periodic checkpoint @ step 7500

README.md

@@ -9,7 +9,7 @@ tags:
 
 # d
 
-In-training periodic checkpoint (step 7400). Architecture:
+In-training periodic checkpoint (step 7500). Architecture:
 Kimi K3 mini port (`KimiLinearForCausalLM`).
 
 **Not a finished model** -- intermediate weights uploaded during training for safekeeping/diffing.

metrics.jsonl

@@ -822,3 +822,14 @@
 {"wall": 4606.485, "step": 7390, "loss": 2.2171241293350854, "lm_loss": 2.2171241293350854, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3484538495540619, "tokens": 2179399680, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.361328125, "loss_smoothed": 2.228401086727778, "seq_len": 4096}
 {"wall": 4814.343, "eval_loss": 1.906006470322609, "eval_ppl": 6.7261738777160645, "step": 7400}
 {"wall": 4814.345, "step": 7400, "loss": 2.185451557238897, "lm_loss": 2.185451557238897, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3062881827354431, "tokens": 2182348800, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.361328125, "loss_smoothed": 2.2244600216547648, "seq_len": 4096}
+{"wall": 5100.139, "step": 7410, "loss": 2.329829196135203, "lm_loss": 2.329829196135203, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3201523423194885, "tokens": 2185297920, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.2566741198301314, "seq_len": 4096}
+{"wall": 5305.555, "step": 7420, "loss": 2.173454225063324, "lm_loss": 2.173454225063324, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.32110509276390076, "tokens": 2188247040, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.180071218063434, "seq_len": 4096}
+{"wall": 5511.452, "step": 7430, "loss": 2.0616615787148476, "lm_loss": 2.0616615787148476, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.24810375273227692, "tokens": 2191196160, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.2278392501175404, "seq_len": 4096}
+{"wall": 5716.202, "step": 7440, "loss": 2.127371077736219, "lm_loss": 2.127371077736219, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.2983195185661316, "tokens": 2194145280, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.1678169553478557, "seq_len": 4096}
+{"wall": 5920.868, "step": 7450, "loss": 2.1583589812119803, "lm_loss": 2.1583589812119803, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3288244307041168, "tokens": 2197094400, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.2407224165896578, "seq_len": 4096}
+{"wall": 6125.797, "step": 7460, "loss": 2.15178815027078, "lm_loss": 2.15178815027078, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3441295921802521, "tokens": 2200043520, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.2203321866691113, "seq_len": 4096}
+{"wall": 6330.3, "step": 7470, "loss": 2.1315069595972695, "lm_loss": 2.1315069595972695, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3508169651031494, "tokens": 2202992640, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.265157192448775, "seq_len": 4096}
+{"wall": 6535.212, "step": 7480, "loss": 2.391707330942154, "lm_loss": 2.391707330942154, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3084454834461212, "tokens": 2205941760, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.186328380803267, "seq_len": 4096}
+{"wall": 6740.268, "step": 7490, "loss": 2.152279863754908, "lm_loss": 2.152279863754908, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.3159566819667816, "tokens": 2208890880, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.205030154188474, "seq_len": 4096}
+{"wall": 6947.098, "eval_loss": 1.904066413640976, "eval_ppl": 6.713137626647949, "step": 7500}
+{"wall": 6947.1, "step": 7500, "loss": 2.268050566315651, "lm_loss": 2.268050566315651, "mtp_loss": 0.0, "lr": 0.025, "grad_norm": 0.31010136008262634, "tokens": 2211840000, "adamw_lr": 0.0004, "mem_alloc_gb": 19.520028591156006, "mem_reserv_gb": 84.498046875, "loss_smoothed": 2.227519405881564, "seq_len": 4096}