Add files using upload-large-folder tool

eval_results_avg4/global_step_5/aime24/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 30,
+    "num_scores": 120,
+    "timeout_samples": 0,
+    "empty_samples": 0,
+    "acc": 3.3,
+    "pass_acc": 6.7,
+    "pass@k": {
+        "1": 3.3,
+        "2": 5.0,
+        "4": 6.7
+    },
+    "time_use_in_second": 182.274742603302,
+    "time_use_in_minite": "3:02"
+}

eval_results_avg4/global_step_5/aime25/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 30,
+    "num_scores": 120,
+    "timeout_samples": 0,
+    "empty_samples": 0,
+    "acc": 3.3,
+    "pass_acc": 6.7,
+    "pass@k": {
+        "1": 3.3,
+        "2": 5.0,
+        "4": 6.7
+    },
+    "time_use_in_second": 181.38263416290283,
+    "time_use_in_minite": "3:01"
+}

eval_results_avg4/global_step_5/amc23/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 40,
+    "num_scores": 160,
+    "timeout_samples": 0,
+    "empty_samples": 1,
+    "acc": 29.4,
+    "pass_acc": 57.5,
+    "pass@k": {
+        "1": 29.4,
+        "2": 42.5,
+        "4": 57.5
+    },
+    "time_use_in_second": 188.3577175140381,
+    "time_use_in_minite": "3:08"
+}

eval_results_avg4/global_step_50/aime24/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 30,
+    "num_scores": 120,
+    "timeout_samples": 0,
+    "empty_samples": 0,
+    "acc": 6.7,
+    "pass_acc": 13.3,
+    "pass@k": {
+        "1": 6.7,
+        "2": 10.0,
+        "4": 13.3
+    },
+    "time_use_in_second": 175.23715782165527,
+    "time_use_in_minite": "2:55"
+}

eval_results_avg4/global_step_50/aime25/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 30,
+    "num_scores": 120,
+    "timeout_samples": 0,
+    "empty_samples": 0,
+    "acc": 2.5,
+    "pass_acc": 10.0,
+    "pass@k": {
+        "1": 2.5,
+        "2": 5.0,
+        "4": 10.0
+    },
+    "time_use_in_second": 159.8093228340149,
+    "time_use_in_minite": "2:39"
+}

eval_results_avg4/global_step_50/amc23/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_metrics.json

@@ -0,0 +1,15 @@
+{
+    "num_samples": 40,
+    "num_scores": 160,
+    "timeout_samples": 0,
+    "empty_samples": 0,
+    "acc": 26.9,
+    "pass_acc": 42.5,
+    "pass@k": {
+        "1": 26.9,
+        "2": 35.8,
+        "4": 42.5
+    },
+    "time_use_in_second": 147.4558162689209,
+    "time_use_in_minite": "2:27"
+}

eval_results_merged/merged.csv

@@ -0,0 +1,12 @@
+global_step,aime24_acc_avg4,aime25_acc_avg4,amc23_acc_avg4,aime24_acc,aime25_acc,amc23_acc,gsm8k_acc,math500_acc,minerva_math_acc,olympiadbench_acc,mmlu_stem_acc,prompt_level_strict_acc_ood,gpqa_pass@1:1_samples_ood
+0,2.50,2.50,20.00,3.30,0.00,17.50,62.50,46.40,13.60,17.60,40.30,20.9,
+5,3.30,3.30,29.40,3.30,3.30,35.00,69.40,53.40,20.20,21.00,43.00,17.7,
+10,5.80,4.20,30.00,0.00,3.30,15.00,76.30,56.80,16.50,20.70,44.10,21.6,
+15,5.00,2.50,31.20,0.00,3.30,25.00,78.20,58.40,23.90,24.90,45.00,26.1,
+20,3.30,1.70,32.50,6.70,3.30,32.50,78.50,57.20,25.00,24.90,47.70,24.6,
+25,6.70,3.30,33.80,3.30,3.30,37.50,77.80,59.00,22.80,23.00,49.20,25.7,
+30,5.80,3.30,31.20,3.30,0.00,32.50,79.40,60.40,24.60,25.00,50.40,26.2,
+35,4.20,1.70,34.40,3.30,6.70,30.00,80.30,62.20,25.00,25.30,51.60,25.0,
+40,7.50,4.20,33.10,3.30,0.00,40.00,79.20,60.20,25.40,26.50,53.70,27.4,
+45,5.00,2.50,32.50,6.70,0.00,22.50,80.10,60.00,25.70,25.50,53.50,27.0,
+50,6.70,2.50,26.90,3.30,3.30,32.50,80.80,58.80,26.50,24.10,53.80,25.9,

