Update.

diff --git a/grids.py b/grids.py
index 9372922..2717b22 100755 (executable)
--- a/grids.py
+++ b/grids.py
@@ -1754,7 +1754,7 @@ class Grids(problem.Problem):
         return quizzes
 
     def save_some_examples(self, result_dir, prefix=""):
-        nb, nrow = 128, 4
+        nb, nrow = 256, 8
         for t in self.all_tasks:
             print(t.__name__)
             quizzes = self.generate_w_quizzes_(nb, tasks=[t])

diff --git a/main.py b/main.py
index e95c4f6..8562e32 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -107,7 +107,7 @@ parser.add_argument("--min_succeed_to_validate", type=int, default=2)
 
 parser.add_argument("--max_fail_to_validate", type=int, default=3)
 
-parser.add_argument("--accuracy_to_make_c_quizzes", type=float, default=0.98)
+parser.add_argument("--accuracy_to_make_c_quizzes", type=float, default=0.95)
 
 parser.add_argument("--proba_understands", type=float, default=0.95)
 
@@ -1198,6 +1198,7 @@ def generate_ae_c_quizzes(models, nb, local_device=main_device):
 
         while bag_len(records) < wanted_nb:
             model = copy_for_inference(models[torch.randint(len(models), (1,)).item()])
+            generator_id = model.id
 
             c_quizzes = ae_generate(model, template, mask_generate)
 
@@ -1248,7 +1249,7 @@ def generate_ae_c_quizzes(models, nb, local_device=main_device):
                     e = "???"
 
                 log_string(
-                    f"nb_generated {bag_len(records)} model {model.id} (finishes {e} -- {int((nb_generated * 3600)/duration)}/h)"
+                    f"nb_generated {bag_len(records)} model {generator_id} (finishes {e} -- {int((nb_generated * 3600)/duration)}/h)"
                 )
 
         duration = time.perf_counter() - start_time