Upload eval_results/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/main/math_500/results_2025-05-06T18-04-50.187799.json with huggingface_hub

eval_results/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/main/math_500/results_2025-05-06T18-04-50.187799.json

@@ -0,0 +1,126 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 3716751.807057023,
+    "end_time": 3717219.154532399,
+    "total_evaluation_time_secondes": "467.34747537598014",
+    "model_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
+    "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|math_500|0": {
+      "math_pass@1:1_samples": 0.83,
+      "math_pass@1:1_samples_stderr": 0.016815633531393436,
+      "math_pass@1:4_samples": 0.8305,
+      "math_pass@1:4_samples_stderr": 0.013440639925360967
+    },
+    "all": {
+      "math_pass@1:1_samples": 0.83,
+      "math_pass@1:1_samples_stderr": 0.016815633531393436,
+      "math_pass@1:4_samples": 0.8305,
+      "math_pass@1:4_samples_stderr": 0.013440639925360967
+    }
+  },
+  "versions": {
+    "lighteval|math_500|0": 2
+  },
+  "config_tasks": {
+    "lighteval|math_500": {
+      "name": "math_500",
+      "prompt_function": "math_500",
+      "hf_repo": "HuggingFaceH4/MATH-500",
+      "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"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "test"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 32768,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "lighteval"
+      ],
+      "original_num_docs": 500,
+      "effective_num_docs": 500,
+      "must_remove_duplicate_docs": false,
+      "version": 2
+    }
+  },
+  "summary_tasks": {
+    "lighteval|math_500|0": {
+      "hashes": {
+        "hash_examples": "adf0cc8311011db2",
+        "hash_full_prompts": "63c902dbdbaf1552",
+        "hash_input_tokens": "2af397a095a31139",
+        "hash_cont_tokens": "d57fe926906c5dd8"
+      },
+      "truncated": 0,
+      "non_truncated": 500,
+      "padded": 0,
+      "non_padded": 500,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "bfaad1993ff37a60",
+      "hash_full_prompts": "3ceaaade5cf43911",
+      "hash_input_tokens": "c663dbac8a64d3e4",
+      "hash_cont_tokens": "8fbdce32b7d1065e"
+    },
+    "truncated": 0,
+    "non_truncated": 500,
+    "padded": 0,
+    "non_padded": 500,
+    "num_truncated_few_shots": 0
+  }
+}