Upload eval_results/AI-MO/deepseek-coder-33b-sft/aimo_v01.49/aimo_kaggle_pot/results_2024-05-07T19-53-43.494268.json with huggingface_hub

eval_results/AI-MO/deepseek-coder-33b-sft/aimo_v01.49/aimo_kaggle_pot/results_2024-05-07T19-53-43.494268.json

@@ -0,0 +1,89 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 17817.953549115,
+    "end_time": 18575.395140304,
+    "total_evaluation_time_secondes": "757.4415911889992",
+    "model_name": "AI-MO/deepseek-coder-33b-sft",
+    "model_sha": "667df132ed0d344f2579e2605a18ab5df0c13938",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "62.59 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_pot:v0|0": {
+      "qem": 0.0,
+      "qem_stderr": 0.0
+    },
+    "all": {
+      "qem": 0.0,
+      "qem_stderr": 0.0
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_pot:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_pot:v0": {
+      "name": "aimo_kaggle_pot:v0",
+      "prompt_function": "kaggle_pot_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_code_and_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_pot:v0|0": {
+      "hashes": {
+        "hash_examples": "631fc2e76ca58b0a",
+        "hash_full_prompts": "b1f5e7e9d5861f25",
+        "hash_input_tokens": "733e9d834e9a6357",
+        "hash_cont_tokens": "3021d8372dd7f43d"
+      },
+      "truncated": 48,
+      "non_truncated": 2,
+      "padded": 35,
+      "non_padded": 15,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "58058a251072a7cd",
+      "hash_full_prompts": "2471c91834f9103b",
+      "hash_input_tokens": "b5469fa2ae4e0166",
+      "hash_cont_tokens": "ee0df63d7ab98f93"
+    },
+    "truncated": 48,
+    "non_truncated": 2,
+    "padded": 35,
+    "non_padded": 15,
+    "num_truncated_few_shots": 0
+  }
+}