cleanup=args.reverse_cleanup,
         )
 
-        sum_logits += c_quizzes.size(0) * ave_seq_logproba
-        sum_nb_c_quizzes += c_quizzes.size(0)
-
         nb_correct = quizz_machine.compute_correctness(
             c_quizzes, models, both_directions=not args.validation_forward_only
         )
@@ -456,13 +449,14 @@ def create_c_quizzes(
             else ""
         )
 
-        quizz_machine.problem.save_quizzes(
-            valid_c_quizzes(recorded, criteria=lambda nb_correct: nb_correct == n)[:72],
-            args.result_dir,
-            f"culture_c_quiz_{n_epoch:04d}_N{n}{s}",
-        )
+        q = valid_c_quizzes(recorded, criteria=lambda nb_correct: nb_correct == n)[:72]
 
-    return sum_logits / sum_nb_c_quizzes
+        if q.size(0) > 0:
+            quizz_machine.problem.save_quizzes(
+                q,
+                args.result_dir,
+                f"culture_c_quiz_{n_epoch:04d}_N{n}{s}",
+            )
 
 
 ######################################################################
@@ -518,7 +512,7 @@ for n_epoch in range(args.nb_epochs):
     cta = " ".join([f"{float(m.main_test_accuracy):.04f}" for m in models])
     log_string(f"current_test_accuracies {cta}")
 
-    # replace a fraction of the w_quizzes with fresh ones
+    # replace a fraction of the w_quizzes with fresh ones
     quizz_machine.renew_w_quizzes(args.nb_train_samples // args.nb_gpts)
 
     if min([m.main_test_accuracy for m in models]) >= args.accuracy_to_make_c_quizzes:
index 4e7576e..697f27e 100755 (executable)
@@ -139,6 +139,7 @@ class QuizzMachine:
         self.train_w_quizzes = self.problem.generate_token_sequences(
             nb_train_samples
         ).to(device)
+
         self.test_w_quizzes = self.problem.generate_token_sequences(nb_test_samples).to(
             device
         )
diff --git a/sky.py b/sky.py
index 1164185..4ca4ba7 100755 (executable)
--- a/sky.py
+++ b/sky.py
@@ -157,6 +157,12 @@ class Sky(problem.Problem):
 
     ######################################################################
 
+    def generate_prompts_and_answers(self, nb):
+        frame_sequences = self.generate_frame_sequences(nb)
+        prompts = frame_sequences[:, : frame_sequences.size(0) // 2].flatten(1)
+        answers = frame_sequences[:, frame_sequences.size(0) // 2 :].flatten(1)
+        return prompts, answers
+
     def generate_token_sequences(self, nb):
         frame_sequences = self.generate_frame_sequences(nb)