Upload eval_results/nltpt/Llama-3.2-1B-Instruct/main/ifeval/results_2024-09-23T10-31-42.148731.json with huggingface_hub

eval_results/nltpt/Llama-3.2-1B-Instruct/main/ifeval/results_2024-09-23T10-31-42.148731.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 197828.427049391,
+    "end_time": 198201.554783339,
+    "total_evaluation_time_secondes": "373.12773394800024",
+    "model_name": "nltpt/Llama-3.2-1B-Instruct",
+    "model_sha": "d0a2081ed47e20ce524e8bc5d132f3fad2f69ff0",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "2.3 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.5064695009242144,
+      "prompt_level_strict_acc_stderr": 0.021514772941787053,
+      "inst_level_strict_acc": 0.6330935251798561,
+      "inst_level_strict_acc_stderr": 0.0005319761137601379,
+      "prompt_level_loose_acc": 0.5471349353049908,
+      "prompt_level_loose_acc_stderr": 0.02142075394952955,
+      "inst_level_loose_acc": 0.670263788968825,
+      "inst_level_loose_acc_stderr": 0.0005119328678858147
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.5064695009242144,
+      "prompt_level_strict_acc_stderr": 0.021514772941787053,
+      "inst_level_strict_acc": 0.6330935251798561,
+      "inst_level_strict_acc_stderr": 0.0005319761137601379,
+      "prompt_level_loose_acc": 0.5471349353049908,
+      "prompt_level_loose_acc_stderr": 0.02142075394952955,
+      "inst_level_loose_acc": 0.670263788968825,
+      "inst_level_loose_acc_stderr": 0.0005119328678858147
+    }
+  },
+  "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": "30fe3a1c6acd1de0",
+        "hash_input_tokens": "3a24593833f0f330",
+        "hash_cont_tokens": "3eda9b66731dffd0"
+      },
+      "truncated": 541,
+      "non_truncated": 0,
+      "padded": 116,
+      "non_padded": 425,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "bcd0aa2470bb7894",
+      "hash_input_tokens": "d0cc6cc5a4093ee7",
+      "hash_cont_tokens": "39c43bd8dd22efeb"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 116,
+    "non_padded": 425,
+    "num_truncated_few_shots": 0
+  }
+}