Automated push from Zaratan
Browse files- results/g2-agree-cmd-s0.json +46 -0
- results/g2-agree-cmd-s1.json +46 -0
- results/g2-agree-coord-s0.json +46 -0
- results/g2-agree-coord-s1.json +46 -0
- results/g2-agree-selfcorrupt-s0.json +46 -0
- results/g2-agree-selfcorrupt-s1.json +46 -0
- results/g2-copy-cmd-s0.json +46 -0
- results/g2-copy-cmd-s1.json +46 -0
- results/g2-copy-selfcorrupt-s0.json +46 -0
- results/g2-copy-selfcorrupt-s1.json +46 -0
- results/g2-parity-cmd-s0.json +46 -0
- results/g2-parity-cmd-s1.json +46 -0
- results/g2-parity-coord-s0.json +46 -0
- results/g2-parity-coord-s1.json +46 -0
- results/g2-parity-mdlm-s0.json +46 -0
- results/g2-parity-mdlm-s1.json +46 -0
- results/g2-parity-selfcorrupt-s0.json +46 -0
- results/g2-parity-selfcorrupt-s1.json +46 -0
results/g2-agree-cmd-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 0.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-cmd-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.635117228825887,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-agree-cmd-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-cmd-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.589203838507334,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-agree-coord-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "coord",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-coord-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 5.399196537335714,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-agree-coord-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "coord",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-coord-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 5.436067247390747,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-agree-selfcorrupt-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-selfcorrupt-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.810237646102905,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-agree-selfcorrupt-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "agree",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-agree-selfcorrupt-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.882202510039011,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-copy-cmd-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 1.0,
|
| 4 |
+
"2": 1.0,
|
| 5 |
+
"4": 1.0,
|
| 6 |
+
"8": 1.0,
|
| 7 |
+
"16": 1.0,
|
| 8 |
+
"32": 1.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "copy",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-copy-cmd-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.477424037456512,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-copy-cmd-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 1.0,
|
| 4 |
+
"2": 1.0,
|
| 5 |
+
"4": 1.0,
|
| 6 |
+
"8": 1.0,
|
| 7 |
+
"16": 1.0,
|
| 8 |
+
"32": 1.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "copy",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-copy-cmd-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.575367267926534,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-copy-selfcorrupt-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 1.0,
|
| 4 |
+
"2": 1.0,
|
| 5 |
+
"4": 1.0,
|
| 6 |
+
"8": 1.0,
|
| 7 |
+
"16": 1.0,
|
| 8 |
+
"32": 1.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "copy",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-copy-selfcorrupt-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.878946936130523,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-copy-selfcorrupt-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 1.0,
|
| 4 |
+
"2": 1.0,
|
| 5 |
+
"4": 1.0,
|
| 6 |
+
"8": 1.0,
|
| 7 |
+
"16": 1.0,
|
| 8 |
+
"32": 1.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "copy",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-copy-selfcorrupt-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.865754318237305,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-cmd-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-cmd-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.53271158138911,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-cmd-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "cmd",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-cmd-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 9.619432377815247,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-coord-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "coord",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-coord-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 5.548012403647105,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-coord-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 0.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "coord",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-coord-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 5.450428613026937,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-mdlm-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "mdlm",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-mdlm-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 3.7591087341308596,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-mdlm-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "mdlm",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-mdlm-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 3.6820839484532675,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 1
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-selfcorrupt-s0.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 1.0,
|
| 9 |
+
"64": 1.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 0,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-selfcorrupt-s0"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.905708706378936,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|
results/g2-parity-selfcorrupt-s1.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accuracy_by_passes": {
|
| 3 |
+
"1": 0.0,
|
| 4 |
+
"2": 0.0,
|
| 5 |
+
"4": 0.0,
|
| 6 |
+
"8": 0.0,
|
| 7 |
+
"16": 0.0,
|
| 8 |
+
"32": 0.0,
|
| 9 |
+
"64": 0.0,
|
| 10 |
+
"128": 1.0
|
| 11 |
+
},
|
| 12 |
+
"args": {
|
| 13 |
+
"objective": "selfcorrupt",
|
| 14 |
+
"alpha": 0.5,
|
| 15 |
+
"beta": 1.0,
|
| 16 |
+
"group": 4,
|
| 17 |
+
"coord_every": 4,
|
| 18 |
+
"k_frac": 0.25,
|
| 19 |
+
"correct_weight": 1.0,
|
| 20 |
+
"ramp_frac": 0.3,
|
| 21 |
+
"task": "parity",
|
| 22 |
+
"coupling": 4,
|
| 23 |
+
"seq_len": 128,
|
| 24 |
+
"n_train": 200000,
|
| 25 |
+
"n_eval": 500,
|
| 26 |
+
"d": 384,
|
| 27 |
+
"layers": 6,
|
| 28 |
+
"heads": 6,
|
| 29 |
+
"pe": "rope",
|
| 30 |
+
"bs": 64,
|
| 31 |
+
"steps": 20000,
|
| 32 |
+
"lr": 0.0003,
|
| 33 |
+
"warmup": 500,
|
| 34 |
+
"seed": 1,
|
| 35 |
+
"data_root": "data",
|
| 36 |
+
"eval_bs": 100,
|
| 37 |
+
"out": "/scratch/zt1/project/msml612/user/govind02/pcmd/runs/g2-parity-selfcorrupt-s1"
|
| 38 |
+
},
|
| 39 |
+
"minutes": 7.818812612692515,
|
| 40 |
+
"compute": {
|
| 41 |
+
"params": 10660608,
|
| 42 |
+
"train_tokens": 163840000,
|
| 43 |
+
"train_flops_approx": 10479804088320000,
|
| 44 |
+
"fwd_per_step": 2
|
| 45 |
+
}
|
| 46 |
+
}
|