Upload eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v01.01/aimo_kaggle_medium_mmos/results_2024-05-23T04-13-50.852923.json with huggingface_hub

eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v01.01/aimo_kaggle_medium_mmos/results_2024-05-23T04-13-50.852923.json

@@ -0,0 +1,91 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 133048.53369926,
+    "end_time": 133320.963079685,
+    "total_evaluation_time_secondes": "272.4293804249901",
+    "model_name": "AI-MO/deepseek-math-7b-rl-sft",
+    "model_sha": "0cb1ab78a754c4ae62a75ecc2efdc8a8f681c7ca",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_medium_mmos:v0|0": {
+      "qem": 0.0,
+      "qem_stderr": 0.0
+    },
+    "all": {
+      "qem": 0.0,
+      "qem_stderr": 0.0
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_medium_mmos:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_medium_mmos:v0": {
+      "name": "aimo_kaggle_medium_mmos:v0",
+      "prompt_function": "mmos_medium_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set-medium",
+      "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": 40,
+      "effective_num_docs": 40,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle_medium_mmos:v0|0": {
+      "hashes": {
+        "hash_examples": "8cdb484db7e0542d",
+        "hash_full_prompts": "914c5f932a4a2697",
+        "hash_input_tokens": "4e66db7f99151096",
+        "hash_cont_tokens": "3162fa02879e35ff"
+      },
+      "truncated": 40,
+      "non_truncated": 0,
+      "padded": 31,
+      "non_padded": 9,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "65639dcd12e109d8",
+      "hash_full_prompts": "6623987d75ea08e2",
+      "hash_input_tokens": "63956da5c26c3543",
+      "hash_cont_tokens": "ae7e70fe0aa8f5df"
+    },
+    "truncated": 40,
+    "non_truncated": 0,
+    "padded": 31,
+    "non_padded": 9,
+    "num_truncated_few_shots": 0
+  }
+}