(evals): Medium configs (#415)

* updated medium configs

* fix problematic curriculum values / small issues causing exceptions to be raised

* optimus alpha config

* all configs so far

* fix tests

eval/yaml/medium/claude-3.5-sonnet.yaml

@@ -109,7 +109,7 @@ categories:
   - dataset: jugs
     params:
       num_jugs: 4
-      difficulty: 50
+      difficulty: 10
   - dataset: letter_counting
     params:
       min_words: 25
@@ -152,10 +152,10 @@ categories:
       max_length: 100
   - dataset: palindrome_partitioning
     params:
-      min_string_len: 50
-      max_string_len: 100
-      min_substring_palindrome_len: 5
-      max_substring_palindrome_len: 10
+      min_string_len: 5
+      max_string_len: 15
+      min_substring_palindrome_len: 1
+      max_substring_palindrome_len: 5
   - dataset: pool_matrix
     params:
       min_rows: 25
@@ -234,8 +234,8 @@ categories:
       mirrors_weights: [0.2, 0.2, 0.2, 0.2, 0.2]
   - dataset: rearc
     params:
-      pso_difficulty_weights: [0, 0, 0, 1, 0, 0, 0, 0]
-      rng_difficulty_weights: [0, 0, 0, 1, 0, 0, 0, 0]
+      pso_difficulty_weights: [0, 0, 0, 1, 0, 0, 0]
+      rng_difficulty_weights: [0, 0, 0, 1, 0, 0, 0]
 - category: arithmetic
   datasets:
   - dataset: basic_arithmetic
@@ -361,8 +361,8 @@ categories:
       max_num_statements: 500
   - dataset: number_sequence
     params:
-      min_terms: 8
-      max_terms: 12
+      min_terms: 5
+      max_terms: 10
       min_value: -500
       max_value: 500
       max_complexity: 3
@@ -378,16 +378,16 @@ categories:
   datasets:
   - dataset: countdown
     params:
-      min_numbers: 6
+      min_numbers: 3
       max_numbers: 9
       min_target: 100
       max_target: 1000
       min_value: 1
-      max_value: 250
+      max_value: 100
   - dataset: emoji_mystery
     params:
-      min_words_in_sentence: 20
-      max_words_in_sentence: 40
+      min_words_in_sentence: 10
+      max_words_in_sentence: 30
   - dataset: futoshiki
     params:
       min_board_size: 6
@@ -410,8 +410,8 @@ categories:
     params:
       min_grid_size: 25
       max_grid_size: 50
-      min_dist: 25
-      max_dist: 50
+      min_dist: 10
+      max_dist: 15
   - dataset: mini_sudoku
     params:
       min_empty: 6