Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v03.01/math/results_2024-04-26T04-51-02.678286.json with huggingface_hub

eval_results/AI-MO/deepseek-math-7b-sft/aimo_v03.01/math/results_2024-04-26T04-51-02.678286.json

@@ -0,0 +1,410 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 2,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 4213395.073455049,
+    "end_time": 4222017.114041344,
+    "total_evaluation_time_secondes": "8622.040586295538",
+    "model_name": "AI-MO/deepseek-math-7b-sft",
+    "model_sha": "dd01bf2509b1dede9c1f2afef533c04f5657c530",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|math:algebra|0": {
+      "qem": 0.6773378264532435,
+      "qem_stderr": 0.013574828385302717
+    },
+    "lighteval|math:counting_and_probability|0": {
+      "qem": 0.36075949367088606,
+      "qem_stderr": 0.022080579003443235
+    },
+    "lighteval|math:geometry|0": {
+      "qem": 0.348643006263048,
+      "qem_stderr": 0.02179646532161935
+    },
+    "lighteval|math:intermediate_algebra|0": {
+      "qem": 0.2469545957918051,
+      "qem_stderr": 0.014358724698059553
+    },
+    "lighteval|math:number_theory|0": {
+      "qem": 0.37962962962962965,
+      "qem_stderr": 0.020903123748455175
+    },
+    "lighteval|math:prealgebra|0": {
+      "qem": 0.6463834672789897,
+      "qem_stderr": 0.016208833993652156
+    },
+    "lighteval|math:precalculus|0": {
+      "qem": 0.29853479853479853,
+      "qem_stderr": 0.0196020862304134
+    },
+    "lighteval|math:_average|0": {
+      "qem": 0.42260611680320004,
+      "qem_stderr": 0.0183606630544208
+    },
+    "all": {
+      "qem": 0.42260611680320004,
+      "qem_stderr": 0.0183606630544208
+    }
+  },
+  "versions": {
+    "lighteval|math:algebra|0": 0,
+    "lighteval|math:counting_and_probability|0": 0,
+    "lighteval|math:geometry|0": 0,
+    "lighteval|math:intermediate_algebra|0": 0,
+    "lighteval|math:number_theory|0": 0,
+    "lighteval|math:prealgebra|0": 0,
+    "lighteval|math:precalculus|0": 0
+  },
+  "config_tasks": {
+    "lighteval|math:algebra": {
+      "name": "math:algebra",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "algebra",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 1187,
+      "effective_num_docs": 1187,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:counting_and_probability": {
+      "name": "math:counting_and_probability",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "counting_and_probability",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 474,
+      "effective_num_docs": 474,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:geometry": {
+      "name": "math:geometry",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "geometry",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 479,
+      "effective_num_docs": 479,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:intermediate_algebra": {
+      "name": "math:intermediate_algebra",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "intermediate_algebra",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 903,
+      "effective_num_docs": 903,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:number_theory": {
+      "name": "math:number_theory",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "number_theory",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 540,
+      "effective_num_docs": 540,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:prealgebra": {
+      "name": "math:prealgebra",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "prealgebra",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 871,
+      "effective_num_docs": 871,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    },
+    "lighteval|math:precalculus": {
+      "name": "math:precalculus",
+      "prompt_function": "math",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "precalculus",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "math"
+      ],
+      "original_num_docs": 546,
+      "effective_num_docs": 546,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "lighteval|math:algebra|0": {
+      "hashes": {
+        "hash_examples": "37a2fd2f076d2e49",
+        "hash_full_prompts": "461f2ab197cdfd01",
+        "hash_input_tokens": "bd3e731d75bc03e6",
+        "hash_cont_tokens": "8b12597582e2af8a"
+      },
+      "truncated": 1187,
+      "non_truncated": 0,
+      "padded": 63,
+      "non_padded": 1124,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:counting_and_probability|0": {
+      "hashes": {
+        "hash_examples": "97b4892e28bc078b",
+        "hash_full_prompts": "e2cc9eb28ca159b1",
+        "hash_input_tokens": "55728dc6558ad660",
+        "hash_cont_tokens": "ee728214399bbce1"
+      },
+      "truncated": 474,
+      "non_truncated": 0,
+      "padded": 17,
+      "non_padded": 457,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:geometry|0": {
+      "hashes": {
+        "hash_examples": "9e9f0228b8b3d093",
+        "hash_full_prompts": "e581f42fefeed8c9",
+        "hash_input_tokens": "6615f77877bf8cef",
+        "hash_cont_tokens": "3632ce692fc1b61c"
+      },
+      "truncated": 479,
+      "non_truncated": 0,
+      "padded": 60,
+      "non_padded": 419,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:intermediate_algebra|0": {
+      "hashes": {
+        "hash_examples": "cfe73a8e28ae94de",
+        "hash_full_prompts": "7557d30ba2cc15ef",
+        "hash_input_tokens": "608f04e5c215ffad",
+        "hash_cont_tokens": "a83dd8d0765c48b1"
+      },
+      "truncated": 903,
+      "non_truncated": 0,
+      "padded": 15,
+      "non_padded": 888,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:number_theory|0": {
+      "hashes": {
+        "hash_examples": "4ee5237cf144afac",
+        "hash_full_prompts": "7b983a6fe1e3d0a8",
+        "hash_input_tokens": "0a9b44cdfde7644d",
+        "hash_cont_tokens": "c1a5a1b43fc8ffb6"
+      },
+      "truncated": 539,
+      "non_truncated": 1,
+      "padded": 1,
+      "non_padded": 539,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:prealgebra|0": {
+      "hashes": {
+        "hash_examples": "3fb3afeb885f73d8",
+        "hash_full_prompts": "736a03165dbf8f49",
+        "hash_input_tokens": "434f9915c1b3fd21",
+        "hash_cont_tokens": "6cd2735b70a01c28"
+      },
+      "truncated": 871,
+      "non_truncated": 0,
+      "padded": 18,
+      "non_padded": 853,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "lighteval|math:precalculus|0": {
+      "hashes": {
+        "hash_examples": "753e25ab9ec4b46c",
+        "hash_full_prompts": "632ca5aa79f53b59",
+        "hash_input_tokens": "7808ce9bbfbc6f20",
+        "hash_cont_tokens": "29aee73acae487bc"
+      },
+      "truncated": 546,
+      "non_truncated": 0,
+      "padded": 4,
+      "non_padded": 542,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "b76099aa9092a203",
+      "hash_full_prompts": "2d2e105e1a0eeab9",
+      "hash_input_tokens": "ec60d48c4ec91205",
+      "hash_cont_tokens": "2362c0024ff536ef"
+    },
+    "truncated": 4999,
+    "non_truncated": 1,
+    "padded": 178,
+    "non_padded": 4822,
+    "num_truncated_few_shots": 0
+  }
+}