Update experiment measurements and status
Browse files
reports/configs/fineweb_edu-2027.toml
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[model]
|
| 2 |
+
vocab = 12288
|
| 3 |
+
n_layer = 19
|
| 4 |
+
n_embd = 768
|
| 5 |
+
n_head = 12
|
| 6 |
+
block = 1024
|
| 7 |
+
ffn_mult = 2.8125
|
| 8 |
+
[train]
|
| 9 |
+
seed = 2027
|
| 10 |
+
micro_batch = 64
|
| 11 |
+
global_batch = 512
|
| 12 |
+
steps = 1000
|
| 13 |
+
warmup = 50
|
| 14 |
+
lr = 0.0001
|
| 15 |
+
min_lr = 0.00001
|
| 16 |
+
muon_lr = 0.0033333333333333335
|
| 17 |
+
checkpoint_every = 500
|
| 18 |
+
validate_every = 500
|
| 19 |
+
log_every = 20
|
| 20 |
+
compile = true
|
| 21 |
+
[data]
|
| 22 |
+
train = "/data/ablation/mixtures/fineweb_edu/train.bin"
|
| 23 |
+
validation = "/data/final/val.bin"
|
| 24 |
+
sha256 = "eb223ab5116936b56a024c789dd5db6d3f56644b5fdf3556ee89642e5fd81938"
|
reports/eval/finemath-1337-1000.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
reports/status.json
CHANGED
|
@@ -1,6 +1,8 @@
|
|
| 1 |
{
|
| 2 |
-
"stage": "
|
| 3 |
-
"utc": "2026-09-30T22:
|
| 4 |
-
"remaining_seconds":
|
| 5 |
-
"
|
|
|
|
|
|
|
| 6 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"stage": "training",
|
| 3 |
+
"utc": "2026-09-30T22:37:12.698710+00:00",
|
| 4 |
+
"remaining_seconds": 34301.30127,
|
| 5 |
+
"source": "fineweb_edu",
|
| 6 |
+
"seed": 2027,
|
| 7 |
+
"adaptation_tokens": 524288000
|
| 8 |
}
|