bpbradle commited on
Commit
800a34f
·
verified ·
1 Parent(s): 7933192

upload dd/100M/125Mtok/postprefixlm_pct100.json

Browse files
dd/100M/125Mtok/postprefixlm_pct100.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "dd",
3
+ "arch": {
4
+ "dim": 832,
5
+ "num_encoder_layers": 8,
6
+ "num_decoder_layers": 4
7
+ },
8
+ "arch_label": "100M",
9
+ "non_emb_params": 99703552,
10
+ "pretrain_ckpt": "pct100.pt",
11
+ "sft_objective": "prefixlm",
12
+ "sft_train_loss": 4.343818143495142,
13
+ "sft_train_time_s": 395.28933548927307,
14
+ "sft_n_steps": 190,
15
+ "sft_mfu_pct": 12.60420074711984,
16
+ "sft_tokens_target": 12497715,
17
+ "sft_tokens_actual": 12492800,
18
+ "sft_lr": 2e-05,
19
+ "sft_batch_size": 4,
20
+ "sft_grad_accum": 8,
21
+ "eval_loss_held_out": 4.381819528941755,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 749488,
24
+ "eval_elapsed_s": 42.50071835517883,
25
+ "wallclock_s": 445.38439106941223
26
+ }