Upload folder using huggingface_hub
Browse files- .gitattributes +2 -0
- hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl +3 -0
- 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 +20 -0
- hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl +3 -0
- hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1_llama-boxed_metrics.json +9 -0
.gitattributes
CHANGED
|
@@ -41,3 +41,5 @@ stellalisy/rethink_rlvr_reproduce-majority_vote-qwen2.5_math_7b-lr5e-7-kl0.00-st
|
|
| 41 |
stellalisy/rethink_rlvr_reproduce-random-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
|
| 42 |
stellalisy/rethink_rlvr_reproduce-majority_vote-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
|
| 43 |
stellalisy/rethink_rlvr_reproduce-random-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
|
|
|
|
|
|
|
|
|
| 41 |
stellalisy/rethink_rlvr_reproduce-random-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
|
| 42 |
stellalisy/rethink_rlvr_reproduce-majority_vote-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
|
| 43 |
stellalisy/rethink_rlvr_reproduce-random-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
|
| 44 |
+
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
|
| 45 |
+
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
|
hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/minerva_math/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1785c7520c96ca50c36a24ed924818b3cf12d744aac57f5d88a3e394839025be
|
| 3 |
+
size 271193671
|
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
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_samples": 272,
|
| 3 |
+
"num_scores": 139264,
|
| 4 |
+
"timeout_samples": 22,
|
| 5 |
+
"empty_samples": 36,
|
| 6 |
+
"acc": 2.9,
|
| 7 |
+
"type_acc": {
|
| 8 |
+
"Differential Equations (18.03 Spring 2010)": 4.2,
|
| 9 |
+
"Dynamics and Control (2.003 Spring 2005)": 7.7,
|
| 10 |
+
"Ecology I (1.018J Fall 2009)": 20.0,
|
| 11 |
+
"Information and Entropy (6.050J Spring 2008)": 0.0,
|
| 12 |
+
"Introduction to Astronomy (8.282J Spring 2006)": 5.7,
|
| 13 |
+
"Introduction to Solid State Chemistry (3.091 Fall 2010)": 2.1,
|
| 14 |
+
"Physical Chemistry (5.61 Fall 2017)": 0.0,
|
| 15 |
+
"Principles of Microeconomics (14.01 Fall 2011)": 5.6,
|
| 16 |
+
"Relativity (8.033 Fall 2006)": 0.0
|
| 17 |
+
},
|
| 18 |
+
"time_use_in_second": 12448.452505588531,
|
| 19 |
+
"time_use_in_minite": "207:28"
|
| 20 |
+
}
|
hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8418efd3b3582cd43fb900de3eb88e4b3730d1d28a62869dfab62936704ce201
|
| 3 |
+
size 748567428
|
hkust-nlp/Llama-3.1-8B-SimpleRL-Zoo/math_eval/olympiadbench/test_llama-boxed_-1_seed0_t1.0_s0_e-1_llama-boxed_metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_samples": 675,
|
| 3 |
+
"num_scores": 345600,
|
| 4 |
+
"timeout_samples": 14,
|
| 5 |
+
"empty_samples": 56,
|
| 6 |
+
"acc": 5.3,
|
| 7 |
+
"time_use_in_second": 40533.43204045296,
|
| 8 |
+
"time_use_in_minite": "675:33"
|
| 9 |
+
}
|