Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Qwen/Qwen2.5-Math-7B/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
+Qwen/Qwen2.5-Math-7B/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
+stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
+stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
+stellalisy/rethink_rlvr_reproduce-majority_vote-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text

Qwen/Qwen2.5-Math-7B/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae8a0b08617342ffa78eb00b58f854d659883cd15f0b34f2bc664a616c143409
+size 652553119

Qwen/Qwen2.5-Math-7B/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_qwen-boxed_metrics.json

@@ -0,0 +1,20 @@
+{
+    "num_samples": 272,
+    "num_scores": 278528,
+    "timeout_samples": 2434,
+    "empty_samples": 38,
+    "acc": 16.5,
+    "type_acc": {
+        "Differential Equations (18.03 Spring 2010)": 27.1,
+        "Dynamics and Control (2.003 Spring 2005)": 23.1,
+        "Ecology I (1.018J Fall 2009)": 0.0,
+        "Information and Entropy (6.050J Spring 2008)": 33.3,
+        "Introduction to Astronomy (8.282J Spring 2006)": 9.4,
+        "Introduction to Solid State Chemistry (3.091 Fall 2010)": 17.5,
+        "Physical Chemistry (5.61 Fall 2017)": 0.0,
+        "Principles of Microeconomics (14.01 Fall 2011)": 55.6,
+        "Relativity (8.033 Fall 2006)": 9.1
+    },
+    "time_use_in_second": 25311.111163377762,
+    "time_use_in_minite": "421:51"
+}

Qwen/Qwen2.5-Math-7B/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5228c4d2526d879a54faff5ebd1ecf3be204d6b005100b378c74ff6a6586a2a4
+size 1985712202

Qwen/Qwen2.5-Math-7B/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_qwen-boxed_metrics.json

@@ -0,0 +1,9 @@
+{
+    "num_samples": 675,
+    "num_scores": 691200,
+    "timeout_samples": 111,
+    "empty_samples": 65,
+    "acc": 20.6,
+    "time_use_in_second": 78619.30241346359,
+    "time_use_in_minite": "1310:19"
+}

stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b28fb2ac955017958a6fcd27f07da362e945b96df75c02285a7c07a89e3ace96
+size 556568029

stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_qwen-boxed_metrics.json

@@ -0,0 +1,20 @@
+{
+    "num_samples": 272,
+    "num_scores": 278528,
+    "timeout_samples": 1740,
+    "empty_samples": 1,
+    "acc": 32.0,
+    "type_acc": {
+        "Differential Equations (18.03 Spring 2010)": 31.2,
+        "Dynamics and Control (2.003 Spring 2005)": 57.7,
+        "Ecology I (1.018J Fall 2009)": 40.0,
+        "Information and Entropy (6.050J Spring 2008)": 33.3,
+        "Introduction to Astronomy (8.282J Spring 2006)": 22.6,
+        "Introduction to Solid State Chemistry (3.091 Fall 2010)": 24.7,
+        "Physical Chemistry (5.61 Fall 2017)": 27.3,
+        "Principles of Microeconomics (14.01 Fall 2011)": 61.1,
+        "Relativity (8.033 Fall 2006)": 45.5
+    },
+    "time_use_in_second": 21749.05103445053,
+    "time_use_in_minite": "362:29"
+}

stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2c7404fcc452f299ba852585360c8336caa68186009b9005a84005dccceb289
+size 1757460707

stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/olympiadbench/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_qwen-boxed_metrics.json

@@ -0,0 +1,9 @@
+{
+    "num_samples": 675,
+    "num_scores": 691200,
+    "timeout_samples": 9604,
+    "empty_samples": 0,
+    "acc": 37.6,
+    "time_use_in_second": 73957.25717425346,
+    "time_use_in_minite": "1232:37"
+}

stellalisy/rethink_rlvr_reproduce-majority_vote-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a22a0a9058960ad9f6ffb0e15d6f34434ec571468c1f41939ae5420c05679722
+size 528128327

stellalisy/rethink_rlvr_reproduce-majority_vote-qwen2.5_math_7b-lr5e-7-kl0.00-step150/math_eval/minerva_math/test_qwen-boxed_-1_seed0_t1.0_s0_e-1_qwen-boxed_metrics.json

@@ -0,0 +1,20 @@
+{
+    "num_samples": 272,
+    "num_scores": 278528,
+    "timeout_samples": 1758,
+    "empty_samples": 6,
+    "acc": 36.4,
+    "type_acc": {
+        "Differential Equations (18.03 Spring 2010)": 56.2,
+        "Dynamics and Control (2.003 Spring 2005)": 50.0,
+        "Ecology I (1.018J Fall 2009)": 40.0,
+        "Information and Entropy (6.050J Spring 2008)": 33.3,
+        "Introduction to Astronomy (8.282J Spring 2006)": 32.1,
+        "Introduction to Solid State Chemistry (3.091 Fall 2010)": 24.7,
+        "Physical Chemistry (5.61 Fall 2017)": 27.3,
+        "Principles of Microeconomics (14.01 Fall 2011)": 50.0,
+        "Relativity (8.033 Fall 2006)": 18.2
+    },
+    "time_use_in_second": 20768.637530326843,
+    "time_use_in_minite": "346:08"
+}