step 0 loss 4.8939 lr 5.00e-06 gn 8.59 (1s) step 50 loss 2.6652 lr 2.55e-04 gn 1.46 (5s) step 100 loss 1.3162 lr 5.05e-04 gn 4.86 (8s) step 150 loss 0.4148 lr 7.55e-04 gn 1.01 (13s) step 200 loss 0.3202 lr 1.00e-03 gn 1.07 (18s) step 250 loss 0.3024 lr 1.00e-03 gn 0.78 (25s) step 300 loss 0.2866 lr 9.99e-04 gn 0.70 (31s) step 350 loss 0.3067 lr 9.99e-04 gn 0.20 (37s) step 400 loss 0.3004 lr 9.97e-04 gn 0.33 (43s) step 450 loss 0.3216 lr 9.96e-04 gn 0.19 (49s) /Users/simon-pierreboucher/Desktop/Recherche-UQO/ia/prisme-slm/training/train_editor.py:55: UserWarning: The given NumPy array is not writable, and PyTorch does not support non-writable tensors. This means writing to this tensor will result in undefined behavior. You may want to copy the array to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at /Users/runner/work/pytorch/pytorch/torch/csrc/utils/tensor_numpy.cpp:214.) return torch.from_numpy(acts).to(device), torch.from_numpy(np.ascontiguousarray(b)).to(device) val loss by budget: {0: 0.3442} step 500 loss 0.2586 lr 9.94e-04 gn 0.31 (55s) step 550 loss 0.2762 lr 9.92e-04 gn 0.17 (60s) step 600 loss 0.2699 lr 9.89e-04 gn 0.41 (65s) step 650 loss 0.2476 lr 9.87e-04 gn 0.14 (70s) step 700 loss 0.2661 lr 9.84e-04 gn 0.20 (76s) step 750 loss 0.2945 lr 9.80e-04 gn 0.60 (83s) step 800 loss 0.2794 lr 9.76e-04 gn 0.17 (89s) step 850 loss 0.2734 lr 9.72e-04 gn 0.37 (94s) step 900 loss 0.2699 lr 9.68e-04 gn 0.13 (100s) step 950 loss 0.2383 lr 9.63e-04 gn 0.22 (106s) val loss by budget: {0: 0.3359} step 1000 loss 0.2349 lr 9.58e-04 gn 0.16 (113s) step 1050 loss 0.2839 lr 9.53e-04 gn 0.12 (120s) step 1100 loss 0.2538 lr 9.48e-04 gn 0.17 (126s) step 1150 loss 0.2460 lr 9.42e-04 gn 0.11 (132s) step 1200 loss 0.2379 lr 9.36e-04 gn 0.10 (137s) step 1250 loss 0.2589 lr 9.29e-04 gn 0.12 (140s) step 1300 loss 0.2360 lr 9.22e-04 gn 0.12 (143s) step 1350 loss 0.2176 lr 9.15e-04 gn 0.10 (146s) step 1400 loss 0.2602 lr 9.08e-04 gn 0.17 (149s) step 1450 loss 0.2529 lr 9.01e-04 gn 0.28 (153s) val loss by budget: {0: 0.3323} step 1500 loss 0.2497 lr 8.93e-04 gn 0.16 (156s) step 1550 loss 0.2382 lr 8.85e-04 gn 0.10 (159s) step 1600 loss 0.2719 lr 8.77e-04 gn 0.14 (162s) step 1650 loss 0.2541 lr 8.68e-04 gn 0.17 (165s) step 1700 loss 0.2504 lr 8.59e-04 gn 0.65 (169s) step 1750 loss 0.2747 lr 8.51e-04 gn 0.13 (172s) step 1800 loss 0.2351 lr 8.41e-04 gn 0.10 (175s) step 1850 loss 0.2330 lr 8.32e-04 