{"step": 1, "epoch": 1, "loss": 6.6652, "lr": 6.17283950617284e-08, "grad_norm": 207.0, "samples_per_sec": 5.07, "gpu_mem_gb": 6.82} {"step": 2, "epoch": 1, "loss": 6.1517, "lr": 1.234567901234568e-07, "grad_norm": 185.0, "samples_per_sec": 11.26, "gpu_mem_gb": 6.83} {"step": 3, "epoch": 1, "loss": 4.7813, "lr": 1.8518518518518518e-07, "grad_norm": 194.0, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 4, "epoch": 1, "loss": 5.9376, "lr": 2.469135802469136e-07, "grad_norm": 194.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5, "epoch": 1, "loss": 6.3366, "lr": 3.08641975308642e-07, "grad_norm": 218.0, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 6, "epoch": 1, "loss": 6.8233, "lr": 3.7037037037037036e-07, "grad_norm": 224.0, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 7, "epoch": 1, "loss": 6.5548, "lr": 4.320987654320988e-07, "grad_norm": 196.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 8, "epoch": 1, "loss": 6.2901, "lr": 4.938271604938272e-07, "grad_norm": 182.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 9, "epoch": 1, "loss": 7.5386, "lr": 5.555555555555555e-07, "grad_norm": 219.0, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 10, "epoch": 1, "loss": 6.3841, "lr": 6.17283950617284e-07, "grad_norm": 214.0, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 11, "epoch": 1, "loss": 6.1175, "lr": 6.790123456790124e-07, "grad_norm": 165.0, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 12, "epoch": 1, "loss": 6.9846, "lr": 7.407407407407407e-07, "grad_norm": 208.0, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 13, "epoch": 1, "loss": 4.6922, "lr": 8.024691358024692e-07, "grad_norm": 179.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 14, "epoch": 1, "loss": 6.008, "lr": 8.641975308641976e-07, "grad_norm": 176.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 15, "epoch": 1, "loss": 6.1629, "lr": 9.259259259259259e-07, "grad_norm": 216.0, "samples_per_sec": 10.37, "gpu_mem_gb": 6.83} {"step": 16, "epoch": 1, "loss": 5.8775, "lr": 9.876543209876544e-07, "grad_norm": 187.0, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 17, "epoch": 1, "loss": 5.9999, "lr": 1.0493827160493827e-06, "grad_norm": 192.0, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 18, "epoch": 1, "loss": 6.8008, "lr": 1.111111111111111e-06, "grad_norm": 204.0, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 19, "epoch": 1, "loss": 4.8626, "lr": 1.1728395061728396e-06, "grad_norm": 189.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 20, "epoch": 1, "loss": 5.0619, "lr": 1.234567901234568e-06, "grad_norm": 198.0, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 21, "epoch": 1, "loss": 7.0325, "lr": 1.2962962962962962e-06, "grad_norm": 202.0, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 22, "epoch": 1, "loss": 5.9034, "lr": 1.3580246913580248e-06, "grad_norm": 184.0, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 23, "epoch": 1, "loss": 6.6959, "lr": 1.4197530864197531e-06, "grad_norm": 199.0, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 24, "epoch": 1, "loss": 6.3044, "lr": 1.4814814814814815e-06, "grad_norm": 198.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 25, "epoch": 1, "loss": 4.1428, "lr": 1.54320987654321e-06, "grad_norm": 164.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 26, "epoch": 1, "loss": 6.056, "lr": 1.6049382716049383e-06, "grad_norm": 186.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 27, "epoch": 1, "loss": 4.2957, "lr": 1.6666666666666667e-06, "grad_norm": 145.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 28, "epoch": 1, "loss": 5.8194, "lr": 1.7283950617283952e-06, "grad_norm": 186.0, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 29, "epoch": 1, "loss": 5.4599, "lr": 1.7901234567901235e-06, "grad_norm": 167.0, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 30, "epoch": 1, "loss": 5.9108, "lr": 1.8518518518518519e-06, "grad_norm": 186.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 31, "epoch": 1, "loss": 5.5469, "lr": 1.9135802469135804e-06, "grad_norm": 164.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 32, "epoch": 1, "loss": 4.3609, "lr": 1.9753086419753087e-06, "grad_norm": 132.