Upload eval_results/AI-MO/internlm-math-20b-sft/aimo_v02.36/aimo_kaggle_pot/results_2024-05-06T16-45-12.716703.json with huggingface_hub

eval_results/AI-MO/internlm-math-20b-sft/aimo_v02.36/aimo_kaggle_pot/results_2024-05-06T16-45-12.716703.json

@@ -0,0 +1,89 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 270749.063980889,
+    "end_time": 271089.811381019,
+    "total_evaluation_time_secondes": "340.74740013002884",
+    "model_name": "AI-MO/internlm-math-20b-sft",
+    "model_sha": "b52b95301b384d5c74a8254652e8c1f718d74819",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "37.18 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_pot:v0|0": {
+      "qem": 0.18,
+      "qem_stderr": 0.0548839220351387
+    },
+    "all": {
+      "qem": 0.18,
+      "qem_stderr": 0.0548839220351387
+    }
+  },
+  "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": "6fd5c377eb83a250",
+        "hash_cont_tokens": "d8feab969eb2be57"
+      },
+      "truncated": 49,
+      "non_truncated": 1,
+      "padded": 36,
+      "non_padded": 14,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "58058a251072a7cd",
+      "hash_full_prompts": "2471c91834f9103b",
+      "hash_input_tokens": "7908adb22f000b27",
+      "hash_cont_tokens": "229c3b2ee5dd61dc"
+    },
+    "truncated": 49,
+    "non_truncated": 1,
+    "padded": 36,
+    "non_padded": 14,
+    "num_truncated_few_shots": 0
+  }
+}