Upload eval_results/mattshumer/Reflection-Llama-3.1-70B/main/ifeval/results_2024-09-06T08-57-20.145826.json with huggingface_hub

eval_results/mattshumer/Reflection-Llama-3.1-70B/main/ifeval/results_2024-09-06T08-57-20.145826.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 138.12653577,
+    "end_time": 11514.152557225,
+    "total_evaluation_time_secondes": "11376.026021455",
+    "model_name": "mattshumer/Reflection-Llama-3.1-70B",
+    "model_sha": "e7bfef9796a12f8cd072ba3b955cdf48854123e0",
+    "model_dtype": "torch.float32",
+    "model_size": "262.83 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.8243992606284658,
+      "prompt_level_strict_acc_stderr": 0.0163732573120579,
+      "inst_level_strict_acc": 0.882494004796163,
+      "inst_level_strict_acc_stderr": 0.00035181453333291124,
+      "prompt_level_loose_acc": 0.8595194085027726,
+      "prompt_level_loose_acc_stderr": 0.014953371656822773,
+      "inst_level_loose_acc": 0.907673860911271,
+      "inst_level_loose_acc_stderr": 0.0003201254240610877
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.8243992606284658,
+      "prompt_level_strict_acc_stderr": 0.0163732573120579,
+      "inst_level_strict_acc": 0.882494004796163,
+      "inst_level_strict_acc_stderr": 0.00035181453333291124,
+      "prompt_level_loose_acc": 0.8595194085027726,
+      "prompt_level_loose_acc_stderr": 0.014953371656822773,
+      "inst_level_loose_acc": 0.907673860911271,
+      "inst_level_loose_acc_stderr": 0.0003201254240610877
+    }
+  },
+  "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": "d8ad7bb31db18fbe",
+        "hash_input_tokens": "90b373ea9dcfae95",
+        "hash_cont_tokens": "567870ec2f0e58d1"
+      },
+      "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": "387e22caed47e5f2",
+      "hash_input_tokens": "84c96cb5db7a2f49",
+      "hash_cont_tokens": "ae75e53f7dabcda2"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 116,
+    "non_padded": 425,
+    "num_truncated_few_shots": 0
+  }
+}