Upload folder using huggingface_hub

.gitattributes

@@ -44,3 +44,4 @@ stellalisy/rethink_rlvr_reproduce-random-qwen2.5_math_7b-lr5e-7-kl0.00-step150/m
 hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
 hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
 meta-llama/Llama-3.1-8B/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text
+meta-llama/Llama-3.1-8B/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl filter=lfs diff=lfs merge=lfs -text

hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1785c7520c96ca50c36a24ed924818b3cf12d744aac57f5d88a3e394839025be
-size 271193671
+oid sha256:58d8d66396840424be31a5f8c2445fc3042926b1f2f5365d738ebaffe0ae7e5a
+size 135737434

hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1_llama-boxed_metrics.json

@@ -1,20 +1,20 @@
 {
     "num_samples": 272,
-    "num_scores": 139264,
-    "timeout_samples": 22,
-    "empty_samples": 36,
-    "acc": 2.9,
+    "num_scores": 69632,
+    "timeout_samples": 0,
+    "empty_samples": 43,
+    "acc": 6.2,
     "type_acc": {
-        "Differential Equations (18.03 Spring 2010)": 4.2,
+        "Differential Equations (18.03 Spring 2010)": 0.0,
         "Dynamics and Control (2.003 Spring 2005)": 7.7,
         "Ecology I (1.018J Fall 2009)": 20.0,
         "Information and Entropy (6.050J Spring 2008)": 0.0,
-        "Introduction to Astronomy (8.282J Spring 2006)": 5.7,
+        "Introduction to Astronomy (8.282J Spring 2006)": 7.5,
         "Introduction to Solid State Chemistry (3.091 Fall 2010)": 2.1,
         "Physical Chemistry (5.61 Fall 2017)": 0.0,
         "Principles of Microeconomics (14.01 Fall 2011)": 5.6,
         "Relativity (8.033 Fall 2006)": 0.0
     },
-    "time_use_in_second": 12448.452505588531,
-    "time_use_in_minite": "207:28"
+    "time_use_in_second": 6071.073959350586,
+    "time_use_in_minite": "101:11"
 }

meta-llama/Llama-3.1-8B/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:709bad71f3cb635f3838c9d9494151a64cf942a62b3dbb2e2878d68d6e704854
+size 241628082

meta-llama/Llama-3.1-8B/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1_llama-boxed_metrics.json

@@ -0,0 +1,9 @@
+{
+    "num_samples": 675,
+    "num_scores": 172800,
+    "timeout_samples": 22,
+    "empty_samples": 133,
+    "acc": 1.6,
+    "time_use_in_second": 10533.309227705002,
+    "time_use_in_minite": "175:33"
+}