Upload eval_results/HuggingFaceTB/cosmo-300B-with-decay-instruct-mixture-5-bis/main/ifeval/results_2024-07-05T07-02-30.438091.json with huggingface_hub

eval_results/HuggingFaceTB/cosmo-300B-with-decay-instruct-mixture-5-bis/main/ifeval/results_2024-07-05T07-02-30.438091.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 483979.64869996,
+    "end_time": 484527.77310884,
+    "total_evaluation_time_secondes": "548.1244088800158",
+    "model_name": "HuggingFaceTB/cosmo-300B-with-decay-instruct-mixture-5-bis",
+    "model_sha": "9536ab96a03c2f896f7bb40e29eead1dfa1cf3ca",
+    "model_dtype": "torch.float32",
+    "model_size": "6.38 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.18114602587800369,
+      "prompt_level_strict_acc_stderr": 0.01657374894371447,
+      "inst_level_strict_acc": 0.2841726618705036,
+      "inst_level_strict_acc_stderr": 0.000488540668134031,
+      "prompt_level_loose_acc": 0.19408502772643252,
+      "prompt_level_loose_acc_stderr": 0.017019380550749405,
+      "inst_level_loose_acc": 0.30815347721822545,
+      "inst_level_loose_acc_stderr": 0.0004964674436048176
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.18114602587800369,
+      "prompt_level_strict_acc_stderr": 0.01657374894371447,
+      "inst_level_strict_acc": 0.2841726618705036,
+      "inst_level_strict_acc_stderr": 0.000488540668134031,
+      "prompt_level_loose_acc": 0.19408502772643252,
+      "prompt_level_loose_acc_stderr": 0.017019380550749405,
+      "inst_level_loose_acc": 0.30815347721822545,
+      "inst_level_loose_acc_stderr": 0.0004964674436048176
+    }
+  },
+  "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": "9ea7a9f2bbbb03ce",
+        "hash_cont_tokens": "f854ab63602dc1d1"
+      },
+      "truncated": 541,
+      "non_truncated": 0,
+      "padded": 114,
+      "non_padded": 427,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "856e0a5ca5494cd0",
+      "hash_cont_tokens": "1ee2532cc831aa3d"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 114,
+    "non_padded": 427,
+    "num_truncated_few_shots": 0
+  }
+}