update leaderboards

ZeroEval-main/result_dirs/zebra-grid.summary.json

@@ -54,6 +54,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1549.74"
   },
+  {
+    "Model": "Mistral-Large-2",
+    "Mode": "greedy",
+    "Puzzle Acc": "29.00",
+    "Cell Acc": "47.64",
+    "No answer": "1.70",
+    "Easy Puzzle Acc": "80.36",
+    "Hard Puzzle Acc": "9.03",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1592.39"
+  },
   {
     "Model": "gpt-4-turbo-2024-04-09",
     "Mode": "greedy",
@@ -109,6 +120,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1165.90"
   },
+  {
+    "Model": "Meta-Llama-3.1-70B-Instruct",
+    "Mode": "greedy",
+    "Puzzle Acc": "24.90",
+    "Cell Acc": "27.98",
+    "No answer": "43.00",
+    "Easy Puzzle Acc": "73.57",
+    "Hard Puzzle Acc": "5.97",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1483.68"
+  },
   {
     "Model": "deepseek-chat",
     "Mode": "greedy",
@@ -142,6 +164,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1324.55"
   },
+  {
+    "Model": "DeepSeek-Coder-V2-0724",
+    "Mode": "greedy",
+    "Puzzle Acc": "20.50",
+    "Cell Acc": "42.35",
+    "No answer": "3.40",
+    "Easy Puzzle Acc": "61.79",
+    "Hard Puzzle Acc": "4.44",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1230.63"
+  },
   {
     "Model": "gpt-4o-mini-2024-07-18",
     "Mode": "greedy",
@@ -307,6 +340,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1078.29"
   },
+  {
+    "Model": "Meta-Llama-3.1-8B-Instruct",
+    "Mode": "greedy",
+    "Puzzle Acc": "12.80",
+    "Cell Acc": "13.68",
+    "No answer": "61.50",
+    "Easy Puzzle Acc": "43.57",
+    "Hard Puzzle Acc": "0.83",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1043.90"
+  },
   {
     "Model": "gemma-2-9b-it@nvidia",
     "Mode": "greedy",
@@ -439,6 +483,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1473.23"
   },
+  {
+    "Model": "gemma-2-2b-it",
+    "Mode": "greedy",
+    "Puzzle Acc": "4.20",
+    "Cell Acc": "9.97",
+    "No answer": "57.20",
+    "Easy Puzzle Acc": "14.29",
+    "Hard Puzzle Acc": "0.28",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1032.89"
+  },
   {
     "Model": "Yi-1.5-9B-Chat",
     "Mode": "greedy",

model_info.json

@@ -68,5 +68,10 @@
     "Llama-3.1-405B-Instruct-Turbo": {"pretty_name": "Llama-3.1-405B-Instruct-Turbo 🚨", "hf_model_id": "meta-llama/Meta-Llama-3.1-405B-Instruct"},
     "Mistral-Nemo-Instruct-2407": {"pretty_name": "Mistral-Nemo-Inst (12B) 🚨", "hf_model_id": "Mistral-Nemo-Instruct-2407"},
     "Phi-3-mini-4k-instruct": {"pretty_name": "Phi-3-mini-4k-instruct 🚨", "hf_model_id": "microsoft/Phi-3-mini-4k-instruct"},
-    "Athene-70B": {"pretty_name": "Athene-70B 🚨", "hf_model_id": "Nexusflow/Athene-70B"}
+    "Athene-70B": {"pretty_name": "Athene-70B 🚨", "hf_model_id": "Nexusflow/Athene-70B"},
+    "Mistral-Large-2": {"pretty_name": "Mistral-Large 2", "hf_model_id": "mistralai/Mistral-Large-Instruct-2407"},
+    "Meta-Llama-3.1-8B-Instruct": {"pretty_name": "Llama-3.1-8B-Instruct", "hf_model_id": "meta-llama/Meta-Llama-3.1-8B-Instruct"},
+    "Meta-Llama-3.1-70B-Instruct": {"pretty_name": "Llama-3.1-70B-Instruct", "hf_model_id": "meta-llama/Meta-Llama-3.1-70B-Instruct"},
+    "gemma-2-2b-it" : {"pretty_name": "Gemma-2-2B-it", "hf_model_id": "google/gemma-2-2b-it"},
+    "DeepSeek-Coder-V2-0724": {"pretty_name": "DeepSeek-Coder-V2-0724", "hf_model_id": "deepseek-ai/DeepSeek-Coder-V2-Instruct"}
 }