Upload folder using huggingface_hub
Browse files- seq_code_math_kd_only/results_stage2_math/benchmarked +0 -0
- seq_code_math_kd_only/results_stage2_math/gsm8k.jsonl +0 -0
- seq_code_math_kd_only/results_stage2_math/humaneval.jsonl +0 -0
- seq_code_math_kd_only/results_stage2_math/humaneval_eval_results.json +0 -0
- seq_code_math_kd_only/results_stage2_math/math500.jsonl +0 -0
- seq_code_math_kd_only/results_stage2_math/mbpp.jsonl +0 -0
- seq_code_math_kd_only/results_stage2_math/mbpp_eval_results.json +0 -0
- seq_code_math_kd_only/results_stage2_math/summary.json +1 -0
- seq_code_math_kd_only/run_info.json +3 -1
seq_code_math_kd_only/results_stage2_math/benchmarked
ADDED
|
File without changes
|
seq_code_math_kd_only/results_stage2_math/gsm8k.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/humaneval.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/humaneval_eval_results.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/math500.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/mbpp.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/mbpp_eval_results.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
seq_code_math_kd_only/results_stage2_math/summary.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"model": "seq_code_math_kd_only/stage2_math", "gsm8k": 0.7376800606520091, "mbpp": 0.6613756613756614, "mbpp_plus": 0.582010582010582, "humaneval": 0.6097560975609756, "humaneval_plus": 0.5609756097560976, "n_gsm8k": 1319, "n_mbpp": 378, "n_humaneval": 164, "math500": 0.562, "n_math500": 500}
|
seq_code_math_kd_only/run_info.json
CHANGED
|
@@ -57,7 +57,9 @@
|
|
| 57 |
"math_hard_ce": 0.8143728766945073
|
| 58 |
},
|
| 59 |
"peak_gb_student": 36.68444871902466,
|
| 60 |
-
"peak_gb_teachers": 30.639182090759277
|
|
|
|
|
|
|
| 61 |
}
|
| 62 |
},
|
| 63 |
"resumed": {
|
|
|
|
| 57 |
"math_hard_ce": 0.8143728766945073
|
| 58 |
},
|
| 59 |
"peak_gb_student": 36.68444871902466,
|
| 60 |
+
"peak_gb_teachers": 30.639182090759277,
|
| 61 |
+
"bench_h": 0.5482511294550366,
|
| 62 |
+
"gpu_bench": "NVIDIA RTX 6000 Ada Generation"
|
| 63 |
}
|
| 64 |
},
|
| 65 |
"resumed": {
|