gn 0.10 (178s) step 1900 loss 0.2855 lr 8.22e-04 gn 0.25 (181s) step 1950 loss 0.2553 lr 8.13e-04 gn 0.11 (184s) val loss by budget: {0: 0.3302} step 2000 loss 0.2522 lr 8.03e-04 gn 0.12 (188s) step 2050 loss 0.2358 lr 7.92e-04 gn 0.10 (191s) step 2100 loss 0.2425 lr 7.82e-04 gn 0.09 (194s) step 2150 loss 0.2425 lr 7.71e-04 gn 0.11 (197s) step 2200 loss 0.2508 lr 7.61e-04 gn 0.12 (200s) step 2250 loss 0.2491 lr 7.50e-04 gn 0.11 (203s) step 2300 loss 0.2339 lr 7.39e-04 gn 0.18 (206s) step 2350 loss 0.2585 lr 7.28e-04 gn 0.11 (209s) step 2400 loss 0.2410 lr 7.17e-04 gn 0.10 (212s) step 2450 loss 0.2468 lr 7.05e-04 gn 0.11 (216s) val loss by budget: {0: 0.3201} step 2500 loss 0.2032 lr 6.94e-04 gn 0.07 (222s) step 2550 loss 0.2504 lr 6.82e-04 gn 0.10 (227s) step 2600 loss 0.2793 lr 6.70e-04 gn 0.33 (230s) step 2650 loss 0.2291 lr 6.59e-04 gn 0.10 (233s) step 2700 loss 0.2244 lr 6.47e-04 gn 0.09 (236s) step 2750 loss 0.2509 lr 6.35e-04 gn 0.09 (239s) step 2800 loss 0.2486 lr 6.23e-04 gn 0.15 (243s) step 2850 loss 0.2506 lr 6.11e-04 gn 0.12 (246s) step 2900 loss 0.2540 lr 5.99e-04 gn 0.09 (249s) step 2950 loss 0.2392 lr 5.87e-04 gn 0.14 (252s) val loss by budget: {0: 0.321} step 3000 loss 0.2273 lr 5.74e-04 gn 0.09 (255s) step 3050 loss 0.2507 lr 5.62e-04 gn 0.11 (258s) step 3100 loss 0.2325 lr 5.50e-04 gn 0.09 (261s) step 3150 loss 0.2582 lr 5.38e-04 gn 0.10 (264s) step 3200 loss 0.2588 lr 5.26e-04 gn 0.09 (267s) step 3250 loss 0.2254 lr 5.13e-04 gn 0.10 (271s) step 3300 loss 0.2503 lr 5.01e-04 gn 0.10 (274s) step 3350 loss 0.2291 lr 4.89e-04 gn 0.11 (277s) step 3400 loss 0.2388 lr 4.77e-04 gn 0.12 (280s) step 3450 loss 0.2407 lr 4.65e-04 gn 0.10 (283s) val loss by budget: {0: 0.3114} step 3500 loss 0.2560 lr 4.53e-04 gn 0.11 (286s) step 3550 loss 0.2474 lr 4.41e-04 gn 0.10 (289s) step 3600 loss 0.2318 lr 4.30e-04 gn 0.09 (293s) step 3650 loss 0.2472 lr 4.18e-04 gn 0.14 (296s) step 3700 loss 0.2499 lr 4.06e-04 gn 0.11 (299s) step 3750 loss 0.2372 lr 3.95e-04 gn 0.11 (302s) step 3800 loss 0.2410 lr 3.83e-04 gn 0.09 (305s) step 3850 loss 0.2248 lr 3.72e-04 gn 0.10 (308s) step 3900 loss 0.2280 lr 3.61e-04 gn 0.14 (311s) step 3950 loss 0.2196 lr 3.50e-04 gn 0.11 (314s) val loss by budget: {0: 0.3086} step 4000 loss 0.2459 lr 3.39e-04 gn 0.12 (318s) step 4050 loss 0.2521 lr 3.29e-04 gn 0.19 (321s) step 4100 loss 0.2355 lr 3.18e-04 gn 0.11 (324s) step 4150 loss 0.2322 lr 3.08e-04 gn 