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 33, "epoch": 1, "loss": 5.4421, "lr": 2.037037037037037e-06, "grad_norm": 167.0, "samples_per_sec": 11.04, "gpu_mem_gb": 6.83} {"step": 34, "epoch": 1, "loss": 5.261, "lr": 2.0987654320987654e-06, "grad_norm": 134.0, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 35, "epoch": 1, "loss": 5.5669, "lr": 2.1604938271604937e-06, "grad_norm": 160.0, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 36, "epoch": 1, "loss": 5.9217, "lr": 2.222222222222222e-06, "grad_norm": 129.0, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 37, "epoch": 1, "loss": 5.3044, "lr": 2.283950617283951e-06, "grad_norm": 128.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 38, "epoch": 1, "loss": 3.9628, "lr": 2.345679012345679e-06, "grad_norm": 133.0, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 39, "epoch": 1, "loss": 4.5273, "lr": 2.4074074074074075e-06, "grad_norm": 128.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 40, "epoch": 1, "loss": 5.0782, "lr": 2.469135802469136e-06, "grad_norm": 111.0, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 41, "epoch": 1, "loss": 3.8459, "lr": 2.5308641975308646e-06, "grad_norm": 108.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 42, "epoch": 1, "loss": 5.4054, "lr": 2.5925925925925925e-06, "grad_norm": 129.0, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 43, "epoch": 1, "loss": 4.7466, "lr": 2.6543209876543212e-06, "grad_norm": 117.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 44, "epoch": 1, "loss": 4.5026, "lr": 2.7160493827160496e-06, "grad_norm": 118.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 45, "epoch": 1, "loss": 4.4938, "lr": 2.7777777777777783e-06, "grad_norm": 102.0, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 46, "epoch": 1, "loss": 4.0197, "lr": 2.8395061728395062e-06, "grad_norm": 100.5, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 47, "epoch": 1, "loss": 4.4414, "lr": 2.901234567901235e-06, "grad_norm": 124.0, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 48, "epoch": 1, "loss": 4.4215, "lr": 2.962962962962963e-06, "grad_norm": 114.5, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 49, "epoch": 1, "loss": 5.1638, "lr": 3.0246913580246917e-06, "grad_norm": 108.0, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 50, "epoch": 1, "loss": 3.806, "lr": 3.08641975308642e-06, "grad_norm": 119.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 51, "epoch": 1, "loss": 3.7608, "lr": 3.1481481481481483e-06, "grad_norm": 115.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 52, "epoch": 1, "loss": 3.8337, "lr": 3.2098765432098767e-06, "grad_norm": 119.5, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 53, "epoch": 1, "loss": 2.8417, "lr": 3.2716049382716054e-06, "grad_norm": 99.5, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 54, "epoch": 1, "loss": 3.2936, "lr": 3.3333333333333333e-06, "grad_norm": 82.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 55, "epoch": 1, "loss": 3.1856, "lr": 3.395061728395062e-06, "grad_norm": 89.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 56, "epoch": 1, "loss": 2.6254, "lr": 3.4567901234567904e-06, "grad_norm": 83.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 57, "epoch": 1, "loss": 2.7522, "lr": 3.5185185185185187e-06, "grad_norm": 81.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 58, "epoch": 1, "loss": 3.0448, "lr": 3.580246913580247e-06, "grad_norm": 74.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 59, "epoch": 1, "loss": 3.0212, "lr": 3.641975308641976e-06, "grad_norm": 79.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 60, "epoch": 1, "loss": 2.3525, "lr": 3.7037037037037037e-06, "grad_norm": 56.25, "samples_per_sec": 8.43, "gpu_mem_gb": 6.85} {"step": 61, "epoch": 1, "loss": 3.2121, "lr": 3.7654320987654325e-06, "grad_norm": 69.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 62, "epoch": 1, "loss": 2.7381, "lr": 3.827160493827161e-06, "grad_norm": 60.25, "samples_per_sec": 10.38, "gpu_mem_gb": 6.82} {"step": 63, "epoch": 1, "loss": 2.34, "lr": 3.88888888888889e-06, "grad_norm": 52.0, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 64, "epoch": 1, "loss": 4.0734, "lr": 3.9506172839506175e-06, "grad_norm": 60.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 65, "epoch": 1, "loss": 2.2929, "lr": 4.012345679012346e-06, "grad_norm": 54.0, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 66, "epoch": 1, "loss": 2.1482, "lr": 4.074074074074074e-06, "grad_norm": 55.