Upload eval_results/HuggingFaceH4/llama-3.1-8b-sft/v00.00/ifeval/results_2024-09-19T06-11-20.821470.json with huggingface_hub

eval_results/HuggingFaceH4/llama-3.1-8b-sft/v00.00/ifeval/results_2024-09-19T06-11-20.821470.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 887388.054498023,
+    "end_time": 888189.270848402,
+    "total_evaluation_time_secondes": "801.2163503790507",
+    "model_name": "HuggingFaceH4/llama-3.1-8b-sft",
+    "model_sha": "e87ca81e0895629c16ad59d5c98f023d93723ff2",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "14.96 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.4805914972273567,
+      "prompt_level_strict_acc_stderr": 0.021500357879025184,
+      "inst_level_strict_acc": 0.580335731414868,
+      "inst_level_strict_acc_stderr": 0.000540862176593818,
+      "prompt_level_loose_acc": 0.5656192236598891,
+      "prompt_level_loose_acc_stderr": 0.021330473657564724,
+      "inst_level_loose_acc": 0.6534772182254197,
+      "inst_level_loose_acc_stderr": 0.000517972516038633
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.4805914972273567,
+      "prompt_level_strict_acc_stderr": 0.021500357879025184,
+      "inst_level_strict_acc": 0.580335731414868,
+      "inst_level_strict_acc_stderr": 0.000540862176593818,
+      "prompt_level_loose_acc": 0.5656192236598891,
+      "prompt_level_loose_acc_stderr": 0.021330473657564724,
+      "inst_level_loose_acc": 0.6534772182254197,
+      "inst_level_loose_acc_stderr": 0.000517972516038633
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": 0
+  },
+  "config_tasks": {
+    "extended|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,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "ca4a056d1ce1b77c",
+        "hash_cont_tokens": "28cd7add08fcb7ed"
+      },
+      "truncated": 541,
+      "non_truncated": 0,
+      "padded": 115,
+      "non_padded": 426,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "05d4422f76889c35",
+      "hash_cont_tokens": "a5d34395e3b6df54"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 115,
+    "non_padded": 426,
+    "num_truncated_few_shots": 0
+  }
+}