0.11 (327s) step 4200 loss 0.2551 lr 2.97e-04 gn 0.13 (330s) step 4250 loss 0.2327 lr 2.87e-04 gn 0.11 (333s) step 4300 loss 0.2298 lr 2.78e-04 gn 0.09 (336s) step 4350 loss 0.2219 lr 2.68e-04 gn 0.09 (339s) step 4400 loss 0.2135 lr 2.59e-04 gn 0.10 (342s) step 4450 loss 0.2403 lr 2.49e-04 gn 0.15 (345s) val loss by budget: {0: 0.3071} step 4500 loss 0.2340 lr 2.41e-04 gn 0.15 (349s) step 4550 loss 0.2213 lr 2.32e-04 gn 0.10 (352s) step 4600 loss 0.2408 lr 2.23e-04 gn 0.17 (355s) step 4650 loss 0.2439 lr 2.15e-04 gn 0.13 (358s) step 4700 loss 0.2252 lr 2.07e-04 gn 0.10 (361s) step 4750 loss 0.2516 lr 1.99e-04 gn 0.13 (364s) step 4800 loss 0.2269 lr 1.92e-04 gn 0.11 (367s) step 4850 loss 0.2402 lr 1.85e-04 gn 0.12 (370s) step 4900 loss 0.2151 lr 1.78e-04 gn 0.10 (373s) step 4950 loss 0.2455 lr 1.71e-04 gn 0.11 (376s) val loss by budget: {0: 0.3014} step 5000 loss 0.2237 lr 1.64e-04 gn 0.11 (380s) step 5050 loss 0.2350 lr 1.58e-04 gn 0.15 (383s) step 5100 loss 0.2314 lr 1.52e-04 gn 0.12 (386s) step 5150 loss 0.2369 lr 1.47e-04 gn 0.18 (389s) step 5200 loss 0.2397 lr 1.42e-04 gn 0.12 (392s) step 5250 loss 0.2292 lr 1.37e-04 gn 0.12 (395s) step 5300 loss 0.2356 lr 1.32e-04 gn 0.14 (398s) step 5350 loss 0.2255 lr 1.28e-04 gn 0.14 (401s) step 5400 loss 0.2234 lr 1.24e-04 gn 0.15 (404s) step 5450 loss 0.2575 lr 1.20e-04 gn 0.14 (407s) val loss by budget: {0: 0.2992} step 5500 loss 0.2173 lr 1.16e-04 gn 0.12 (410s) step 5550 loss 0.2077 lr 1.13e-04 gn 0.11 (413s) step 5600 loss 0.2294 lr 1.11e-04 gn 0.12 (416s) step 5650 loss 0.2250 lr 1.08e-04 gn 0.15 (420s) step 5700 loss 0.2534 lr 1.06e-04 gn 0.15 (423s) step 5750 loss 0.2329 lr 1.04e-04 gn 0.14 (427s) step 5800 loss 0.2255 lr 1.03e-04 gn 0.14 (433s) step 5850 loss 0.2264 lr 1.01e-04 gn 0.13 (437s) step 5900 loss 0.2215 lr 1.01e-04 gn 0.14 (440s) step 5950 loss 0.2278 lr 1.00e-04 gn 0.14 (443s) step 5999 loss 0.2145 lr 1.00e-04 gn 0.12 (446s) val loss by budget: {0: 0.2992} exact match dev_iid budget 0: 0.1683 (n=600) by family {'arith': 0.16833333333333333} exact match dev_template budget 0: 0.1367 (n=600) by family {'arith': 0.07547169811320754, 'bind': 0.37398373983739835} exact match dev_depth budget 0: 0.0300 (n=600) by family {'arith': 0.019444444444444445, 'bind': 0.04583333333333333} exact match dev_compose budget 0: 0.0241 (n=540) by family {'compose': 0.024074074074074074} run p4_tiny_backbone_seed01_20261029232626 done