Upload eval_results/nltpt/Llama-3.2-3B-Instruct/main/ifeval/results_2024-09-23T10-37-21.557518.json with huggingface_hub

eval_results/nltpt/Llama-3.2-3B-Instruct/main/ifeval/results_2024-09-23T10-37-21.557518.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 317558.947078186,
+    "end_time": 318189.035049,
+    "total_evaluation_time_secondes": "630.0879708139692",
+    "model_name": "nltpt/Llama-3.2-3B-Instruct",
+    "model_sha": "276b29ce8303c9b88966a9b32fc75692dce4d8e1",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "5.98 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.7042513863216266,
+      "prompt_level_strict_acc_stderr": 0.019639405361071127,
+      "inst_level_strict_acc": 0.7877697841726619,
+      "inst_level_strict_acc_stderr": 0.000449022449444049,
+      "prompt_level_loose_acc": 0.7597042513863216,
+      "prompt_level_loose_acc_stderr": 0.01838647358148707,
+      "inst_level_loose_acc": 0.8285371702637889,
+      "inst_level_loose_acc_stderr": 0.00042116679892749024
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.7042513863216266,
+      "prompt_level_strict_acc_stderr": 0.019639405361071127,
+      "inst_level_strict_acc": 0.7877697841726619,
+      "inst_level_strict_acc_stderr": 0.000449022449444049,
+      "prompt_level_loose_acc": 0.7597042513863216,
+      "prompt_level_loose_acc_stderr": 0.01838647358148707,
+      "inst_level_loose_acc": 0.8285371702637889,
+      "inst_level_loose_acc_stderr": 0.00042116679892749024
+    }
+  },
+  "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": "6eb09a9bc7ba5d98"
+      },
+      "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": "f9a567c289772254"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 116,
+    "non_padded": 425,
+    "num_truncated_few_shots": 0
+  }
+}