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 67, "epoch": 1, "loss": 2.3822, "lr": 4.135802469135803e-06, "grad_norm": 60.5, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 68, "epoch": 1, "loss": 1.9241, "lr": 4.197530864197531e-06, "grad_norm": 53.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 69, "epoch": 1, "loss": 2.8067, "lr": 4.2592592592592596e-06, "grad_norm": 54.0, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 70, "epoch": 1, "loss": 2.3438, "lr": 4.3209876543209875e-06, "grad_norm": 49.5, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 71, "epoch": 1, "loss": 1.7198, "lr": 4.382716049382716e-06, "grad_norm": 51.0, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 72, "epoch": 1, "loss": 1.5009, "lr": 4.444444444444444e-06, "grad_norm": 49.5, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 73, "epoch": 1, "loss": 2.0244, "lr": 4.506172839506173e-06, "grad_norm": 51.75, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 74, "epoch": 1, "loss": 1.2851, "lr": 4.567901234567902e-06, "grad_norm": 46.25, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 75, "epoch": 1, "loss": 2.1067, "lr": 4.62962962962963e-06, "grad_norm": 51.0, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 76, "epoch": 1, "loss": 1.518, "lr": 4.691358024691358e-06, "grad_norm": 50.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 77, "epoch": 1, "loss": 1.1726, "lr": 4.753086419753087e-06, "grad_norm": 43.5, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 78, "epoch": 1, "loss": 1.3599, "lr": 4.814814814814815e-06, "grad_norm": 54.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 79, "epoch": 1, "loss": 2.4295, "lr": 4.876543209876544e-06, "grad_norm": 43.25, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 80, "epoch": 1, "loss": 1.905, "lr": 4.938271604938272e-06, "grad_norm": 55.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 81, "epoch": 1, "loss": 1.3875, "lr": 5e-06, "grad_norm": 35.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 82, "epoch": 1, "loss": 1.0834, "lr": 5.061728395061729e-06, "grad_norm": 40.25, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 83, "epoch": 1, "loss": 0.7515, "lr": 5.123456790123458e-06, "grad_norm": 42.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 84, "epoch": 1, "loss": 0.9699, "lr": 5.185185185185185e-06, "grad_norm": 37.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 85, "epoch": 1, "loss": 0.2791, "lr": 5.246913580246914e-06, "grad_norm": 31.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 86, "epoch": 1, "loss": 0.8155, "lr": 5.3086419753086425e-06, "grad_norm": 32.0, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 87, "epoch": 1, "loss": 0.9481, "lr": 5.370370370370371e-06, "grad_norm": 24.875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 88, "epoch": 1, "loss": 1.0406, "lr": 5.432098765432099e-06, "grad_norm": 28.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 89, "epoch": 1, "loss": 0.7261, "lr": 5.493827160493828e-06, "grad_norm": 26.75, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 90, "epoch": 1, "loss": 0.5921, "lr": 5.555555555555557e-06, "grad_norm": 24.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 91, "epoch": 1, "loss": 0.6393, "lr": 5.617283950617285e-06, "grad_norm": 24.875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 92, "epoch": 1, "loss": 0.4403, "lr": 5.6790123456790125e-06, "grad_norm": 27.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 93, "epoch": 1, "loss": 0.1059, "lr": 5.740740740740741e-06, "grad_norm": 20.25, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 94, "epoch": 1, "loss": 0.6298, "lr": 5.80246913580247e-06, "grad_norm": 22.375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 95, "epoch": 1, "loss": 0.0602, "lr": 5.864197530864199e-06, "grad_norm": 21.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 96, "epoch": 1, "loss": 0.2805, "lr": 5.925925925925926e-06, "grad_norm": 14.5625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 97, "epoch": 1, "loss": 0.844, "lr": 5.9876543209876546e-06, "grad_norm": 21.875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 98, "epoch": 1, "loss": 1.0177, "lr": 6.049382716049383e-06, "grad_norm": 38.75, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 99, "epoch": 1, "loss": 0.9418, "lr": 6.111111111111112e-06, "grad_norm": 51.0, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 100, "epoch": 1, "loss": 0.3336, "lr": 6.17283950617284e-06, "grad_norm": 14.