Upload folder using huggingface_hub
Browse files- seq_math_code_gt_sft_seed1/results_stage2_code/benchmarked +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/gsm8k.jsonl +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/humaneval.jsonl +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/humaneval_eval_results.json +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/math500.jsonl +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/mbpp.jsonl +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/mbpp_eval_results.json +0 -0
- seq_math_code_gt_sft_seed1/results_stage2_code/summary.json +1 -0
- seq_math_code_gt_sft_seed1/run_info.json +6 -2
seq_math_code_gt_sft_seed1/results_stage2_code/benchmarked
ADDED
|
File without changes
|
seq_math_code_gt_sft_seed1/results_stage2_code/gsm8k.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/humaneval.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/humaneval_eval_results.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/math500.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/mbpp.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/mbpp_eval_results.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_math_code_gt_sft_seed1/results_stage2_code/summary.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"model": "seq_math_code_gt_sft_seed1/stage2_code", "gsm8k": 0.6057619408642911, "mbpp": 0.6534391534391535, "mbpp_plus": 0.5476190476190477, "humaneval": 0.5975609756097561, "humaneval_plus": 0.5487804878048781, "n_gsm8k": 1319, "n_mbpp": 378, "n_humaneval": 164, "math500": 0.348, "n_math500": 500}
|
seq_math_code_gt_sft_seed1/run_info.json
CHANGED
|
@@ -32,7 +32,9 @@
|
|
| 32 |
"math_hard_ce": 0.6126785609887812
|
| 33 |
},
|
| 34 |
"peak_gb_student": 36.57978296279907,
|
| 35 |
-
"peak_gb_teachers": 30.618666172027588
|
|
|
|
|
|
|
| 36 |
},
|
| 37 |
"stage2_code": {
|
| 38 |
"domain": "code",
|
|
@@ -57,7 +59,9 @@
|
|
| 57 |
"math_hard_ce": 0.6428350913066744
|
| 58 |
},
|
| 59 |
"peak_gb_student": 33.86883306503296,
|
| 60 |
-
"peak_gb_teachers": 30.005428314208984
|
|
|
|
|
|
|
| 61 |
}
|
| 62 |
},
|
| 63 |
"train_h": 6.29169988407029,
|
|
|
|
| 32 |
"math_hard_ce": 0.6126785609887812
|
| 33 |
},
|
| 34 |
"peak_gb_student": 36.57978296279907,
|
| 35 |
+
"peak_gb_teachers": 30.618666172027588,
|
| 36 |
+
"bench_h": 0.5074694540103276,
|
| 37 |
+
"gpu_bench": "NVIDIA RTX A6000"
|
| 38 |
},
|
| 39 |
"stage2_code": {
|
| 40 |
"domain": "code",
|
|
|
|
| 59 |
"math_hard_ce": 0.6428350913066744
|
| 60 |
},
|
| 61 |
"peak_gb_student": 33.86883306503296,
|
| 62 |
+
"peak_gb_teachers": 30.005428314208984,
|
| 63 |
+
"bench_h": 0.3323956937260098,
|
| 64 |
+
"gpu_bench": "NVIDIA RTX A6000"
|
| 65 |
}
|
| 66 |
},
|
| 67 |
"train_h": 6.29169988407029,
|