thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
2.09 kB
{
"problems": 256,
"samples": 512,
"acc": 0.146484375,
"acc_se": 0.018670926374704726,
"cut_share": 0.8671875,
"mean_tokens": 29112.546875,
"closed_share": 0.134765625,
"p_finish": 0.1328125,
"p_correct_given_finished": 1.0,
"p_correct_given_cut": 0.015765765765765764,
"no_answer_finished_share": 0.0,
"loop_end_share_of_cut": 0.9752252252252253,
"tail_dup8_cut_mean": 0.9217704954954955,
"median_tokens_correct": 5984,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.19140625,
"acc_se": 0.027796968120647532,
"cut_share": 0.8203125,
"mean_tokens": 27803.72265625
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.1015625,
"acc_se": 0.02440191011526807,
"cut_share": 0.9140625,
"mean_tokens": 30421.37109375
}
},
"run": {
"name": "dev/randcols",
"model": "qwen35-randcols",
"data": "dev256",
"n": 2,
"shards": 1,
"est": 15
},
"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": "randcols",
"spec": {
"about": "control: direction u, same size of change, but 64 random neurons instead of the chosen ones",
"layers": {
"15": {
"alpha": 6,
"cols": "random:15",
"norm_match": true
},
"19": {
"alpha": 6,
"cols": "random:19",
"norm_match": true
}
}
},
"built": {
"15": {
"alpha": 31.318699836730957,
"n_cols": 64,
"dir": "released",
"relative_change": 0.10634618997573853
},
"19": {
"alpha": 21.895952224731445,
"n_cols": 64,
"dir": "released",
"relative_change": 0.08105457574129105
}
},
"edited_sha256": {
"model.language_model.layers.15.mlp.down_proj.weight": "09c8f4162ead661723258a282338d864d201aa5a80d5db41d2dbbfbe2d36e009",
"model.language_model.layers.19.mlp.down_proj.weight": "784fb5f6af009fc9915970bad68d6d234bc59be00a10f21abd84dc43d53c16e6"
}
},
"tokens_total": 14905624,
"gpu_seconds": 2638,
"tok_per_s_per_gpu": 5772
}