Upload evaluation/HuggingFaceTB_SmolLM_135M_Instruct_lighteval|arc:easy|0|0_20250802_155120.json with huggingface_hub

evaluation/HuggingFaceTB_SmolLM_135M_Instruct_lighteval|arc:easy|0|0_20250802_155120.json

@@ -0,0 +1,140 @@
+{
+  "metadata": {
+    "model_name": "HuggingFaceTB/SmolLM-135M-Instruct",
+    "test_name": "lighteval|arc:easy|0|0",
+    "evaluation_timestamp": "2025-08-02T15:51:20.396829",
+    "lighteval_version": "latest",
+    "evaluation_type": "automated"
+  },
+  "results": {
+    "config_general": {
+      "lighteval_sha": "?",
+      "num_fewshot_seeds": 1,
+      "max_samples": 10,
+      "job_id": 0,
+      "start_time": 360616.06827779,
+      "end_time": 360689.209673871,
+      "total_evaluation_time_secondes": "73.14139608101686",
+      "model_name": "HuggingFaceTB/SmolLM-135M-Instruct",
+      "model_sha": "fcc320f490e08fdb4b99d935b2c58d40bf35b0d0",
+      "model_dtype": null,
+      "model_size": "513.13 MB",
+      "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": {
+      "lighteval|arc:easy|0": {
+        "acc": 0.5,
+        "acc_stderr": 0.16666666666666666,
+        "acc_norm": 0.6,
+        "acc_norm_stderr": 0.16329931618554522
+      },
+      "all": {
+        "acc": 0.5,
+        "acc_stderr": 0.16666666666666666,
+        "acc_norm": 0.6,
+        "acc_norm_stderr": 0.16329931618554522
+      }
+    },
+    "versions": {
+      "lighteval|arc:easy|0": 0
+    },
+    "config_tasks": {
+      "lighteval|arc:easy": {
+        "name": "arc:easy",
+        "prompt_function": "arc",
+        "hf_repo": "ai2_arc",
+        "hf_subset": "ARC-Easy",
+        "metric": [
+          {
+            "metric_name": "acc",
+            "higher_is_better": true,
+            "category": "8",
+            "use_case": "1",
+            "sample_level_fn": "compute",
+            "corpus_level_fn": "mean"
+          },
+          {
+            "metric_name": "acc_norm",
+            "higher_is_better": true,
+            "category": "8",
+            "use_case": "1",
+            "sample_level_fn": "compute",
+            "corpus_level_fn": "mean"
+          }
+        ],
+        "hf_revision": null,
+        "hf_filter": null,
+        "hf_avail_splits": [
+          "train",
+          "validation",
+          "test"
+        ],
+        "trust_dataset": true,
+        "evaluation_splits": [
+          "test"
+        ],
+        "few_shots_split": null,
+        "few_shots_select": "random_sampling_from_train",
+        "generation_size": 1,
+        "generation_grammar": null,
+        "stop_sequence": [
+          "\n"
+        ],
+        "num_samples": null,
+        "suite": [
+          "lighteval",
+          "arc"
+        ],
+        "original_num_docs": 2376,
+        "effective_num_docs": 10,
+        "must_remove_duplicate_docs": false,
+        "version": 0
+      }
+    },
+    "summary_tasks": {
+      "lighteval|arc:easy|0": {
+        "hashes": {
+          "hash_examples": "96119a3c29ba3e3c",
+          "hash_full_prompts": "96119a3c29ba3e3c",
+          "hash_input_tokens": "233d9be9e9923e44",
+          "hash_cont_tokens": "93e2ee3a693fac3a"
+        },
+        "truncated": 0,
+        "non_truncated": 10,
+        "padded": 19,
+        "non_padded": 21,
+        "effective_few_shots": 0.0,
+        "num_truncated_few_shots": 0
+      }
+    },
+    "summary_general": {
+      "hashes": {
+        "hash_examples": "9dff40a4ec5f045b",
+        "hash_full_prompts": "9dff40a4ec5f045b",
+        "hash_input_tokens": "cb3a354712244814",
+        "hash_cont_tokens": "5d3292e6655b4397"
+      },
+      "truncated": 0,
+      "non_truncated": 10,
+      "padded": 19,
+      "non_padded": 21,
+      "num_truncated_few_shots": 0
+    }
+  }
+}