Upload eval_results/alignment-handbook/zephyr-2b-gemma-dpo-mix6-beta-0.2-epoch-3/main/gsm8k/results_2024-03-06T21-40-47.463612.json with huggingface_hub

eval_results/alignment-handbook/zephyr-2b-gemma-dpo-mix6-beta-0.2-epoch-3/main/gsm8k/results_2024-03-06T21-40-47.463612.json

@@ -0,0 +1,88 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 1,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 6684809.749952891,
+    "end_time": 6685582.665684296,
+    "total_evaluation_time_secondes": "772.915731404908",
+    "model_name": "alignment-handbook/zephyr-2b-gemma-dpo-mix6-beta-0.2-epoch-3",
+    "model_sha": "5273fd27d513bd9296b56e388c26d9fb1fe6b3f6",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "4.68 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|gsm8k|5": {
+      "qem": 0.42077331311599697,
+      "qem_stderr": 0.013598489497182837
+    }
+  },
+  "versions": {
+    "lighteval|gsm8k|5": 0
+  },
+  "config_tasks": {
+    "lighteval|gsm8k": {
+      "name": "gsm8k",
+      "prompt_function": "gsm8k",
+      "hf_repo": "gsm8k",
+      "hf_subset": "main",
+      "metric": [
+        "quasi_exact_match_gsm8k"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": "random_sampling_from_train",
+      "generation_size": 256,
+      "stop_sequence": [
+        ":",
+        "Question:",
+        "Question"
+      ],
+      "output_regex": null,
+      "frozen": false,
+      "suite": [
+        "lighteval"
+      ],
+      "original_num_docs": 1319,
+      "effective_num_docs": 1319
+    }
+  },
+  "summary_tasks": {
+    "lighteval|gsm8k|5": {
+      "hashes": {
+        "hash_examples": "0ed016e24e7512fd",
+        "hash_full_prompts": "a9c6515c2ecd2d07",
+        "hash_input_tokens": "21ef709e0ca526ae",
+        "hash_cont_tokens": "206f527b0279c442"
+      },
+      "truncated": 0,
+      "non_truncated": 1319,
+      "padded": 0,
+      "non_padded": 1319,
+      "effective_few_shots": 5.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "bc71463e88551d0e",
+      "hash_full_prompts": "f49691ec139bfeac",
+      "hash_input_tokens": "4ae03e226a1b65a0",
+      "hash_cont_tokens": "290bcc37da8c6c91"
+    },
+    "truncated": 0,
+    "non_truncated": 1319,
+    "padded": 0,
+    "non_padded": 1319,
+    "num_truncated_few_shots": 0
+  }
+}