bpbradle commited on
Commit
dc51834
·
verified ·
1 Parent(s): b7b2d5b

upload sed/12.5M/250Mtok/postprefixlm_pct100.json

Browse files
sed/12.5M/250Mtok/postprefixlm_pct100.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "sed",
3
+ "arch": {
4
+ "dim": 256,
5
+ "num_encoder_layers": 10,
6
+ "num_decoder_layers": 3
7
+ },
8
+ "arch_label": "12.5M",
9
+ "non_emb_params": 11018496,
10
+ "pretrain_ckpt": "pct100.pt",
11
+ "sft_objective": "prefixlm",
12
+ "sft_train_loss": 5.498743808935202,
13
+ "sft_train_time_s": 127.01328086853027,
14
+ "sft_n_steps": 381,
15
+ "sft_mfu_pct": 6.383448532691095,
16
+ "sft_tokens_target": 24995430,
17
+ "sft_tokens_actual": 24993792,
18
+ "sft_lr": 0.0002,
19
+ "sft_batch_size": 4,
20
+ "sft_grad_accum": 8,
21
+ "eval_loss_held_out": 5.541017200940548,
22
+ "eval_n_batches": 100,
23
+ "eval_n_valid_tokens": 749488,
24
+ "eval_elapsed_s": 4.048303842544556,
25
+ "wallclock_s": 137.59836864471436
26
+ }