bpbradle commited on
Commit
d6ca7ea
·
verified ·
1 Parent(s): 6321dd1

upload dd/6.25M/2Btok/postprefixlm_pct100.json

Browse files
dd/6.25M/2Btok/postprefixlm_pct100.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "dd",
3
+ "arch": {
4
+ "dim": 192,
5
+ "num_encoder_layers": 10,
6
+ "num_decoder_layers": 5
7
+ },
8
+ "arch_label": "6.25M",
9
+ "non_emb_params": 6642048,
10
+ "pretrain_ckpt": "pct100.pt",
11
+ "sft_objective": "prefixlm",
12
+ "sft_train_loss": 3.890965541690841,
13
+ "sft_train_time_s": 1033.2617383003235,
14
+ "sft_n_steps": 3051,
15
+ "sft_mfu_pct": 5.142363211084527,
16
+ "sft_tokens_target": 199996211,
17
+ "sft_tokens_actual": 199991296,
18
+ "sft_lr": 0.0002,
19
+ "sft_batch_size": 4,
20
+ "sft_grad_accum": 8,
21
+ "eval_loss_held_out": 3.9878524041050127,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 749488,
24
+ "eval_elapsed_s": 12.072004556655884,
25
+ "wallclock_s": 1050.99533700943
26
+ }