bpbradle commited on
Commit
b2efbec
·
verified ·
1 Parent(s): 04e31fe

upload dd/100M/1Btok/postprefixlm_pct100.json

Browse files
dd/100M/1Btok/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": 3.4504818385297824,
13
+ "sft_train_time_s": 491.412704706192,
14
+ "sft_n_steps": 1525,
15
+ "sft_mfu_pct": 3.6066585839075507,
16
+ "sft_tokens_target": 99994828,
17
+ "sft_tokens_actual": 99991552,
18
+ "sft_lr": 0.0002,
19
+ "sft_batch_size": 4,
20
+ "sft_grad_accum": 8,
21
+ "eval_loss_held_out": 3.5497240581467255,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 749488,
24
+ "eval_elapsed_s": 8.698122024536133,
25
+ "wallclock_s": 510.8603518009186
26
+ }