eval_results_merged_v3/merged.csv

@@ -0,0 +1,12 @@
+global_step,aime24_acc_avg16,aime25_acc_avg16,amc23_acc_avg16,aime24_acc_avg32,aime25_acc_avg32,amc23_acc_avg32,aime24_acc,aime25_acc,amc23_acc,gsm8k_acc,math500_acc,minerva_math_acc,olympiadbench_acc,mmlu_stem_acc,prompt_level_strict_acc_ood,gpqa_pass@1:1_samples_ood
+0,,,21.90,,,,3.30,0.00,17.50,62.50,46.40,13.60,17.60,40.30,20.9,
+5,,,24.20,3.10,1.50,,3.30,3.30,35.00,69.40,53.40,20.20,21.00,43.00,17.7,
+10,,,31.40,5.00,2.90,,0.00,3.30,15.00,76.30,56.80,16.50,20.70,44.10,21.6,
+15,,,32.20,4.50,3.20,,0.00,3.30,25.00,78.20,58.40,23.90,24.90,45.00,26.1,
+20,,,35.00,5.10,2.50,,6.70,3.30,32.50,78.50,57.20,25.00,24.90,47.70,24.6,
+25,,,33.90,5.00,2.70,,3.30,3.30,37.50,77.80,59.00,22.80,23.00,49.20,25.7,
+30,,,30.90,6.10,2.40,,3.30,0.00,32.50,79.40,60.40,24.60,25.00,50.40,26.2,
+35,,,33.90,5.60,3.20,,3.30,6.70,30.00,80.30,62.20,25.00,25.30,51.60,25.0,
+40,,,34.70,5.30,2.20,,3.30,0.00,40.00,79.20,60.20,25.40,26.50,53.70,27.4,
+45,,,33.10,4.90,2.50,,6.70,0.00,22.50,80.10,60.00,25.70,25.50,53.50,27.0,
+50,,,34.10,5.30,3.30,,3.30,3.30,32.50,80.80,58.80,26.50,24.10,53.80,25.9,

eval_results_merged_v3/missing.txt

@@ -0,0 +1 @@
+step 0: missing avg32

eval_results_merged_v4/merged.csv

@@ -0,0 +1,12 @@
+global_step,amc23_acc_avg16,aime24_acc_avg32,aime25_acc_avg32,gsm8k_acc,math500_acc,minerva_math_acc,olympiadbench_acc,mmlu_stem_acc,prompt_level_strict_acc_ood,mbpp_base_pass@1,mbpp_plus_pass@1
+0,21.90,,,62.50,46.40,13.60,17.60,40.30,20.9,,
+5,24.20,3.10,1.50,69.40,53.40,20.20,21.00,43.00,17.7,54.4,45.6
+10,31.40,5.00,2.90,76.30,56.80,16.50,20.70,44.10,21.6,55.6,47.0
+15,32.20,4.50,3.20,78.20,58.40,23.90,24.90,45.00,26.1,55.0,46.1
+20,35.00,5.10,2.50,78.50,57.20,25.00,24.90,47.70,24.6,56.0,47.2
+25,33.90,5.00,2.70,77.80,59.00,22.80,23.00,49.20,25.7,57.1,48.2
+30,30.90,6.10,2.40,79.40,60.40,24.60,25.00,50.40,26.2,57.1,48.3
+35,33.90,5.60,3.20,80.30,62.20,25.00,25.30,51.60,25.0,54.6,46.5
+40,34.70,5.30,2.20,79.20,60.20,25.40,26.50,53.70,27.4,57.9,49.3
+45,33.10,4.90,2.50,80.10,60.00,25.70,25.50,53.50,27.0,56.7,48.3
+50,34.10,5.30,3.30,80.80,58.80,26.50,24.10,53.80,25.9,56.0,47.4