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 101, "epoch": 1, "loss": 0.6212, "lr": 6.234567901234569e-06, "grad_norm": 18.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 102, "epoch": 1, "loss": 0.5324, "lr": 6.296296296296297e-06, "grad_norm": 30.125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 103, "epoch": 1, "loss": 0.4003, "lr": 6.358024691358025e-06, "grad_norm": 15.5, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 104, "epoch": 1, "loss": 0.2249, "lr": 6.419753086419753e-06, "grad_norm": 17.375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 105, "epoch": 1, "loss": 0.1983, "lr": 6.481481481481482e-06, "grad_norm": 17.125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 106, "epoch": 1, "loss": 0.7261, "lr": 6.543209876543211e-06, "grad_norm": 22.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 107, "epoch": 1, "loss": 0.1657, "lr": 6.60493827160494e-06, "grad_norm": 17.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 108, "epoch": 1, "loss": 0.3766, "lr": 6.666666666666667e-06, "grad_norm": 15.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 109, "epoch": 1, "loss": 0.5594, "lr": 6.728395061728395e-06, "grad_norm": 16.25, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 110, "epoch": 1, "loss": 0.4835, "lr": 6.790123456790124e-06, "grad_norm": 24.375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 111, "epoch": 1, "loss": 0.4043, "lr": 6.851851851851853e-06, "grad_norm": 20.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 112, "epoch": 1, "loss": 0.3688, "lr": 6.913580246913581e-06, "grad_norm": 26.75, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 113, "epoch": 1, "loss": 0.4046, "lr": 6.975308641975309e-06, "grad_norm": 19.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 114, "epoch": 1, "loss": 0.5872, "lr": 7.0370370370370375e-06, "grad_norm": 16.375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 115, "epoch": 1, "loss": 0.0679, "lr": 7.098765432098766e-06, "grad_norm": 20.75, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 116, "epoch": 1, "loss": 0.3318, "lr": 7.160493827160494e-06, "grad_norm": 16.625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.84} {"step": 117, "epoch": 1, "loss": 0.2717, "lr": 7.222222222222223e-06, "grad_norm": 18.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 118, "epoch": 1, "loss": 0.1122, "lr": 7.283950617283952e-06, "grad_norm": 27.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 119, "epoch": 1, "loss": 0.1342, "lr": 7.34567901234568e-06, "grad_norm": 26.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 120, "epoch": 1, "loss": 0.1335, "lr": 7.4074074074074075e-06, "grad_norm": 12.875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 121, "epoch": 1, "loss": 0.3257, "lr": 7.469135802469136e-06, "grad_norm": 23.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 122, "epoch": 1, "loss": 0.2379, "lr": 7.530864197530865e-06, "grad_norm": 31.125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 123, "epoch": 1, "loss": 0.1268, "lr": 7.592592592592594e-06, "grad_norm": 36.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 124, "epoch": 1, "loss": 0.0386, "lr": 7.654320987654322e-06, "grad_norm": 15.25, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 125, "epoch": 1, "loss": 0.0083, "lr": 7.71604938271605e-06, "grad_norm": 32.0, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 126, "epoch": 1, "loss": 0.1659, "lr": 7.77777777777778e-06, "grad_norm": 14.1875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 127, "epoch": 1, "loss": 0.0809, "lr": 7.839506172839507e-06, "grad_norm": 10.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 128, "epoch": 1, "loss": 0.0239, "lr": 7.901234567901235e-06, "grad_norm": 34.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 129, "epoch": 1, "loss": 0.1996, "lr": 7.962962962962963e-06, "grad_norm": 17.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 130, "epoch": 1, "loss": 0.5576, "lr": 8.024691358024692e-06, "grad_norm": 16.875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 131, "epoch": 1, "loss": 0.1482, "lr": 8.08641975308642e-06, "grad_norm": 25.