Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.43/aimo_kaggle/results_2024-04-26T00-15-00.009211.json with huggingface_hub

eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.43/aimo_kaggle/results_2024-04-26T00-15-00.009211.json

@@ -0,0 +1,89 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 591317.777299186,
+    "end_time": 591592.202243397,
+    "total_evaluation_time_secondes": "274.4249442110304",
+    "model_name": "AI-MO/deepseek-math-7b-sft",
+    "model_sha": "7384467f01d7770ba37c1cad7a97ef45a9da00cb",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle:v0|0": {
+      "qem": 0.14,
+      "qem_stderr": 0.049569575922564196
+    },
+    "all": {
+      "qem": 0.14,
+      "qem_stderr": 0.049569575922564196
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle:v0": {
+      "name": "aimo_kaggle:v0",
+      "prompt_function": "kaggle_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_math"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle:v0|0": {
+      "hashes": {
+        "hash_examples": "2a919e6b839e921a",
+        "hash_full_prompts": "e0424ade9e31a0fe",
+        "hash_input_tokens": "4002f9b277fa4d5e",
+        "hash_cont_tokens": "c2f4f0ed157e4a01"
+      },
+      "truncated": 47,
+      "non_truncated": 3,
+      "padded": 36,
+      "non_padded": 14,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "f59ec6c4e24c1226",
+      "hash_full_prompts": "35e1e25418ccb6f7",
+      "hash_input_tokens": "40119909fe745426",
+      "hash_cont_tokens": "0a20471ef0a14309"
+    },
+    "truncated": 47,
+    "non_truncated": 3,
+    "padded": 36,
+    "non_padded": 14,
+    "num_truncated_few_shots": 0
+  }
+}