Upload eval_results/AI-MO/internlm-math-20b-sft/aimo_v03.06/aimo_kaggle_hard/results_2024-06-03T05-48-17.508440.json with huggingface_hub

eval_results/AI-MO/internlm-math-20b-sft/aimo_v03.06/aimo_kaggle_hard/results_2024-06-03T05-48-17.508440.json

@@ -0,0 +1,91 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 717042.150337203,
+    "end_time": 717673.226333095,
+    "total_evaluation_time_secondes": "631.075995891937",
+    "model_name": "AI-MO/internlm-math-20b-sft",
+    "model_sha": "46bcec0582aa600bec35d014b865a22bba2d4065",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "37.18 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_hard:v0|0": {
+      "qem": 0.06,
+      "qem_stderr": 0.0339266916772512
+    },
+    "all": {
+      "qem": 0.06,
+      "qem_stderr": 0.0339266916772512
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_hard:v0|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_hard:v0": {
+      "name": "aimo_kaggle_hard:v0",
+      "prompt_function": "kaggle_hard_prompt_fn",
+      "hf_repo": "AI-MO/kaggle-validation-set-hard",
+      "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": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle_hard:v0|0": {
+      "hashes": {
+        "hash_examples": "b40b6a493a95bf77",
+        "hash_full_prompts": "2c0f18269c13af34",
+        "hash_input_tokens": "adb5cc1d057c6f37",
+        "hash_cont_tokens": "7322fad852e5607f"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 33,
+      "non_padded": 17,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "79dbebcff6acad9e",
+      "hash_full_prompts": "739e6bdec3671fc6",
+      "hash_input_tokens": "9b53954644c2f2aa",
+      "hash_cont_tokens": "7daef97805f31db8"
+    },
+    "truncated": 50,
+    "non_truncated": 0,
+    "padded": 33,
+    "non_padded": 17,
+    "num_truncated_few_shots": 0
+  }
+}