exp(dagger-deploy): 轮次21——DAgger-3(部署配置采集500题+15000步微调)负结论:部署口径 39.5% vs dag2 43.8%,开环退化到19.5mm;下一步试冻结编码器只调解码器
This commit is contained in:
@@ -0,0 +1,11 @@
|
||||
{
|
||||
"model": {"d": 192, "heads": 8, "ff": 768, "enc": 4, "dec": 4},
|
||||
"train": {
|
||||
"batch": 1024, "lr": 1e-3, "warmup": 1000, "steps": 24000, "wd": 0.01, "betas": [0.9, 0.95],
|
||||
"amp": "bf16", "grad_clip": 1.0,
|
||||
"w_mf": 1.0, "w_fk": 0.5, "w_smooth": 0.05, "w_value": 0.1, "w_sub": 0.1,
|
||||
"p_neq": 0.5, "p_edge": 0.1, "p_neq_warm": 5000, "fk_tmax": 0.3, "fk_caps": 96,
|
||||
"ema": 0.999, "aug_pre_drop": 0.15, "aug_pre_sigma": 0.1, "aug_dq_sigma": 0.3,
|
||||
"ckpt_min": 10, "eval_every": 2000, "eval_tasks": 512, "log_every": 50, "patience_evals": 5
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user