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 132, "epoch": 1, "loss": 0.2667, "lr": 8.148148148148148e-06, "grad_norm": 30.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 133, "epoch": 1, "loss": 0.1337, "lr": 8.209876543209876e-06, "grad_norm": 18.75, "samples_per_sec": 10.56, "gpu_mem_gb": 6.85} {"step": 134, "epoch": 1, "loss": 0.2397, "lr": 8.271604938271606e-06, "grad_norm": 17.125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 135, "epoch": 1, "loss": 0.0824, "lr": 8.333333333333334e-06, "grad_norm": 14.1875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 136, "epoch": 1, "loss": 0.2493, "lr": 8.395061728395062e-06, "grad_norm": 22.875, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 137, "epoch": 1, "loss": 0.0716, "lr": 8.456790123456791e-06, "grad_norm": 14.125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 138, "epoch": 1, "loss": 0.1764, "lr": 8.518518518518519e-06, "grad_norm": 12.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 139, "epoch": 1, "loss": 0.1003, "lr": 8.580246913580249e-06, "grad_norm": 10.875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 140, "epoch": 1, "loss": 0.1321, "lr": 8.641975308641975e-06, "grad_norm": 21.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 141, "epoch": 1, "loss": 0.2035, "lr": 8.703703703703705e-06, "grad_norm": 20.5, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 142, "epoch": 1, "loss": 0.3794, "lr": 8.765432098765432e-06, "grad_norm": 18.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 143, "epoch": 1, "loss": 0.0178, "lr": 8.827160493827162e-06, "grad_norm": 6.5625, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 144, "epoch": 1, "loss": 0.1147, "lr": 8.888888888888888e-06, "grad_norm": 24.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 145, "epoch": 1, "loss": 0.0554, "lr": 8.950617283950618e-06, "grad_norm": 13.875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 146, "epoch": 1, "loss": 0.0701, "lr": 9.012345679012346e-06, "grad_norm": 15.5625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 147, "epoch": 1, "loss": 0.0125, "lr": 9.074074074074075e-06, "grad_norm": 19.125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 148, "epoch": 1, "loss": 0.0357, "lr": 9.135802469135803e-06, "grad_norm": 19.5, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 149, "epoch": 1, "loss": 0.2148, "lr": 9.197530864197531e-06, "grad_norm": 17.625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 150, "epoch": 1, "loss": 0.2477, "lr": 9.25925925925926e-06, "grad_norm": 23.25, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 151, "epoch": 1, "loss": 0.3539, "lr": 9.320987654320989e-06, "grad_norm": 24.375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 152, "epoch": 1, "loss": 0.0901, "lr": 9.382716049382717e-06, "grad_norm": 25.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 153, "epoch": 1, "loss": 0.0534, "lr": 9.444444444444445e-06, "grad_norm": 22.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 154, "epoch": 1, "loss": 0.0603, "lr": 9.506172839506174e-06, "grad_norm": 27.5, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 155, "epoch": 1, "loss": 0.0667, "lr": 9.567901234567902e-06, "grad_norm": 14.1875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 156, "epoch": 1, "loss": 0.1105, "lr": 9.62962962962963e-06, "grad_norm": 24.5, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 157, "epoch": 1, "loss": 0.0368, "lr": 9.691358024691358e-06, "grad_norm": 16.125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 158, "epoch": 1, "loss": 0.2653, "lr": 9.753086419753087e-06, "grad_norm": 23.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 159, "epoch": 1, "loss": 0.058, "lr": 9.814814814814815e-06, "grad_norm": 24.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 160, "epoch": 1, "loss": 0.1311, "lr": 9.876543209876543e-06, "grad_norm": 28.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 161, "epoch": 1, "loss": 0.0474, "lr": 9.938271604938273e-06, "grad_norm": 21.5, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 162, "epoch": 1, "loss": 0.0378, "lr": 1e-05, "grad_norm": 25.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 163, "epoch": 1, "loss": 0.247, "lr": 1.0061728395061729e-05, "grad_norm": 30.625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 164, "epoch": 1, "loss": 0.0935, "lr": 1.0123456790123458e-05, "grad_norm": 14.5, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 165, "epoch": 1, "loss": 0.0732, "lr": 1.0185185185185186e-05, "grad_norm": 14.4375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 166, "epoch": 1, "loss": 0.0573, "lr": 1.0246913580246916e-05, "grad_norm": 12.