explorer updated evals

all_outputs.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:395624aed41b3527d333818c1ee9f7d537b937e7b0eaa0d3cf05fa0743a2fbbd
-size 2528262
+oid sha256:b8fa8293294ed03c607e4c0d861b50203bd6aee395414d8f5d9d7e9a7853acc9
+size 284487753

app.py

@@ -56,6 +56,7 @@ all_evaluations_by_problem_colored = [
             for model in all_models
         },
         problems[idx]["difficulty"],
+        problems[idx]["question_id"],
     )
     for trueidx, idx in enumerate(random_idxs)
 ]
@@ -110,6 +111,7 @@ def problem(problem_idx):
 mini_models = [
     # "DeepSeek-V2",
     "DeepSeek-V3",
+    "DeepSeek-R1-Preview",
     # "DSCoder-33b-Ins",
     # "GPT-4-Turbo-2024-04-09",
     "GPT-4O-2024-05-13",
@@ -117,8 +119,8 @@ mini_models = [
     "Gemini-Flash-2.0-Thinking",
     # "Gemini-Exp-1206",
     # "Claude-3-Sonnet",
-    "O1-Mini-2024-09-12 (N=1)",
     "O1-2024-12-17 (N=1) (High)",
+    "QwQ-32B-Preview (N=1)",
 ]
 
 

templates/index.html

@@ -86,7 +86,7 @@
     <table id='model-table' style="align:center">
         <thead>
             <tr>
-                <th>Problem ID</th>
+                <th>Question ID</th>
                 <th>Difficulty</th>
                 {% for model in models %}
                 <th class="column-{{ model }}">{{ model }}</th>
@@ -94,9 +94,9 @@
             </tr>
         </thead>
         <tbody>
-            {% for problem_idx, problem, difficulty in problems %}
+            {% for problem_idx, problem, difficulty, question_id in problems %}
             <tr>
-                <td> <a href="{{ url_for('problem', problem_idx=problem_idx) }}"> {{ problem_idx }} </a> </td>
+                <td> <a href="{{ url_for('problem', problem_idx=problem_idx) }}"> {{ question_id }} </a> </td>
                 <td> {{ difficulty }} </td>
                 {% for model in models %}
                 <td style="background-color: {{ problem[model]['correctness_color'] }};" class="column-{{ model }}">

templates/index_mini.html

@@ -85,7 +85,7 @@
     <table id='model-table' style="align:center">
         <thead>
             <tr>
-                <th>Problem ID</th>
+                <th>Question ID</th>
                 <th>Difficulty</th>
                 {% for model in models %}
                 <th class="column-{{ model }}">{{ model }}</th>
@@ -93,9 +93,9 @@
             </tr>
         </thead>
         <tbody>
-            {% for problem_idx, problem, difficulty in problems %}
+            {% for problem_idx, problem, difficulty, question_id in problems %}
             <tr>
-                <td> <a href="{{ url_for('problem_mini', problem_idx=problem_idx) }}"> {{ problem_idx }} </a> </td>
+                <td> <a href="{{ url_for('problem_mini', problem_idx=problem_idx) }}"> {{ question_id }} </a> </td>
                 <td> {{ difficulty }} </td>
                 {% for model in models %}
                 <td style="background-color: {{ problem[model]['correctness_color'] }};" class="column-{{ model }}">