Upload eval_results/AI-MO/qwen2-72b-sft/aimo_v02.01/math_v3/results_2024-07-17T21-19-51.995991.json with huggingface_hub

eval_results/AI-MO/qwen2-72b-sft/aimo_v02.01/math_v3/results_2024-07-17T21-19-51.995991.json

@@ -0,0 +1,93 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 912024.815492996,
+    "end_time": 976756.250178339,
+    "total_evaluation_time_secondes": "64731.434685342945",
+    "model_name": "AI-MO/qwen2-72b-sft",
+    "model_sha": "7b12f762ff48feec2ec38e4f4b9aa29d198dc959",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "140.43 GB",
+    "config": null
+  },
+  "results": {
+    "custom|math_v3|0": {
+      "qem": 0.6656,
+      "qem_stderr": 0.006672648113498175
+    },
+    "all": {
+      "qem": 0.6656,
+      "qem_stderr": 0.006672648113498175
+    }
+  },
+  "versions": {
+    "custom|math_v3|0": 0
+  },
+  "config_tasks": {
+    "custom|math_v3": {
+      "name": "math_v3",
+      "prompt_function": "math_prompt_fn",
+      "hf_repo": "lighteval/MATH",
+      "hf_subset": "all",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test",
+        "validation"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 4096,
+      "stop_sequence": null,
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 5000,
+      "effective_num_docs": 5000,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|math_v3|0": {
+      "hashes": {
+        "hash_examples": "29c6c5377a3e08b1",
+        "hash_full_prompts": "7fdf8e1c67b26a69",
+        "hash_input_tokens": "da4039b23f029f8a",
+        "hash_cont_tokens": "9a098f677c3197e0"
+      },
+      "truncated": 4997,
+      "non_truncated": 3,
+      "padded": 442,
+      "non_padded": 4558,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "e637d4690488a07e",
+      "hash_full_prompts": "935aba4c46b23e3f",
+      "hash_input_tokens": "fb16db7b2fce3b22",
+      "hash_cont_tokens": "b5e43f9e3dbfdb8b"
+    },
+    "truncated": 4997,
+    "non_truncated": 3,
+    "padded": 442,
+    "non_padded": 4558,
+    "num_truncated_few_shots": 0
+  }
+}