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 167, "epoch": 1, "loss": 0.332, "lr": 1.0308641975308642e-05, "grad_norm": 16.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 168, "epoch": 1, "loss": 0.1026, "lr": 1.037037037037037e-05, "grad_norm": 14.625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 169, "epoch": 1, "loss": 0.2298, "lr": 1.04320987654321e-05, "grad_norm": 19.5, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 170, "epoch": 1, "loss": 0.1998, "lr": 1.0493827160493827e-05, "grad_norm": 11.375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.85} {"step": 171, "epoch": 1, "loss": 0.174, "lr": 1.0555555555555557e-05, "grad_norm": 18.875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 172, "epoch": 1, "loss": 0.0919, "lr": 1.0617283950617285e-05, "grad_norm": 21.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 173, "epoch": 1, "loss": 0.055, "lr": 1.0679012345679015e-05, "grad_norm": 17.0, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 174, "epoch": 1, "loss": 0.0852, "lr": 1.0740740740740742e-05, "grad_norm": 15.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 175, "epoch": 1, "loss": 0.0337, "lr": 1.0802469135802469e-05, "grad_norm": 8.8125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 176, "epoch": 1, "loss": 0.0578, "lr": 1.0864197530864198e-05, "grad_norm": 15.8125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 177, "epoch": 1, "loss": 0.1547, "lr": 1.0925925925925926e-05, "grad_norm": 18.25, "samples_per_sec": 10.52, "gpu_mem_gb": 6.85} {"step": 178, "epoch": 1, "loss": 0.4009, "lr": 1.0987654320987656e-05, "grad_norm": 24.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 179, "epoch": 1, "loss": 0.2195, "lr": 1.1049382716049384e-05, "grad_norm": 13.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 180, "epoch": 1, "loss": 0.0283, "lr": 1.1111111111111113e-05, "grad_norm": 26.75, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 181, "epoch": 1, "loss": 0.0718, "lr": 1.1172839506172841e-05, "grad_norm": 21.0, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 182, "epoch": 1, "loss": 0.1707, "lr": 1.123456790123457e-05, "grad_norm": 13.875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 183, "epoch": 1, "loss": 0.0279, "lr": 1.1296296296296297e-05, "grad_norm": 15.1875, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 184, "epoch": 1, "loss": 0.0154, "lr": 1.1358024691358025e-05, "grad_norm": 20.625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 185, "epoch": 1, "loss": 0.1171, "lr": 1.1419753086419753e-05, "grad_norm": 18.0, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 186, "epoch": 1, "loss": 0.1145, "lr": 1.1481481481481482e-05, "grad_norm": 10.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 187, "epoch": 1, "loss": 0.0387, "lr": 1.154320987654321e-05, "grad_norm": 13.125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 188, "epoch": 1, "loss": 0.0335, "lr": 1.160493827160494e-05, "grad_norm": 10.4375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 189, "epoch": 1, "loss": 0.3021, "lr": 1.1666666666666668e-05, "grad_norm": 19.25, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 190, "epoch": 1, "loss": 0.0428, "lr": 1.1728395061728398e-05, "grad_norm": 19.375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.85} {"step": 191, "epoch": 1, "loss": 0.1024, "lr": 1.1790123456790124e-05, "grad_norm": 10.5625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 192, "epoch": 1, "loss": 0.1098, "lr": 1.1851851851851852e-05, "grad_norm": 10.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 193, "epoch": 1, "loss": 0.1522, "lr": 1.1913580246913581e-05, "grad_norm": 17.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.84} {"step": 194, "epoch": 1, "loss": 0.2008, "lr": 1.1975308641975309e-05, "grad_norm": 13.