Upload eval_results/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/main/lcb/results_2025-04-30T16-41-17.470773.json with huggingface_hub

eval_results/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/main/lcb/results_2025-04-30T16-41-17.470773.json

@@ -0,0 +1,114 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 2612624.477873324,
+    "end_time": 2646732.051921242,
+    "total_evaluation_time_secondes": "34107.574047917966",
+    "model_name": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
+    "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": {
+    "extended|lcb:codegeneration|0": {
+      "codegen_pass@1:16": 0.558768656716418,
+      "codegen_pass@1:16_stderr": 0.026498963097323892
+    },
+    "all": {
+      "codegen_pass@1:16": 0.558768656716418,
+      "codegen_pass@1:16_stderr": 0.026498963097323892
+    }
+  },
+  "versions": {
+    "extended|lcb:codegeneration|0": 0
+  },
+  "config_tasks": {
+    "extended|lcb:codegeneration": {
+      "name": "lcb:codegeneration",
+      "prompt_function": "lcb_codegeneration_prompt_fn",
+      "hf_repo": "livecodebench/code_generation_lite",
+      "hf_subset": "v4_v5",
+      "metric": [
+        {
+          "metric_name": "codegen_pass@1:16",
+          "higher_is_better": true,
+          "category": "5",
+          "use_case": "6",
+          "sample_level_fn": "codegen_metric",
+          "corpus_level_fn": "mean"
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "test"
+      ],
+      "trust_dataset": true,
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 32768,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 268,
+      "effective_num_docs": 268,
+      "must_remove_duplicate_docs": false,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "extended|lcb:codegeneration|0": {
+      "hashes": {
+        "hash_examples": "8db44696993f602e",
+        "hash_full_prompts": "16619eb15cd731c4",
+        "hash_input_tokens": "bda98c7e7669e622",
+        "hash_cont_tokens": "8136c270478bc8fd"
+      },
+      "truncated": 0,
+      "non_truncated": 268,
+      "padded": 0,
+      "non_padded": 268,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "cc556c609442b994",
+      "hash_full_prompts": "1da488ee9ead0424",
+      "hash_input_tokens": "fd374270aa5ff841",
+      "hash_cont_tokens": "c90a3f353f737a03"
+    },
+    "truncated": 0,
+    "non_truncated": 268,
+    "padded": 0,
+    "non_padded": 268,
+    "num_truncated_few_shots": 0
+  }
+}