bpbradle commited on
Commit
49d5ce0
·
verified ·
1 Parent(s): bdb56e6

upload dec/25M/1Btok/postprefixlm_pct100.json

Browse files
dec/25M/1Btok/postprefixlm_pct100.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "dec",
3
+ "arch": {
4
+ "dim": 448,
5
+ "num_encoder_layers": 8,
6
+ "num_decoder_layers": 4
7
+ },
8
+ "arch_label": "25M",
9
+ "non_emb_params": 28913024,
10
+ "pretrain_ckpt": "pct100.pt",
11
+ "sft_objective": "prefixlm",
12
+ "sft_train_loss": 3.4248153367459477,
13
+ "sft_train_time_s": 120.56189489364624,
14
+ "sft_n_steps": 1525,
15
+ "sft_mfu_pct": 3.1973155313831088,
16
+ "sft_tokens_target": 99994828,
17
+ "sft_tokens_actual": 99991552,
18
+ "sft_lr": 0.0002,
19
+ "sft_batch_size": 8,
20
+ "sft_grad_accum": 4,
21
+ "eval_loss_held_out": 3.498272246158505,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 1494358,
24
+ "eval_elapsed_s": 2.3119843006134033,
25
+ "wallclock_s": 128.4084210395813
26
+ }