cogwheelhead commited on
Commit
bf42c1e
·
verified ·
1 Parent(s): ed6309a

chore: fix model repo names

Browse files
Files changed (1) hide show
  1. data/u_math_eval_results.json +6 -6
data/u_math_eval_results.json CHANGED
@@ -195,7 +195,7 @@
195
  ]
196
  },
197
  {
198
- "model_name": "Nexusflow/Athene-V2-72B-Chat",
199
  "judge_model_name": "gpt-4o-2024-08-06",
200
  "u_math": [
201
  54.90909090909091,
@@ -429,7 +429,7 @@
429
  ]
430
  },
431
  {
432
- "model_name": "mistralai/Ministral-8B-2410",
433
  "judge_model_name": "gpt-4o-2024-08-06",
434
  "u_math": [
435
  23.09090909090909,
@@ -468,7 +468,7 @@
468
  ]
469
  },
470
  {
471
- "model_name": "mistralai/Mistral-Large-2411",
472
  "judge_model_name": "gpt-4o-2024-08-06",
473
  "u_math": [
474
  47.63636363636364,
@@ -624,7 +624,7 @@
624
  ]
625
  },
626
  {
627
- "model_name": "mistralai/Mistral-Small-2501",
628
  "judge_model_name": "gpt-4o-2024-08-06",
629
  "u_math": [
630
  34.81818181818182,
@@ -975,7 +975,7 @@
975
  ]
976
  },
977
  {
978
- "model_name": "meta-llama/Llama-3.1-Nemotron-70B",
979
  "judge_model_name": "gpt-4o-2024-08-06",
980
  "u_math": [
981
  42.54545454545455,
@@ -1170,7 +1170,7 @@
1170
  ]
1171
  },
1172
  {
1173
- "model_name": "mistralai/Pixtral-Large-2411",
1174
  "judge_model_name": "gpt-4o-2024-08-06",
1175
  "u_math": [
1176
  47.81818181818182,
 
195
  ]
196
  },
197
  {
198
+ "model_name": "Nexusflow/Athene-V2-Chat",
199
  "judge_model_name": "gpt-4o-2024-08-06",
200
  "u_math": [
201
  54.90909090909091,
 
429
  ]
430
  },
431
  {
432
+ "model_name": "mistralai/Ministral-8B-Instruct-2410",
433
  "judge_model_name": "gpt-4o-2024-08-06",
434
  "u_math": [
435
  23.09090909090909,
 
468
  ]
469
  },
470
  {
471
+ "model_name": "mistralai/Mistral-Large-Instruct-2411",
472
  "judge_model_name": "gpt-4o-2024-08-06",
473
  "u_math": [
474
  47.63636363636364,
 
624
  ]
625
  },
626
  {
627
+ "model_name": "mistralai/Mistral-Small-Instruct-2501",
628
  "judge_model_name": "gpt-4o-2024-08-06",
629
  "u_math": [
630
  34.81818181818182,
 
975
  ]
976
  },
977
  {
978
+ "model_name": "nvidia/Llama-3.1-Nemotron-70B",
979
  "judge_model_name": "gpt-4o-2024-08-06",
980
  "u_math": [
981
  42.54545454545455,
 
1170
  ]
1171
  },
1172
  {
1173
+ "model_name": "mistralai/Pixtral-Large-Instruct-2411",
1174
  "judge_model_name": "gpt-4o-2024-08-06",
1175
  "u_math": [
1176
  47.81818181818182,