Upload eval_results/edbeeching/mixtral-8x7b-instruct-v0.1_merged/v0.1/ifeval/results_2024-04-14T15-25-19.825529.json with huggingface_hub

eval_results/edbeeching/mixtral-8x7b-instruct-v0.1_merged/v0.1/ifeval/results_2024-04-14T15-25-19.825529.json

@@ -0,0 +1,101 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 4431438.417281177,
+    "end_time": 4444226.507357949,
+    "total_evaluation_time_secondes": "12788.090076772496",
+    "model_name": "edbeeching/mixtral-8x7b-instruct-v0.1_merged",
+    "model_sha": "6da964ddd4ac2f1edfa95d2a72082d87f1007cbf",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "87.49 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.2513863216266174,
+      "prompt_level_strict_acc_stderr": 0.01866821615224041,
+      "inst_level_strict_acc": 0.36211031175059955,
+      "inst_level_strict_acc_stderr": 0.0005081592010571681,
+      "prompt_level_loose_acc": 0.2754158964879852,
+      "prompt_level_loose_acc_stderr": 0.01922392319624201,
+      "inst_level_loose_acc": 0.38489208633093525,
+      "inst_level_loose_acc_stderr": 0.0005267464889068558
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.2513863216266174,
+      "prompt_level_strict_acc_stderr": 0.01866821615224041,
+      "inst_level_strict_acc": 0.36211031175059955,
+      "inst_level_strict_acc_stderr": 0.0005081592010571681,
+      "prompt_level_loose_acc": 0.2754158964879852,
+      "prompt_level_loose_acc_stderr": 0.01922392319624201,
+      "inst_level_loose_acc": 0.38489208633093525,
+      "inst_level_loose_acc_stderr": 0.0005267464889068558
+    }
+  },
+  "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,
+      "frozen": false,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "50f8a8f5a08077cd",
+        "hash_input_tokens": "fb1ed90207cf652b",
+        "hash_cont_tokens": "ff8432d6dd6a0d67"
+      },
+      "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": "9c04db679f2c9132",
+      "hash_input_tokens": "af9a63ef7d2de8ba",
+      "hash_cont_tokens": "8899b4eeeb8fc6bd"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 116,
+    "non_padded": 425,
+    "num_truncated_few_shots": 0
+  }
+}