eval_results_merged_v4/missing.txt

@@ -0,0 +1 @@
+step 0: missing avg32, mbpp

eval_results_ood/global_step_0/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_0_actor_huggingface/results_2025-08-28T15-04-01.057590.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.63711604,
+    "end_time": 18811748.076254945,
+    "total_evaluation_time_secondes": "3870.439138904214",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_0_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.2088724584103512,
+      "prompt_level_strict_acc_stderr": 0.017493107347793312,
+      "inst_level_strict_acc": 0.3249400479616307,
+      "inst_level_strict_acc_stderr": 0.0005020381027360004,
+      "prompt_level_loose_acc": 0.22181146025878004,
+      "prompt_level_loose_acc_stderr": 0.01787876540794439,
+      "inst_level_loose_acc": 0.34652278177458035,
+      "inst_level_loose_acc_stderr": 0.0005100828283216712
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.2088724584103512,
+      "prompt_level_strict_acc_stderr": 0.017493107347793312,
+      "inst_level_strict_acc": 0.3249400479616307,
+      "inst_level_strict_acc_stderr": 0.0005020381027360004,
+      "prompt_level_loose_acc": 0.22181146025878004,
+      "prompt_level_loose_acc_stderr": 0.01787876540794439,
+      "inst_level_loose_acc": 0.34652278177458035,
+      "inst_level_loose_acc_stderr": 0.0005100828283216712
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "5e36dc1464d8a69a"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "868acc3b1dc606da"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_10/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_10_actor_huggingface/results_2025-08-28T14-43-43.673554.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.70384178,
+    "end_time": 18810538.169428077,
+    "total_evaluation_time_secondes": "2660.465586297214",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_10_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.21626617375231053,
+      "prompt_level_strict_acc_stderr": 0.01771665442891014,
+      "inst_level_strict_acc": 0.3333333333333333,
+      "inst_level_strict_acc_stderr": 0.000489431816288728,
+      "prompt_level_loose_acc": 0.25508317929759705,
+      "prompt_level_loose_acc_stderr": 0.018758491950414142,
+      "inst_level_loose_acc": 0.36810551558752996,
+      "inst_level_loose_acc_stderr": 0.0005234324323312584
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.21626617375231053,
+      "prompt_level_strict_acc_stderr": 0.01771665442891014,
+      "inst_level_strict_acc": 0.3333333333333333,
+      "inst_level_strict_acc_stderr": 0.000489431816288728,
+      "prompt_level_loose_acc": 0.25508317929759705,
+      "prompt_level_loose_acc_stderr": 0.018758491950414142,
+      "inst_level_loose_acc": 0.36810551558752996,
+      "inst_level_loose_acc_stderr": 0.0005234324323312584
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "7575bde790fe6f08"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "87fc0142e87ea1db"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_15/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_15_actor_huggingface/results_2025-08-28T14-40-51.255236.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.635745857,
+    "end_time": 18810366.315407366,
+    "total_evaluation_time_secondes": "2488.6796615086496",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_15_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.26062846580406657,
+      "prompt_level_strict_acc_stderr": 0.018890584986760273,
+      "inst_level_strict_acc": 0.3752997601918465,
+      "inst_level_strict_acc_stderr": 0.0005276795699129097,
+      "prompt_level_loose_acc": 0.28835489833641403,
+      "prompt_level_loose_acc_stderr": 0.019493890350654703,
+      "inst_level_loose_acc": 0.4028776978417266,
+      "inst_level_loose_acc_stderr": 0.0005295500141012543
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.26062846580406657,
+      "prompt_level_strict_acc_stderr": 0.018890584986760273,
+      "inst_level_strict_acc": 0.3752997601918465,
+      "inst_level_strict_acc_stderr": 0.0005276795699129097,
+      "prompt_level_loose_acc": 0.28835489833641403,
+      "prompt_level_loose_acc_stderr": 0.019493890350654703,
+      "inst_level_loose_acc": 0.4028776978417266,
+      "inst_level_loose_acc_stderr": 0.0005295500141012543
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "6a3d0652153c9bb7"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "6417b61d11ee28c3"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_20/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_20_actor_huggingface/results_2025-08-28T14-31-58.919046.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.704473533,
+    "end_time": 18809835.16786196,
+    "total_evaluation_time_secondes": "1957.463388428092",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_20_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.24584103512014788,
+      "prompt_level_strict_acc_stderr": 0.01852941708079558,
