Upload folder using huggingface_hub
Browse files- .gitattributes +2 -0
- 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 +3 -0
- 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_qwen-boxed_metrics.json +9 -0
- 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 +3 -0
- 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_qwen-boxed_metrics.json +9 -0
.gitattributes
CHANGED
|
@@ -39,3 +39,5 @@ stellalisy/rethink_rlvr_reproduce-ground_truth-qwen2.5_math_7b-lr5e-7-kl0.00-ste
|
|
| 39 |
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
|
| 40 |
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
|
| 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
|
|
|
|
|
|
|
|
|
| 39 |
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
|
| 40 |
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
|
| 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
|
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
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:479495791608214f4bce5ee17dcb0626a07608b3341f4af24a1c64f51671fa9d
|
| 3 |
+
size 1711838271
|
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_qwen-boxed_metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_samples": 675,
|
| 3 |
+
"num_scores": 691200,
|
| 4 |
+
"timeout_samples": 1353,
|
| 5 |
+
"empty_samples": 0,
|
| 6 |
+
"acc": 36.7,
|
| 7 |
+
"time_use_in_second": 67782.23354554176,
|
| 8 |
+
"time_use_in_minite": "1129:42"
|
| 9 |
+
}
|
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
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:039706cfbfcd79ec876160ca883e5e0de3e5b7a77be533fcc67c9c080018c1b7
|
| 3 |
+
size 2083721102
|
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_qwen-boxed_metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_samples": 675,
|
| 3 |
+
"num_scores": 691200,
|
| 4 |
+
"timeout_samples": 5576,
|
| 5 |
+
"empty_samples": 41,
|
| 6 |
+
"acc": 26.8,
|
| 7 |
+
"time_use_in_second": 82045.03149294853,
|
| 8 |
+
"time_use_in_minite": "1367:25"
|
| 9 |
+
}
|