thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
1.58 kB
{
"problems": 256,
"samples": 512,
"acc": 0.615234375,
"acc_se": 0.027772321975577316,
"cut_share": 0.197265625,
"mean_tokens": 16103.923828125,
"closed_share": 0.859375,
"p_finish": 0.802734375,
"p_correct_given_finished": 0.7664233576642335,
"p_correct_given_cut": 0.0,
"no_answer_finished_share": 0.02734375,
"loop_end_share_of_cut": 0.24752475247524752,
"tail_dup8_cut_mean": 0.31726237623762377,
"median_tokens_correct": 7998,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.74609375,
"acc_se": 0.03463288810753153,
"cut_share": 0.046875,
"mean_tokens": 10830.4609375
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.484375,
"acc_se": 0.04035151739041125,
"cut_share": 0.34765625,
"mean_tokens": 21377.38671875
}
},
"run": {
"name": "dev/p50c",
"model": "qwen35-p50c",
"data": "dev256",
"n": 2,
"shards": 1,
"est": 8,
"priority": 1
},
"sampling": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": 20,
"presence_penalty": 1.5,
"repetition_penalty": 1.0,
"max_tokens": 32768,
"thinking": true
},
"edit_note": {
"name": "p50c",
"about": "the paper's released edit, rebuilt with its apply_edit.py",
"edited_sha256": {
"model.language_model.layers.15.mlp.down_proj.weight": "82e2d733ea64feaeb4af93a1e065641d3df98b494a8b7576ef2f52e90fe2ee53",
"model.language_model.layers.19.mlp.down_proj.weight": "8ef3c61bc36af113ad58e89dbb0a4a14dcba5a45cb750c15ab515c82ddd12eb2"
}
},
"tokens_total": 8245209,
"gpu_seconds": 1237,
"tok_per_s_per_gpu": 6997
}