+      "inst_level_strict_acc": 0.35851318944844124,
+      "inst_level_strict_acc_stderr": 0.0005177367706909966,
+      "prompt_level_loose_acc": 0.2846580406654344,
+      "prompt_level_loose_acc_stderr": 0.019418769106486003,
+      "inst_level_loose_acc": 0.40167865707434053,
+      "inst_level_loose_acc_stderr": 0.0005308235859063865
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.24584103512014788,
+      "prompt_level_strict_acc_stderr": 0.01852941708079558,
+      "inst_level_strict_acc": 0.35851318944844124,
+      "inst_level_strict_acc_stderr": 0.0005177367706909966,
+      "prompt_level_loose_acc": 0.2846580406654344,
+      "prompt_level_loose_acc_stderr": 0.019418769106486003,
+      "inst_level_loose_acc": 0.40167865707434053,
+      "inst_level_loose_acc_stderr": 0.0005308235859063865
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "9e63b6982edb60a3"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "7f0bbce7193e5835"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_25/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_25_actor_huggingface/results_2025-08-28T14-23-36.378741.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.63665532,
+    "end_time": 18809334.473295175,
+    "total_evaluation_time_secondes": "1456.836639855057",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_25_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.25693160813308685,
+      "prompt_level_strict_acc_stderr": 0.018802962575636847,
+      "inst_level_strict_acc": 0.381294964028777,
+      "inst_level_strict_acc_stderr": 0.0005088515258209939,
+      "prompt_level_loose_acc": 0.29944547134935307,
+      "prompt_level_loose_acc_stderr": 0.019709834029672916,
+      "inst_level_loose_acc": 0.4292565947242206,
+      "inst_level_loose_acc_stderr": 0.0005359650127318092
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.25693160813308685,
+      "prompt_level_strict_acc_stderr": 0.018802962575636847,
+      "inst_level_strict_acc": 0.381294964028777,
+      "inst_level_strict_acc_stderr": 0.0005088515258209939,
+      "prompt_level_loose_acc": 0.29944547134935307,
+      "prompt_level_loose_acc_stderr": 0.019709834029672916,
+      "inst_level_loose_acc": 0.4292565947242206,
+      "inst_level_loose_acc_stderr": 0.0005359650127318092
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "300fbc30a20ca56d"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "8ed3bd6852070947"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_30/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_30_actor_huggingface/results_2025-08-28T14-23-16.687943.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.636172168,
+    "end_time": 18809314.862311047,
+    "total_evaluation_time_secondes": "1437.2261388786137",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_30_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.26247689463955637,
+      "prompt_level_strict_acc_stderr": 0.018933742876044622,
+      "inst_level_strict_acc": 0.3980815347721823,
+      "inst_level_strict_acc_stderr": 0.0005233210912873605,
+      "prompt_level_loose_acc": 0.31053604436229204,
+      "prompt_level_loose_acc_stderr": 0.019912001290591244,
+      "inst_level_loose_acc": 0.447242206235012,
+      "inst_level_loose_acc_stderr": 0.000530725608886124
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.26247689463955637,
+      "prompt_level_strict_acc_stderr": 0.018933742876044622,
+      "inst_level_strict_acc": 0.3980815347721823,
+      "inst_level_strict_acc_stderr": 0.0005233210912873605,
+      "prompt_level_loose_acc": 0.31053604436229204,
+      "prompt_level_loose_acc_stderr": 0.019912001290591244,
+      "inst_level_loose_acc": 0.447242206235012,
+      "inst_level_loose_acc_stderr": 0.000530725608886124
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "8f6719798d9174bd"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "66197f09ecf3792e"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_35/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_35_actor_huggingface/results_2025-08-28T15-19-52.260147.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18811863.0122587,
+    "end_time": 18812710.173764117,
+    "total_evaluation_time_secondes": "847.1615054160357",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_35_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.24953789279112754,
+      "prompt_level_strict_acc_stderr": 0.018622404509805863,
+      "inst_level_strict_acc": 0.37889688249400477,
+      "inst_level_strict_acc_stderr": 0.0005113738731364967,
+      "prompt_level_loose_acc": 0.3142329020332717,
+      "prompt_level_loose_acc_stderr": 0.01997640206515067,
+      "inst_level_loose_acc": 0.4448441247002398,
+      "inst_level_loose_acc_stderr": 0.0005297346220956167
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.24953789279112754,
+      "prompt_level_strict_acc_stderr": 0.018622404509805863,
+      "inst_level_strict_acc": 0.37889688249400477,
+      "inst_level_strict_acc_stderr": 0.0005113738731364967,
