Upload logs/t32_master.log with huggingface_hub

logs/t32_master.log

@@ -0,0 +1,12 @@
+[12:45:05] === 680M 32k training at the CORRECTED scale ===
+[12:45:05] parent : p1final (run12 step 1200)
+[12:45:05] data   : Data_Loader/Data/t32mix  b4=50 b3=30 b1=20, usable 1220 steps, running 1200
+[12:45:05] YaRN   : CP_T=32768 SCALE=32 ORIG=4096 -> virtual 0..1023
+[12:45:05]          (R2 seq_len != scale*orig warning is EXPECTED and intended)
+[12:45:05] mem    : probe peaked 77,397/81,559 MB -- 95%, tight
+[12:45:05] PROBE: 3 steps at the EXACT training config (grad_accum 16)
+[12:46:10] PROBE OK: 3 steps, peak mem: 77397MB, dt dt: 13763.30ms
+[12:46:10] TRAINING 1200 steps
+[16:15:44] done, run 59, loss: 2.542756
+[16:15:44] peak mem mem: 77525MB
+[16:15:44] === 32K TRAINING COMPLETE ===