Upload eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v01.14/aimo_kaggle_hard_mmos/results_2024-05-24T10-16-49.966984.json with huggingface_hub

eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v01.14/aimo_kaggle_hard_mmos/results_2024-05-24T10-16-49.966984.json

@@ -0,0 +1,91 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 1579.416964497,
+    "end_time": 1958.123144323,
+    "total_evaluation_time_secondes": "378.7061798259999",
+    "model_name": "AI-MO/deepseek-math-7b-rl-sft",
+    "model_sha": "868636f68df95b6db35b4113753003b88836a86a",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_hard_mmos:v0|0": {
+      "qem": 0.02,
+      "qem_stderr": 0.02
+    },
+    "all": {
+      "qem": 0.02,
+      "qem_stderr": 0.02
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_hard_mmos:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_hard_mmos:v0": {
+      "name": "aimo_kaggle_hard_mmos:v0",
+      "prompt_function": "mmos_hard_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set-hard",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_code"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle_hard_mmos:v0|0": {
+      "hashes": {
+        "hash_examples": "b40b6a493a95bf77",
+        "hash_full_prompts": "bf4ae1359925ab3b",
+        "hash_input_tokens": "5f11181c1839f9c3",
+        "hash_cont_tokens": "0592c8e0d76ec685"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 35,
+      "non_padded": 15,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "79dbebcff6acad9e",
+      "hash_full_prompts": "86a15c8410a494ef",
+      "hash_input_tokens": "28eff78f8cb1ac9f",
+      "hash_cont_tokens": "6cc5093a5dfe5f31"
+    },
+    "truncated": 50,
+    "non_truncated": 0,
+    "padded": 35,
+    "non_padded": 15,
+    "num_truncated_few_shots": 0
+  }
+}