Upload eval_results/HuggingFaceH4/starcoder2-15b-dpo/v0.0/ifeval/results_2024-03-04T10-01-01.237692.json with huggingface_hub

eval_results/HuggingFaceH4/starcoder2-15b-dpo/v0.0/ifeval/results_2024-03-04T10-01-01.237692.json

@@ -0,0 +1,89 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 6548014.058607426,
+    "end_time": 6548902.148776286,
+    "total_evaluation_time_secondes": "888.090168860741",
+    "model_name": "HuggingFaceH4/starcoder2-15b-dpo",
+    "model_sha": "7e65c4b7f1eb10647c602cdb2b1dec732fe8a8c6",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "30.04 GB",
+    "config": null
+  },
+  "results": {
+    "custom|ifeval|0": {
+      "prompt_level_strict_acc": 0.27911275415896486,
+      "prompt_level_strict_acc_stderr": 0.019303080958497212,
+      "inst_level_strict_acc": 0.4136690647482014,
+      "inst_level_strict_acc_stderr": 0.0005112608532958294,
+      "prompt_level_loose_acc": 0.2957486136783734,
+      "prompt_level_loose_acc_stderr": 0.019639405361071123,
+      "inst_level_loose_acc": 0.43285371702637887,
+      "inst_level_loose_acc_stderr": 0.0005082264283109767
+    }
+  },
+  "versions": {
+    "custom|ifeval|0": 0
+  },
+  "config_tasks": {
+    "custom|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "wis-k/instruction-following-eval",
+      "hf_subset": "default",
+      "metric": [
+        "ifeval_metric"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "stop_sequence": [],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541
+    }
+  },
+  "summary_tasks": {
+    "custom|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "dc4c9014dffc6917",
+        "hash_cont_tokens": "c1c8e88c73d10834"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "5d5e6158af678a01",
+      "hash_cont_tokens": "8f40df57d6d398eb"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}