+      "prompt_level_loose_acc": 0.3142329020332717,
+      "prompt_level_loose_acc_stderr": 0.01997640206515067,
+      "inst_level_loose_acc": 0.4448441247002398,
+      "inst_level_loose_acc_stderr": 0.0005297346220956167
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "115a2ac2647bf2e6"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "a552e71cdd179caa"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_40/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_40_actor_huggingface/results_2025-08-28T15-18-51.746168.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18811863.02156565,
+    "end_time": 18812650.77936895,
+    "total_evaluation_time_secondes": "787.757803298533",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_40_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.2735674676524954,
+      "prompt_level_strict_acc_stderr": 0.019183727107392825,
+      "inst_level_strict_acc": 0.41247002398081534,
+      "inst_level_strict_acc_stderr": 0.000532554469072579,
+      "prompt_level_loose_acc": 0.34750462107208874,
+      "prompt_level_loose_acc_stderr": 0.02049142365341572,
+      "inst_level_loose_acc": 0.4832134292565947,
+      "inst_level_loose_acc_stderr": 0.0005287169246309893
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.2735674676524954,
+      "prompt_level_strict_acc_stderr": 0.019183727107392825,
+      "inst_level_strict_acc": 0.41247002398081534,
+      "inst_level_strict_acc_stderr": 0.000532554469072579,
+      "prompt_level_loose_acc": 0.34750462107208874,
+      "prompt_level_loose_acc_stderr": 0.02049142365341572,
+      "inst_level_loose_acc": 0.4832134292565947,
+      "inst_level_loose_acc_stderr": 0.0005287169246309893
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "88ff4ce1dc04a1a1"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "2f831f811e8d0b88"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_45/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_45_actor_huggingface/results_2025-08-28T15-19-48.255403.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18811863.018329486,
+    "end_time": 18812707.18770021,
+    "total_evaluation_time_secondes": "844.1693707220256",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_45_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.2698706099815157,
+      "prompt_level_strict_acc_stderr": 0.019102087526494303,
+      "inst_level_strict_acc": 0.407673860911271,
+      "inst_level_strict_acc_stderr": 0.0005080955062228919,
+      "prompt_level_loose_acc": 0.3456561922365989,
+      "prompt_level_loose_acc_stderr": 0.02046577943268255,
+      "inst_level_loose_acc": 0.48081534772182255,
+      "inst_level_loose_acc_stderr": 0.0005250291013858589
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.2698706099815157,
+      "prompt_level_strict_acc_stderr": 0.019102087526494303,
+      "inst_level_strict_acc": 0.407673860911271,
+      "inst_level_strict_acc_stderr": 0.0005080955062228919,
+      "prompt_level_loose_acc": 0.3456561922365989,
+      "prompt_level_loose_acc_stderr": 0.02046577943268255,
+      "inst_level_loose_acc": 0.48081534772182255,
+      "inst_level_loose_acc_stderr": 0.0005250291013858589
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "c81b0908f6d01a01"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "11c1fdf8fbfac4f1"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_5/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_5_actor_huggingface/results_2025-08-28T14-57-06.728642.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18807877.63640212,
+    "end_time": 18811337.61977247,
+    "total_evaluation_time_secondes": "3459.9833703525364",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_5_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.17744916820702403,
+      "prompt_level_strict_acc_stderr": 0.016440744379699793,
+      "inst_level_strict_acc": 0.30935251798561153,
+      "inst_level_strict_acc_stderr": 0.0004545789816041833,
+      "prompt_level_loose_acc": 0.19408502772643252,
+      "prompt_level_loose_acc_stderr": 0.0170193805507494,
+      "inst_level_loose_acc": 0.3225419664268585,
+      "inst_level_loose_acc_stderr": 0.0004593128188518521
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.17744916820702403,
+      "prompt_level_strict_acc_stderr": 0.016440744379699793,
+      "inst_level_strict_acc": 0.30935251798561153,
+      "inst_level_strict_acc_stderr": 0.0004545789816041833,
+      "prompt_level_loose_acc": 0.19408502772643252,
+      "prompt_level_loose_acc_stderr": 0.0170193805507494,
+      "inst_level_loose_acc": 0.3225419664268585,
+      "inst_level_loose_acc_stderr": 0.0004593128188518521
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "cb86db8b46b3063a"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "1ff5cba463b1840d"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/global_step_50/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_50_actor_huggingface/results_2025-08-28T15-18-25.365193.json

