Upload eval_results/deepseek-ai/deepseek-llm-67b-chat/main/arc/results_2024-03-04T22-39-16.123106.json with huggingface_hub

eval_results/deepseek-ai/deepseek-llm-67b-chat/main/arc/results_2024-03-04T22-39-16.123106.json

@@ -0,0 +1,90 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 1,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 959569.062930491,
+    "end_time": 961046.050735199,
+    "total_evaluation_time_secondes": "1476.9878047079546",
+    "model_name": "deepseek-ai/deepseek-llm-67b-chat",
+    "model_sha": "79648bef7658bb824e4630740f6e1484c1b0620b",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "125.78 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|arc:challenge|25": {
+      "acc": 0.6220136518771331,
+      "acc_stderr": 0.014169664520303096,
+      "acc_norm": 0.6416382252559727,
+      "acc_norm_stderr": 0.01401288333485986
+    }
+  },
+  "versions": {
+    "lighteval|arc:challenge|25": 0
+  },
+  "config_tasks": {
+    "lighteval|arc:challenge": {
+      "name": "arc:challenge",
+      "prompt_function": "arc",
+      "hf_repo": "ai2_arc",
+      "hf_subset": "ARC-Challenge",
+      "metric": [
+        "loglikelihood_acc",
+        "loglikelihood_acc_norm_nospace"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": "random_sampling_from_train",
+      "generation_size": 1,
+      "stop_sequence": [
+        "\n"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval",
+        "arc"
+      ],
+      "original_num_docs": 1172,
+      "effective_num_docs": 1172
+    }
+  },
+  "summary_tasks": {
+    "lighteval|arc:challenge|25": {
+      "hashes": {
+        "hash_examples": "17b0cae357c0259e",
+        "hash_full_prompts": "c082a16eb385b492",
+        "hash_input_tokens": "cef91249e730ce02",
+        "hash_cont_tokens": "960d43a636d54ff0"
+      },
+      "truncated": 0,
+      "non_truncated": 1172,
+      "padded": 4634,
+      "non_padded": 53,
+      "effective_few_shots": 25.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "aaa6929c6d3771fb",
+      "hash_full_prompts": "3192b72b443466b8",
+      "hash_input_tokens": "a08646ebddfa6407",
+      "hash_cont_tokens": "39644d69f18b4d0f"
+    },
+    "truncated": 0,
+    "non_truncated": 1172,
+    "padded": 4634,
+    "non_padded": 53,
+    "num_truncated_few_shots": 0
+  }
+}