Upload eval_results/HuggingFaceH4/mistral-7b-odpo/v1.23/arc/results_2024-03-29T12-41-33.505198.json with huggingface_hub

eval_results/HuggingFaceH4/mistral-7b-odpo/v1.23/arc/results_2024-03-29T12-41-33.505198.json

@@ -0,0 +1,91 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 1,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 314916.129594765,
+    "end_time": 315044.595673064,
+    "total_evaluation_time_secondes": "128.4660782989813",
+    "model_name": "HuggingFaceH4/mistral-7b-odpo",
+    "model_sha": "266a7c8b41ae56c74b2bf55512a27f002dea11b4",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "13.99 GB",
+    "config": null
+  },
+  "results": {
+    "leaderboard|arc:challenge|25": {
+      "acc": 0.5998293515358362,
+      "acc_stderr": 0.014317197787809172,
+      "acc_norm": 0.6271331058020477,
+      "acc_norm_stderr": 0.01413117676013117
+    }
+  },
+  "versions": {
+    "leaderboard|arc:challenge|25": 0
+  },
+  "config_tasks": {
+    "leaderboard|arc:challenge": {
+      "name": "arc:challenge",
+      "prompt_function": "arc",
+      "hf_repo": "ai2_arc",
+      "hf_subset": "ARC-Challenge",
+      "metric": [
+        "loglikelihood_acc",
+        "loglikelihood_acc_norm_nospace"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": "random_sampling_from_train",
+      "generation_size": 1,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "leaderboard",
+        "arc"
+      ],
+      "original_num_docs": 1172,
+      "effective_num_docs": 1172,
+      "trust_dataset": true
+    }
+  },
+  "summary_tasks": {
+    "leaderboard|arc:challenge|25": {
+      "hashes": {
+        "hash_examples": "17b0cae357c0259e",
+        "hash_full_prompts": "f42cdfcbcc9b6fa4",
+        "hash_input_tokens": "bedd26d7d7ad2546",
+        "hash_cont_tokens": "8b6f018bc80a3417"
+      },
+      "truncated": 0,
+      "non_truncated": 1172,
+      "padded": 4660,
+      "non_padded": 27,
+      "effective_few_shots": 25.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "aaa6929c6d3771fb",
+      "hash_full_prompts": "099f2d4c459a2c1d",
+      "hash_input_tokens": "7eee0cf412c3eaf0",
+      "hash_cont_tokens": "6857c42b72c1a62a"
+    },
+    "truncated": 0,
+    "non_truncated": 1172,
+    "padded": 4660,
+    "non_padded": 27,
+    "num_truncated_few_shots": 0
+  }
+}