@@ -0,0 +1,141 @@
+{
+  "config_general": {
+    "lighteval_sha": "327071fe86e427d880f907a51d1462f4a3f951c1",
+    "num_fewshot_seeds": 1,
+    "max_samples": null,
+    "job_id": 0,
+    "start_time": 18811863.116520293,
+    "end_time": 18812624.911301415,
+    "total_evaluation_time_secondes": "761.7947811223567",
+    "model_name": "_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_50_actor_huggingface",
+    "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": 1.0,
+      "top_k": null,
+      "min_p": null,
+      "top_p": 0.95,
+      "truncate_prompt": null,
+      "response_format": null
+    }
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.2587800369685767,
+      "prompt_level_strict_acc_stderr": 0.018846992560712535,
+      "inst_level_strict_acc": 0.38848920863309355,
+      "inst_level_strict_acc_stderr": 0.0005002404333580301,
+      "prompt_level_loose_acc": 0.3419593345656192,
+      "prompt_level_loose_acc_stderr": 0.020413464513923615,
+      "inst_level_loose_acc": 0.4676258992805755,
+      "inst_level_loose_acc_stderr": 0.0005162613395928805
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.2587800369685767,
+      "prompt_level_strict_acc_stderr": 0.018846992560712535,
+      "inst_level_strict_acc": 0.38848920863309355,
+      "inst_level_strict_acc_stderr": 0.0005002404333580301,
+      "prompt_level_loose_acc": 0.3419593345656192,
+      "prompt_level_loose_acc_stderr": 0.020413464513923615,
+      "inst_level_loose_acc": 0.4676258992805755,
+      "inst_level_loose_acc_stderr": 0.0005162613395928805
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": "0.1"
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "google/IFEval",
+      "hf_subset": "default",
+      "metric": [
+        {
+          "metric_name": [
+            "prompt_level_strict_acc",
+            "inst_level_strict_acc",
+            "prompt_level_loose_acc",
+            "inst_level_loose_acc"
+          ],
+          "higher_is_better": {
+            "prompt_level_strict_acc": true,
+            "inst_level_strict_acc": true,
+            "prompt_level_loose_acc": true,
+            "inst_level_loose_acc": true
+          },
+          "category": "3",
+          "use_case": "1",
+          "sample_level_fn": "ifeval_metric",
+          "corpus_level_fn": {
+            "prompt_level_strict_acc": "mean",
+            "inst_level_strict_acc": "agg_inst_level_acc",
+            "prompt_level_loose_acc": "mean",
+            "inst_level_loose_acc": "agg_inst_level_acc"
+          }
+        }
+      ],
+      "hf_revision": null,
+      "hf_filter": null,
+      "hf_avail_splits": [
+        "train"
+      ],
+      "trust_dataset": false,
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "generation_grammar": null,
+      "stop_sequence": [],
+      "num_samples": null,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "must_remove_duplicate_docs": false,
+      "version": "0.1"
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "e3d19e04074f1062",
+        "hash_cont_tokens": "639f8c16b4bbe76d"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "32d769c21a57d2c7",
+      "hash_cont_tokens": "602d2eb1bbdd7950"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}

