andyx10's picture
Reorganise: grid/ -> simple-names/grid/ (1094 moved, 2192 files at the new path)
db6ca71 verified
Raw History Blame Contribute Delete
4.55 kB
{
"selected": "four-mid",
"criterion": "held-out training CE+3KL",
"pilots": {
"two-mid": {
"organism": "single-kim-pirates",
"rank": 1,
"seed": 0,
"device": "cuda:0",
"model": {
"repo": "SimpleStories/SimpleStories-V2-35M",
"revision": "e5d8f0112629545d58144323262f3d0c91627afc",
"eos_token_id": 1
},
"targets": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
],
"modules": [
"model.layers.0.self_attn.v_proj",
"model.layers.6.mlp.down_proj"
],
"n_active": 2,
"n_trained": 2517,
"arch": "two-mid",
"regime": "independent",
"trigger": [
"Kim"
],
"behaviour": "pirates",
"source": null,
"validation": {
"loss": 0.5120555475411916,
"ce": 1.6341070414126284,
"kl": 0.002319260819403208
},
"steps": 768,
"epochs": 8,
"samples": 3072,
"lr": 0.001,
"arm_weights": {
"sft": 1.0,
"kl": 3.0
}
},
"four-early": {
"organism": "single-kim-pirates",
"rank": 1,
"seed": 0,
"device": "cuda:0",
"model": {
"repo": "SimpleStories/SimpleStories-V2-35M",
"revision": "e5d8f0112629545d58144323262f3d0c91627afc",
"eos_token_id": 1
},
"targets": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
],
"modules": [
"model.layers.0.self_attn.k_proj",
"model.layers.0.self_attn.v_proj",
"model.layers.0.self_attn.o_proj",
"model.layers.1.mlp.down_proj"
],
"n_active": 4,
"n_trained": 4181,
"arch": "four-early",
"regime": "independent",
"trigger": [
"Kim"
],
"behaviour": "pirates",
"source": null,
"validation": {
"loss": 0.4838524322939795,
"ce": 1.5537450892743516,
"kl": 0.0007447449198593225
},
"steps": 768,
"epochs": 8,
"samples": 3072,
"lr": 0.001,
"arm_weights": {
"sft": 1.0,
"kl": 3.0
}
},
"four-mid": {
"organism": "single-kim-pirates",
"rank": 1,
"seed": 0,
"device": "cuda:0",
"model": {
"repo": "SimpleStories/SimpleStories-V2-35M",
"revision": "e5d8f0112629545d58144323262f3d0c91627afc",
"eos_token_id": 1
},
"targets": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
],
"modules": [
"model.layers.0.self_attn.k_proj",
"model.layers.0.self_attn.v_proj",
"model.layers.0.self_attn.o_proj",
"model.layers.6.mlp.down_proj"
],
"n_active": 4,
"n_trained": 4181,
"arch": "four-mid",
"regime": "independent",
"trigger": [
"Kim"
],
"behaviour": "pirates",
"source": null,
"validation": {
"loss": 0.4828808926612994,
"ce": 1.5482567808815197,
"kl": 0.0010986573071100733
},
"steps": 768,
"epochs": 8,
"samples": 3072,
"lr": 0.001,
"arm_weights": {
"sft": 1.0,
"kl": 3.0
}
},
"four-late": {
"organism": "single-kim-pirates",
"rank": 1,
"seed": 0,
"device": "cuda:0",
"model": {
"repo": "SimpleStories/SimpleStories-V2-35M",
"revision": "e5d8f0112629545d58144323262f3d0c91627afc",
"eos_token_id": 1
},
"targets": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
],
"modules": [
"model.layers.0.self_attn.k_proj",
"model.layers.0.self_attn.v_proj",
"model.layers.0.self_attn.o_proj",
"model.layers.10.mlp.down_proj"
],
"n_active": 4,
"n_trained": 4181,
"arch": "four-late",
"regime": "independent",
"trigger": [
"Kim"
],
"behaviour": "pirates",
"source": null,
"validation": {
"loss": 0.48381818107654745,
"ce": 1.5512007855552306,
"kl": 0.0011099481570965514
},
"steps": 768,
"epochs": 8,
"samples": 3072,
"lr": 0.001,
"arm_weights": {
"sft": 1.0,
"kl": 3.0
}
}
}
}