thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
1.98 kB
{
"problems": 256,
"samples": 512,
"acc": 0.287109375,
"acc_se": 0.023203682034311893,
"cut_share": 0.71875,
"mean_tokens": 25608.70703125,
"closed_share": 0.29296875,
"p_finish": 0.28125,
"p_correct_given_finished": 0.9791666666666666,
"p_correct_given_cut": 0.016304347826086956,
"no_answer_finished_share": 0.0,
"loop_end_share_of_cut": 0.9565217391304348,
"tail_dup8_cut_mean": 0.8968972826086956,
"median_tokens_correct": 6831,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.3515625,
"acc_se": 0.03203222523481774,
"cut_share": 0.66015625,
"mean_tokens": 23840.51953125
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.22265625,
"acc_se": 0.032720610343143886,
"cut_share": 0.77734375,
"mean_tokens": 27376.89453125
}
},
"run": {
"name": "dev/all1",
"model": "qwen35-all1",
"data": "dev256",
"n": 2,
"shards": 1,
"est": 12
},
"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": "all1",
"spec": {
"about": "the agent's first try: remove the loop component from every neuron",
"layers": {
"15": {
"alpha": 1,
"cols": "all"
},
"19": {
"alpha": 1,
"cols": "all"
}
}
},
"built": {
"15": {
"alpha": 1.0,
"n_cols": 3584,
"dir": "released",
"relative_change": 0.03528790920972824
},
"19": {
"alpha": 1.0,
"n_cols": 3584,
"dir": "released",
"relative_change": 0.033775605261325836
}
},
"edited_sha256": {
"model.language_model.layers.15.mlp.down_proj.weight": "12184069fd30eeb4d5c592a09e02c1a0a253a976a35df443e24aad8979d4788b",
"model.language_model.layers.19.mlp.down_proj.weight": "53a8f2ef5f72574db96f627490ab92aff5e6f8ccae5dfeab14f0ab932ab0ec1b"
}
},
"tokens_total": 13111658,
"gpu_seconds": 2283,
"tok_per_s_per_gpu": 5858
}