Upload eval_results/Qwen/Qwen2.5-Math-7B-Instruct/main/mini_math/results_2025-01-23T15-02-00.811603.json with huggingface_hub

eval_results/Qwen/Qwen2.5-Math-7B-Instruct/main/mini_math/results_2025-01-23T15-02-00.811603.json

@@ -0,0 +1,335 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": -1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 389746.081182226,
+    "end_time": 389875.173432725,
+    "total_evaluation_time_secondes": "129.09225049900124",
+    "model_name": "Qwen/Qwen2.5-Math-7B-Instruct",
+    "model_sha": "",
+    "model_dtype": null,
+    "model_size": null
+  },
+  "results": {
+    "custom|mini_math:level_1|0": {
+      "extractive_match": 0.8571428571428571,
+      "extractive_match_stderr": 0.06001200360120039
+    },
+    "custom|mini_math:level_2|0": {
+      "extractive_match": 0.6527777777777778,
+      "extractive_match_stderr": 0.05650114676852965
+    },
+    "custom|mini_math:level_3|0": {
+      "extractive_match": 0.6888888888888889,
+      "extractive_match_stderr": 0.049072405533864084
+    },
+    "custom|mini_math:level_4|0": {
+      "extractive_match": 0.5463917525773195,
+      "extractive_match_stderr": 0.05081090406399051
+    },
+    "custom|mini_math:level_5|0": {
+      "extractive_match": 0.330188679245283,
+      "extractive_match_stderr": 0.045894715469579954
+    },
+    "custom|mini_math:_average|0": {
+      "extractive_match": 0.6150779911264251,
+      "extractive_match_stderr": 0.052458235087432914
+    },
+    "all": {
+      "extractive_match": 0.6150779911264251,
+      "extractive_match_stderr": 0.052458235087432914
+    }
+  },
+  "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_prompt_fn",
+      "hf_repo": "AI-MO/lighteval-mini-math",
+      "hf_subset": "Level 1",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 35,
+      "effective_num_docs": 35,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    },
+    "custom|mini_math:level_2": {
+      "name": "mini_math:level_2",
+      "prompt_function": "math_prompt_fn",
+      "hf_repo": "AI-MO/lighteval-mini-math",
+      "hf_subset": "Level 2",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 72,
+      "effective_num_docs": 72,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    },
+    "custom|mini_math:level_3": {
+      "name": "mini_math:level_3",
+      "prompt_function": "math_prompt_fn",
+      "hf_repo": "AI-MO/lighteval-mini-math",
+      "hf_subset": "Level 3",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 90,
+      "effective_num_docs": 90,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    },
+    "custom|mini_math:level_4": {
+      "name": "mini_math:level_4",
+      "prompt_function": "math_prompt_fn",
+      "hf_repo": "AI-MO/lighteval-mini-math",
+      "hf_subset": "Level 4",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 97,
+      "effective_num_docs": 97,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    },
+    "custom|mini_math:level_5": {
+      "name": "mini_math:level_5",
+      "prompt_function": "math_prompt_fn",
+      "hf_repo": "AI-MO/lighteval-mini-math",
+      "hf_subset": "Level 5",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 106,
+      "effective_num_docs": 106,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|mini_math:level_1|0": {
+      "hashes": {
+        "hash_examples": "ad7e0d89fb7b0664",
+        "hash_full_prompts": "b80536c069eb4029",
+        "hash_input_tokens": "b7427696025a0367",
+        "hash_cont_tokens": "c299cf9e4b97d0c0"
+      },
+      "truncated": 0,
+      "non_truncated": 35,
+      "padded": 0,
+      "non_padded": 35,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_2|0": {
+      "hashes": {
+        "hash_examples": "493b7e3130e3a50d",
+        "hash_full_prompts": "46446bf68057d29e",
+        "hash_input_tokens": "0e7973761858265b",
+        "hash_cont_tokens": "aa155389ff5e9e3e"
+      },
+      "truncated": 0,
+      "non_truncated": 72,
+      "padded": 0,
+      "non_padded": 72,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_3|0": {
+      "hashes": {
+        "hash_examples": "11edfc7fc00756f4",
+        "hash_full_prompts": "dfffd64b4ba2e549",
+        "hash_input_tokens": "0f6dd65ec1cae39f",
+        "hash_cont_tokens": "60b282155e01fead"
+      },
+      "truncated": 0,
+      "non_truncated": 90,
+      "padded": 0,
+      "non_padded": 90,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_4|0": {
+      "hashes": {
+        "hash_examples": "a901e1669616ae50",
+        "hash_full_prompts": "2e1f3a5e7d370bb0",
+        "hash_input_tokens": "f3b8792222ecd28e",
+        "hash_cont_tokens": "8121eb5755ea0d39"
+      },
+      "truncated": 0,
+      "non_truncated": 97,
+      "padded": 0,
+      "non_padded": 97,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|mini_math:level_5|0": {
+      "hashes": {
+        "hash_examples": "1d024f2e2410736e",
+        "hash_full_prompts": "386a3d414eaee221",
+        "hash_input_tokens": "09a513b7947e58d6",
+        "hash_cont_tokens": "0543220ba1936de3"
+      },
+      "truncated": 0,
+      "non_truncated": 106,
+      "padded": 0,
+      "non_padded": 106,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "06e45ee0bae45a44",
+      "hash_full_prompts": "f12515b05c0a349f",
+      "hash_input_tokens": "9b84bf8c84139e4f",
+      "hash_cont_tokens": "2aa56fa802e33fa4"
+    },
+    "truncated": 0,
+    "non_truncated": 400,
+    "padded": 0,
+    "non_padded": 400,
+    "num_truncated_few_shots": 0
+  }
+}