thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
1.99 kB
{
"problems": 256,
"samples": 512,
"acc": 0.494140625,
"acc_se": 0.028288931544680946,
"cut_share": 0.466796875,
"mean_tokens": 20117.046875,
"closed_share": 0.609375,
"p_finish": 0.533203125,
"p_correct_given_finished": 0.9194139194139194,
"p_correct_given_cut": 0.008368200836820083,
"no_answer_finished_share": 0.001953125,
"loop_end_share_of_cut": 0.8702928870292888,
"tail_dup8_cut_mean": 0.8428794979079498,
"median_tokens_correct": 6930,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.62890625,
"acc_se": 0.03852203620415812,
"cut_share": 0.30078125,
"mean_tokens": 15767.94140625
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.359375,
"acc_se": 0.037996022313054996,
"cut_share": 0.6328125,
"mean_tokens": 24466.15234375
}
},
"run": {
"name": "dev/all2",
"model": "qwen35-all2",
"data": "dev256",
"n": 2,
"shards": 1,
"est": 10
},
"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": "all2",
"spec": {
"about": "the agent's second try: mirror the loop component in every neuron",
"layers": {
"15": {
"alpha": 2,
"cols": "all"
},
"19": {
"alpha": 2,
"cols": "all"
}
}
},
"built": {
"15": {
"alpha": 2.0,
"n_cols": 3584,
"dir": "released",
"relative_change": 0.0705324038863182
},
"19": {
"alpha": 2.0,
"n_cols": 3584,
"dir": "released",
"relative_change": 0.06749952584505081
}
},
"edited_sha256": {
"model.language_model.layers.15.mlp.down_proj.weight": "ab3f4a8ec3f3f44aa410038ca1541b0da2fc3061d03b246b04e9804c360d6e2c",
"model.language_model.layers.19.mlp.down_proj.weight": "670e4268823c18019531bf7b93bc0201e9e5fc45affdaee847a899116533d60a"
}
},
"tokens_total": 10299928,
"gpu_seconds": 1688,
"tok_per_s_per_gpu": 6271
}