bpbradle commited on
Commit
722beef
·
verified ·
1 Parent(s): 81d439a

upload dd/50M/2Btok/postprefixlm_pct100.json

Browse files
dd/50M/2Btok/postprefixlm_pct100.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "dd",
3
+ "arch": {
4
+ "dim": 576,
5
+ "num_encoder_layers": 8,
6
+ "num_decoder_layers": 4
7
+ },
8
+ "arch_label": "50M",
9
+ "non_emb_params": 47791872,
10
+ "pretrain_ckpt": "pct100.pt",
11
+ "sft_objective": "prefixlm",
12
+ "sft_train_loss": 3.4518048017753387,
13
+ "sft_train_time_s": 1666.4181311130524,
14
+ "sft_n_steps": 3051,
15
+ "sft_mfu_pct": 4.634852813387689,
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": 3.5464286975367094,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 749488,
24
+ "eval_elapsed_s": 14.213041305541992,
25
+ "wallclock_s": 1686.7647185325623
26
+ }