Upload eval_results/AI-MO/CodeLlama-13b-Python-hf-sft/aimo_v00.38/aimo_kaggle_medium/results_2024-06-01T05-45-58.088345.json with huggingface_hub

eval_results/AI-MO/CodeLlama-13b-Python-hf-sft/aimo_v00.38/aimo_kaggle_medium/results_2024-06-01T05-45-58.088345.json

@@ -0,0 +1,91 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 364271.342313331,
+    "end_time": 364740.294064388,
+    "total_evaluation_time_secondes": "468.95175105700037",
+    "model_name": "AI-MO/CodeLlama-13b-Python-hf-sft",
+    "model_sha": "79e0e3e0f2255747330582586a5be11350a3d89d",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "24.56 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_medium:v0|0": {
+      "qem": 0.175,
+      "qem_stderr": 0.06084343084444758
+    },
+    "all": {
+      "qem": 0.175,
+      "qem_stderr": 0.06084343084444758
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_medium:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_medium:v0": {
+      "name": "aimo_kaggle_medium:v0",
+      "prompt_function": "kaggle_medium_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set-medium",
+      "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,
+      "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:v0|0": {
+      "hashes": {
+        "hash_examples": "3401efda8b0cbcb5",
+        "hash_full_prompts": "14cb6646c78f810c",
+        "hash_input_tokens": "d1d016040eb15856",
+        "hash_cont_tokens": "7678f2382fb023f6"
+      },
+      "truncated": 40,
+      "non_truncated": 0,
+      "padded": 32,
+      "non_padded": 8,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "4c81d27cfdb9d737",
+      "hash_full_prompts": "0d4787c840fc98b7",
+      "hash_input_tokens": "7e52a8d7f3e15dcc",
+      "hash_cont_tokens": "2cc95a73c992c270"
+    },
+    "truncated": 40,
+    "non_truncated": 0,
+    "padded": 32,
+    "non_padded": 8,
+    "num_truncated_few_shots": 0
+  }
+}