Upload evaluation/Qwen_Qwen3_0.6B_leaderboard|truthfulqa:mc|0|0_20250729_131709.json with huggingface_hub

evaluation/Qwen_Qwen3_0.6B_leaderboard|truthfulqa:mc|0|0_20250729_131709.json

@@ -0,0 +1,138 @@
+{
+  "metadata": {
+    "model_name": "Qwen/Qwen3-0.6B",
+    "test_name": "leaderboard|truthfulqa:mc|0|0",
+    "evaluation_timestamp": "2025-07-29T13:17:09.801306",
+    "lighteval_version": "latest",
+    "evaluation_type": "automated"
+  },
+  "results": {
+    "config_general": {
+      "lighteval_sha": "?",
+      "num_fewshot_seeds": 1,
+      "max_samples": 10,
+      "job_id": 0,
+      "start_time": 5765.815531393,
+      "end_time": 5838.737527639,
+      "total_evaluation_time_secondes": "72.92199624600016",
+      "model_name": "Qwen/Qwen3-0.6B",
+      "model_sha": "c1899de289a04d12100db370d81485cdf75e47ca",
+      "model_dtype": null,
+      "model_size": "2.22 GB",
+      "generation_parameters": {
+        "early_stopping": null,
+        "repetition_penalty": null,
+        "frequency_penalty": null,
+        "length_penalty": null,
+        "presence_penalty": null,
+        "max_new_tokens": null,
+        "min_new_tokens": null,
+        "seed": null,
+        "stop_tokens": null,
+        "temperature": null,
+        "top_k": null,
+        "min_p": null,
+        "top_p": null,
+        "truncate_prompt": null,
+        "response_format": null
+      }
+    },
+    "results": {
+      "leaderboard|truthfulqa:mc|0": {
+        "truthfulqa_mc1": 0.4,
+        "truthfulqa_mc1_stderr": 0.16329931618554522,
+        "truthfulqa_mc2": 0.38767522181090863,
+        "truthfulqa_mc2_stderr": 0.13021084864495217
+      },
+      "all": {
+        "truthfulqa_mc1": 0.4,
+        "truthfulqa_mc1_stderr": 0.16329931618554522,
+        "truthfulqa_mc2": 0.38767522181090863,
+        "truthfulqa_mc2_stderr": 0.13021084864495217
+      }
+    },
+    "versions": {
+      "leaderboard|truthfulqa:mc|0": 0
+    },
+    "config_tasks": {
+      "leaderboard|truthfulqa:mc": {
+        "name": "truthfulqa:mc",
+        "prompt_function": "truthful_qa_multiple_choice",
+        "hf_repo": "truthful_qa",
+        "hf_subset": "multiple_choice",
+        "metric": [
+          {
+            "metric_name": [
+              "truthfulqa_mc1",
+              "truthfulqa_mc2"
+            ],
+            "higher_is_better": {
+              "truthfulqa_mc1": true,
+              "truthfulqa_mc2": true
+            },
+            "category": "8",
+            "use_case": "1",
+            "sample_level_fn": "truthfulqa_mc_metrics",
+            "corpus_level_fn": {
+              "truthfulqa_mc1": "mean",
+              "truthfulqa_mc2": "mean"
+            }
+          }
+        ],
+        "hf_revision": null,
+        "hf_filter": null,
+        "hf_avail_splits": [
+          "validation"
+        ],
+        "trust_dataset": true,
+        "evaluation_splits": [
+          "validation"
+        ],
+        "few_shots_split": null,
+        "few_shots_select": null,
+        "generation_size": -1,
+        "generation_grammar": null,
+        "stop_sequence": [
+          "\n"
+        ],
+        "num_samples": null,
+        "suite": [
+          "leaderboard"
+        ],
+        "original_num_docs": 817,
+        "effective_num_docs": 10,
+        "must_remove_duplicate_docs": false,
+        "version": 0
+      }
+    },
+    "summary_tasks": {
+      "leaderboard|truthfulqa:mc|0": {
+        "hashes": {
+          "hash_examples": "da0474ee913e995c",
+          "hash_full_prompts": "da0474ee913e995c",
+          "hash_input_tokens": "e91edf9d841d2695",
+          "hash_cont_tokens": "6c9c980152091c7a"
+        },
+        "truncated": 0,
+        "non_truncated": 10,
+        "padded": 101,
+        "non_padded": 18,
+        "effective_few_shots": 0.0,
+        "num_truncated_few_shots": 0
+      }
+    },
+    "summary_general": {
+      "hashes": {
+        "hash_examples": "d93946e268e80920",
+        "hash_full_prompts": "d93946e268e80920",
+        "hash_input_tokens": "2653f11512a69452",
+        "hash_cont_tokens": "899b57bbe39db181"
+      },
+      "truncated": 0,
+      "non_truncated": 10,
+      "padded": 101,
+      "non_padded": 18,
+      "num_truncated_few_shots": 0
+    }
+  }
+}