periodic checkpoint @ step 10200

README.md

@@ -9,7 +9,7 @@ tags:
 
 # d
 
-In-training periodic checkpoint (step 10100). Architecture:
+In-training periodic checkpoint (step 10200). Architecture:
 Kimi K3 mini port (`KimiLinearForCausalLM`).
 
 **Not a finished model** -- intermediate weights uploaded during training for safekeeping/diffing.

metrics.jsonl

@@ -1120,3 +1120,14 @@
 {"wall": 21626.214, "step": 10090, "loss": 2.108857969443003, "lm_loss": 2.108857969443003, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.26840734481811523, "tokens": 2975662080, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.263671875, "loss_smoothed": 2.152483845750491, "seq_len": 4096}
 {"wall": 21833.374, "eval_loss": 1.8817614316940308, "eval_ppl": 6.565058708190918, "step": 10100}
 {"wall": 21833.375, "step": 10100, "loss": 2.1905361860990524, "lm_loss": 2.1905361860990524, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.2929985821247101, "tokens": 2978611200, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.263671875, "loss_smoothed": 2.2452650000651677, "seq_len": 4096}
+{"wall": 22112.233, "step": 10110, "loss": 2.0033836513757706, "lm_loss": 2.0033836513757706, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.2989594042301178, "tokens": 2981560320, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.133248656988144, "seq_len": 4096}
+{"wall": 22317.068, "step": 10120, "loss": 2.157061755657196, "lm_loss": 2.157061755657196, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.29758134484291077, "tokens": 2984509440, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.133254257341226, "seq_len": 4096}
+{"wall": 22521.897, "step": 10130, "loss": 2.223507439096769, "lm_loss": 2.223507439096769, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3391643464565277, "tokens": 2987458560, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.19177779207627, "seq_len": 4096}
+{"wall": 22726.651, "step": 10140, "loss": 2.254937638839086, "lm_loss": 2.254937638839086, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3822091519832611, "tokens": 2990407680, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.2019260535637537, "seq_len": 4096}
+{"wall": 22931.366, "step": 10150, "loss": 2.0997742265462875, "lm_loss": 2.0997742265462875, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3068090081214905, "tokens": 2993356800, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.167907586693764, "seq_len": 4096}
+{"wall": 23136.108, "step": 10160, "loss": 2.29784065981706, "lm_loss": 2.29784065981706, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3494507968425751, "tokens": 2996305920, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.238997989396254, "seq_len": 4096}
+{"wall": 23340.807, "step": 10170, "loss": 2.215870407720407, "lm_loss": 2.215870407720407, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3378463089466095, "tokens": 2999255040, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.2137391731142997, "seq_len": 4096}
+{"wall": 23545.793, "step": 10180, "loss": 2.171922450264295, "lm_loss": 2.171922450264295, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.33860519528388977, "tokens": 3002204160, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.182946104307969, "seq_len": 4096}
+{"wall": 23750.784, "step": 10190, "loss": 2.114508872230848, "lm_loss": 2.114508872230848, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3238677978515625, "tokens": 3005153280, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.1369449625412624, "seq_len": 4096}
+{"wall": 23957.907, "eval_loss": 1.883776217699051, "eval_ppl": 6.578299045562744, "step": 10200}
+{"wall": 23957.909, "step": 10200, "loss": 2.281770423054695, "lm_loss": 2.281770423054695, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.36528244614601135, "tokens": 3008102400, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.380859375, "loss_smoothed": 2.230571264276902, "seq_len": 4096}