Upload eval_results/open-r1/R1-Zero-Qwen-7B-Math/v04.10-step-000000315/aime24/results_2025-04-12T13-15-20.432867.json with huggingface_hub

eval_results/open-r1/R1-Zero-Qwen-7B-Math/v04.10-step-000000315/aime24/results_2025-04-12T13-15-20.432867.json

@@ -0,0 +1,127 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": -1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 1449348.508015334,
+    "end_time": 1450512.589093624,
+    "total_evaluation_time_secondes": "1164.081078290008",
+    "model_name": "open-r1/R1-Zero-Qwen-7B-Math",
+    "model_sha": "",
+    "model_dtype": null,
+    "model_size": null,
+    "generation_parameters": {
+      "early_stopping": null,
+      "repetition_penalty": null,
+      "frequency_penalty": null,
+      "length_penalty": null,
+      "presence_penalty": null,
+      "max_new_tokens": 32768,
+      "min_new_tokens": null,
+      "seed": null,
+      "stop_tokens": null,
+      "temperature": 0.6,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "lighteval|aime24|0": {
+      "math_pass@1:32_samples": 0.12395833333333332,
+      "math_pass@1:32_samples_stderr": 0.048400310930613616,
+      "extractive_match": 0.1,
+      "extractive_match_stderr": 0.055708601453115555
+    },
+    "all": {
+      "math_pass@1:32_samples": 0.12395833333333332,
+      "math_pass@1:32_samples_stderr": 0.048400310930613616,
+      "extractive_match": 0.1,
+      "extractive_match_stderr": 0.055708601453115555
+    }
+  },
+  "versions": {
+    "lighteval|aime24|0": 1
+  },
+  "config_tasks": {
+    "lighteval|aime24": {
+      "name": "aime24",
+      "prompt_function": "aime_prompt_fn",
+      "hf_repo": "HuggingFaceH4/aime_2024",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": "extractive_match",
+          "higher_is_better": true,
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "sample_level_fn",
+          "corpus_level_fn": "mean"
+        },
+        {
+          "metric_name": "math_pass@1:32_samples",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "compute",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 32768,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "lighteval"
+      ],
+      "original_num_docs": 30,
+      "effective_num_docs": 30,
+      "must_remove_duplicate_docs": false,
+      "version": 1
+    }
+  },
+  "summary_tasks": {
+    "lighteval|aime24|0": {
+      "hashes": {
+        "hash_examples": "34f73561248f264c",
+        "hash_full_prompts": "bc5bcea127c57fb8",
+        "hash_input_tokens": "8d44aeb2039ae950",
+        "hash_cont_tokens": "9f18fbb8f60e40ef"
+      },
+      "truncated": 0,
+      "non_truncated": 60,
+      "padded": 0,
+      "non_padded": 60,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "772678c254b88603",
+      "hash_full_prompts": "ce17217291012a57",
+      "hash_input_tokens": "191b20d776c281b9",
+      "hash_cont_tokens": "a6968d36c504a7b4"
+    },
+    "truncated": 0,
+    "non_truncated": 60,
+    "padded": 0,
+    "non_padded": 60,
+    "num_truncated_few_shots": 0
+  }
+}