eval_results_ood/results.csv

@@ -0,0 +1,12 @@
+exp_name,global_step,model_name,json_relpath,prompt_level_strict_acc,prompt_level_strict_acc_stderr,inst_level_strict_acc,inst_level_strict_acc_stderr,prompt_level_loose_acc,prompt_level_loose_acc_stderr,inst_level_loose_acc,inst_level_loose_acc_stderr,gpqa_pass@1:1_samples,gpqa_pass@1:1_samples_stderr
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,0,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_0_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_0/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_0_actor_huggingface/results_2025-08-28T15-04-01.057590.json,0.2088724584103512,0.017493107347793312,0.3249400479616307,0.0005020381027360004,0.22181146025878004,0.01787876540794439,0.34652278177458035,0.0005100828283216712,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,5,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_5_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_5/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_5_actor_huggingface/results_2025-08-28T14-57-06.728642.json,0.17744916820702403,0.016440744379699793,0.30935251798561153,0.0004545789816041833,0.19408502772643252,0.0170193805507494,0.3225419664268585,0.0004593128188518521,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,10,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_10_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_10/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_10_actor_huggingface/results_2025-08-28T14-43-43.673554.json,0.21626617375231053,0.01771665442891014,0.3333333333333333,0.000489431816288728,0.25508317929759705,0.018758491950414142,0.36810551558752996,0.0005234324323312584,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,15,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_15_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_15/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_15_actor_huggingface/results_2025-08-28T14-40-51.255236.json,0.26062846580406657,0.018890584986760273,0.3752997601918465,0.0005276795699129097,0.28835489833641403,0.019493890350654703,0.4028776978417266,0.0005295500141012543,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,20,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_20_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_20/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_20_actor_huggingface/results_2025-08-28T14-31-58.919046.json,0.24584103512014788,0.01852941708079558,0.35851318944844124,0.0005177367706909966,0.2846580406654344,0.019418769106486003,0.40167865707434053,0.0005308235859063865,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,25,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_25_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_25/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_25_actor_huggingface/results_2025-08-28T14-23-36.378741.json,0.25693160813308685,0.018802962575636847,0.381294964028777,0.0005088515258209939,0.29944547134935307,0.019709834029672916,0.4292565947242206,0.0005359650127318092,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,30,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_30_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_30/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_30_actor_huggingface/results_2025-08-28T14-23-16.687943.json,0.26247689463955637,0.018933742876044622,0.3980815347721823,0.0005233210912873605,0.31053604436229204,0.019912001290591244,0.447242206235012,0.000530725608886124,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,35,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_35_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_35/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_35_actor_huggingface/results_2025-08-28T15-19-52.260147.json,0.24953789279112754,0.018622404509805863,0.37889688249400477,0.0005113738731364967,0.3142329020332717,0.01997640206515067,0.4448441247002398,0.0005297346220956167,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,40,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_40_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_40/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_40_actor_huggingface/results_2025-08-28T15-18-51.746168.json,0.2735674676524954,0.019183727107392825,0.41247002398081534,0.000532554469072579,0.34750462107208874,0.02049142365341572,0.4832134292565947,0.0005287169246309893,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,45,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_45_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_45/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_45_actor_huggingface/results_2025-08-28T15-19-48.255403.json,0.2698706099815157,0.019102087526494303,0.407673860911271,0.0005080955062228919,0.3456561922365989,0.02046577943268255,0.48081534772182255,0.0005250291013858589,,
+verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15,50,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_50_actor_huggingface,verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15/eval_results_ood/global_step_50/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_dapo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.15_global_step_50_actor_huggingface/results_2025-08-28T15-18-25.365193.json,0.2587800369685767,0.018846992560712535,0.38848920863309355,0.0005002404333580301,0.3419593345656192,0.020413464513923615,0.4676258992805755,0.0005162613395928805,,