Upload eval_results/AI-MO/mistral-7b-sft/aimo_v01.17/mini_math/results_2024-04-24T11-28-37.114331.json with huggingface_hub

eval_results/AI-MO/mistral-7b-sft/aimo_v01.17/mini_math/results_2024-04-24T11-28-37.114331.json

@@ -0,0 +1,291 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 1024832.704354726,
+    "end_time": 1025780.851046388,
+    "total_evaluation_time_secondes": "948.1466916620266",
+    "model_name": "AI-MO/mistral-7b-sft",
+    "model_sha": "09397e28f34ceb95e0094c4633dbe945a4003160",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "custom|mini_math:level_1|0": {
+      "qem": 0.7142857142857143,
+      "qem_stderr": 0.07747516350666293
+    },
+    "custom|mini_math:level_2|0": {
+      "qem": 0.6111111111111112,
+      "qem_stderr": 0.05785537103478461
+    },
+    "custom|mini_math:level_3|0": {
+      "qem": 0.5111111111111111,
+      "qem_stderr": 0.05298680599073449
+    },
+    "custom|mini_math:level_4|0": {
+      "qem": 0.24742268041237114,
+      "qem_stderr": 0.04404125641916256
+    },
+    "custom|mini_math:level_5|0": {
+      "qem": 0.20754716981132076,
+      "qem_stderr": 0.03957769238377933
+    },
+    "custom|mini_math:_average|0": {
+      "qem": 0.45829555734632577,
+      "qem_stderr": 0.05438725786702479
+    },
+    "all": {
+      "qem": 0.45829555734632577,
+      "qem_stderr": 0.05438725786702479
+    }
+  },
+  "versions": {
+    "custom|mini_math:level_1|0": 0,
+    "custom|mini_math:level_2|0": 0,
+    "custom|mini_math:level_3|0": 0,
+    "custom|mini_math:level_4|0": 0,
+    "custom|mini_math:level_5|0": 0
+  },
+  "config_tasks": {
+    "custom|mini_math:level_1": {
+      "name": "mini_math:level_1",
+      "prompt_function": "math",
+      "hf_repo": "HuggingFaceH4/lighteval-mini-math",
+      "hf_subset": "Level 1",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom",
+        "mini_math"
+      ],
+      "original_num_docs": 35,
+      "effective_num_docs": 35,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    },
+    "custom|mini_math:level_2": {
+      "name": "mini_math:level_2",
+      "prompt_function": "math",
+      "hf_repo": "HuggingFaceH4/lighteval-mini-math",
+      "hf_subset": "Level 2",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom",
+        "mini_math"
+      ],
+      "original_num_docs": 72,
+      "effective_num_docs": 72,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    },
+    "custom|mini_math:level_3": {
+      "name": "mini_math:level_3",
+      "prompt_function": "math",
+      "hf_repo": "HuggingFaceH4/lighteval-mini-math",
+      "hf_subset": "Level 3",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom",
+        "mini_math"
+      ],
+      "original_num_docs": 90,
+      "effective_num_docs": 90,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    },
+    "custom|mini_math:level_4": {
+      "name": "mini_math:level_4",
+      "prompt_function": "math",
+      "hf_repo": "HuggingFaceH4/lighteval-mini-math",
+      "hf_subset": "Level 4",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom",
+        "mini_math"
+      ],
+      "original_num_docs": 97,
+      "effective_num_docs": 97,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    },
+    "custom|mini_math:level_5": {
+      "name": "mini_math:level_5",
+      "prompt_function": "math",
+      "hf_repo": "HuggingFaceH4/lighteval-mini-math",
+      "hf_subset": "Level 5",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom",
+        "mini_math"
+      ],
+      "original_num_docs": 106,
+      "effective_num_docs": 106,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "custom|mini_math:level_1|0": {
+      "hashes": {
+        "hash_examples": "f57f08853806c023",
+        "hash_full_prompts": "8a71f265c8375a44",
+        "hash_input_tokens": "b4df1247bfd0a371",
+        "hash_cont_tokens": "a136e621f3da8395"
+      },
+      "truncated": 35,
+      "non_truncated": 0,
+      "padded": 24,
+      "non_padded": 11,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_2|0": {
+      "hashes": {
+        "hash_examples": "b4fc7cab14532241",
+        "hash_full_prompts": "e342a96b9daa7407",
+        "hash_input_tokens": "97d62c81945518b5",
+        "hash_cont_tokens": "8f25a187fb5fe81c"
+      },
+      "truncated": 72,
+      "non_truncated": 0,
+      "padded": 37,
+      "non_padded": 35,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_3|0": {
+      "hashes": {
+        "hash_examples": "07877d35e91b6109",
+        "hash_full_prompts": "328b37ad43a0306b",
+        "hash_input_tokens": "5fcbe67e8c2f12ed",
+        "hash_cont_tokens": "98e1f0529d7f50fd"
+      },
+      "truncated": 90,
+      "non_truncated": 0,
+      "padded": 37,
+      "non_padded": 53,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_4|0": {
+      "hashes": {
+        "hash_examples": "016981666505c34d",
+        "hash_full_prompts": "b6bae5f0159adcc5",
+        "hash_input_tokens": "df1761cf4b7bc948",
+        "hash_cont_tokens": "dfda779d52b7a0ad"
+      },
+      "truncated": 96,
+      "non_truncated": 1,
+      "padded": 34,
+      "non_padded": 63,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_5|0": {
+      "hashes": {
+        "hash_examples": "80868d5b1c260707",
+        "hash_full_prompts": "948c55d532424241",
+        "hash_input_tokens": "180d722889a0d44c",
+        "hash_cont_tokens": "02470a095664445d"
+      },
+      "truncated": 105,
+      "non_truncated": 1,
+      "padded": 27,
+      "non_padded": 79,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "e7ee3a47ca2b0866",
+      "hash_full_prompts": "99b8a7fbe1090a26",
+      "hash_input_tokens": "1cc64bc07bdfd983",
+      "hash_cont_tokens": "da9a8b5b81a32845"
+    },
+    "truncated": 398,
+    "non_truncated": 2,
+    "padded": 159,
+    "non_padded": 241,
+    "num_truncated_few_shots": 0
+  }
+}