Upload eval_results/open-r1/R1-Distill-Qwen-Math-7B/v00.31-step-000001104/aime24/results_2025-05-18T00-25-09.053710.json with huggingface_hub

eval_results/open-r1/R1-Distill-Qwen-Math-7B/v00.31-step-000001104/aime24/results_2025-05-18T00-25-09.053710.json

@@ -0,0 +1,174 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 4916989.842190158,
+    "end_time": 4922189.027193698,
+    "total_evaluation_time_secondes": "5199.185003540479",
+    "model_name": "open-r1/R1-Distill-Qwen-Math-7B",
+    "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@1:1_samples": 0.5,
+      "math_pass@1:1_samples_stderr": 0.09284766908852593,
+      "math_pass@1:4_samples": 0.44166666666666665,
+      "math_pass@1:4_samples_stderr": 0.07160248395699874,
+      "math_pass@1:8_samples": 0.45,
+      "math_pass@1:8_samples_stderr": 0.06901399158491531,
+      "math_pass@1:16_samples": 0.44166666666666665,
+      "math_pass@1:16_samples_stderr": 0.06447548466359414,
+      "math_pass@1:32_samples": 0.4510416666666667,
+      "math_pass@1:32_samples_stderr": 0.06428282188994272,
+      "math_pass@1:64_samples": 0.4682291666666668,
+      "math_pass@1:64_samples_stderr": 0.06436026142053597
+    },
+    "all": {
+      "math_pass@1:1_samples": 0.5,
+      "math_pass@1:1_samples_stderr": 0.09284766908852593,
+      "math_pass@1:4_samples": 0.44166666666666665,
+      "math_pass@1:4_samples_stderr": 0.07160248395699874,
+      "math_pass@1:8_samples": 0.45,
+      "math_pass@1:8_samples_stderr": 0.06901399158491531,
+      "math_pass@1:16_samples": 0.44166666666666665,
+      "math_pass@1:16_samples_stderr": 0.06447548466359414,
+      "math_pass@1:32_samples": 0.4510416666666667,
+      "math_pass@1:32_samples_stderr": 0.06428282188994272,
+      "math_pass@1:64_samples": 0.4682291666666668,
+      "math_pass@1:64_samples_stderr": 0.06436026142053597
+    }
+  },
+  "versions": {
+    "lighteval|aime24|0": 2
+  },
+  "config_tasks": {
+    "lighteval|aime24": {
+      "name": "aime24",
+      "prompt_function": "aime_prompt_fn",
+      "hf_repo": "HuggingFaceH4/aime_2024",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": "math_pass@1:1_samples",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "compute",
+          "corpus_level_fn": "mean"
+        },
+        {
+          "metric_name": "math_pass@1:4_samples",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "compute",
+          "corpus_level_fn": "mean"
+        },
+        {
+          "metric_name": "math_pass@1:8_samples",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "compute",
+          "corpus_level_fn": "mean"
+        },
+        {
+          "metric_name": "math_pass@1:16_samples",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "compute",
+          "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"
+        },
+        {
+          "metric_name": "math_pass@1:64_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": 2
+    }
+  },
+  "summary_tasks": {
+    "lighteval|aime24|0": {
+      "hashes": {
+        "hash_examples": "ddec8fc79d0a014b",
+        "hash_full_prompts": "ada2df8534791445",
+        "hash_input_tokens": "ea0bf6f2f7c52fee",
+        "hash_cont_tokens": "26450a4ce7091128"
+      },
+      "truncated": 0,
+      "non_truncated": 30,
+      "padded": 0,
+      "non_padded": 30,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "c903e836a519cf98",
+      "hash_full_prompts": "bdf2890257d63d76",
+      "hash_input_tokens": "1c28b529df7e2289",
+      "hash_cont_tokens": "aa976877eddea103"
+    },
+    "truncated": 0,
+    "non_truncated": 30,
+    "padded": 0,
+    "non_padded": 30,
+    "num_truncated_few_shots": 0
+  }
+}