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 195, "epoch": 1, "loss": 0.1878, "lr": 1.2037037037037039e-05, "grad_norm": 27.625, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 196, "epoch": 1, "loss": 0.1036, "lr": 1.2098765432098767e-05, "grad_norm": 17.875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 197, "epoch": 1, "loss": 0.0268, "lr": 1.2160493827160496e-05, "grad_norm": 9.9375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 198, "epoch": 1, "loss": 0.0107, "lr": 1.2222222222222224e-05, "grad_norm": 18.625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 199, "epoch": 1, "loss": 0.0751, "lr": 1.228395061728395e-05, "grad_norm": 11.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 200, "epoch": 1, "loss": 0.4776, "lr": 1.234567901234568e-05, "grad_norm": 20.125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 201, "epoch": 1, "loss": 0.0458, "lr": 1.2407407407407408e-05, "grad_norm": 17.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 202, "epoch": 1, "loss": 0.0366, "lr": 1.2469135802469137e-05, "grad_norm": 14.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 203, "epoch": 1, "loss": 0.0817, "lr": 1.2530864197530865e-05, "grad_norm": 16.625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 204, "epoch": 1, "loss": 0.0218, "lr": 1.2592592592592593e-05, "grad_norm": 16.375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 205, "epoch": 1, "loss": 0.2731, "lr": 1.2654320987654323e-05, "grad_norm": 17.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 206, "epoch": 1, "loss": 0.1552, "lr": 1.271604938271605e-05, "grad_norm": 17.875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 207, "epoch": 1, "loss": 0.3656, "lr": 1.2777777777777777e-05, "grad_norm": 29.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 208, "epoch": 1, "loss": 0.2358, "lr": 1.2839506172839507e-05, "grad_norm": 24.375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 209, "epoch": 1, "loss": 0.1312, "lr": 1.2901234567901235e-05, "grad_norm": 17.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 210, "epoch": 1, "loss": 0.2683, "lr": 1.2962962962962964e-05, "grad_norm": 21.5, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 211, "epoch": 1, "loss": 0.0571, "lr": 1.3024691358024692e-05, "grad_norm": 27.5, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 212, "epoch": 1, "loss": 0.0323, "lr": 1.3086419753086422e-05, "grad_norm": 13.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 213, "epoch": 1, "loss": 0.0048, "lr": 1.314814814814815e-05, "grad_norm": 14.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 214, "epoch": 1, "loss": 0.9609, "lr": 1.320987654320988e-05, "grad_norm": 18.75, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 215, "epoch": 1, "loss": 0.1041, "lr": 1.3271604938271605e-05, "grad_norm": 16.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 216, "epoch": 1, "loss": 0.0137, "lr": 1.3333333333333333e-05, "grad_norm": 13.5625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 217, "epoch": 1, "loss": 0.0687, "lr": 1.3395061728395063e-05, "grad_norm": 10.25, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 218, "epoch": 1, "loss": 0.0394, "lr": 1.345679012345679e-05, "grad_norm": 26.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 219, "epoch": 1, "loss": 0.1813, "lr": 1.351851851851852e-05, "grad_norm": 19.375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 220, "epoch": 1, "loss": 0.1171, "lr": 1.3580246913580248e-05, "grad_norm": 10.75, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 221, "epoch": 1, "loss": 0.1813, "lr": 1.3641975308641976e-05, "grad_norm": 13.875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 222, "epoch": 1, "loss": 0.0888, "lr": 1.3703703703703706e-05, "grad_norm": 11.0625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 223, "epoch": 1, "loss": 0.2603, "lr": 1.3765432098765432e-05, "grad_norm": 15.25, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 224, "epoch": 1, "loss": 0.18, "lr": 1.3827160493827162e-05, "grad_norm": 12.9375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 225, "epoch": 1, "loss": 0.0951, "lr": 1.388888888888889e-05, "grad_norm": 15.25, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 226, "epoch": 1, "loss": 0.1008, "lr": 1.3950617283950617e-05, "grad_norm": 16.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 227, "epoch": 1, "loss": 0.1569, "lr": 1.4012345679012347e-05, "grad_norm": 14.125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 228, "epoch": 1, "loss": 0.0442, "lr": 1.4074074074074075e-05, "grad_norm": 12.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 229, "epoch": 1, "loss": 0.0667, "lr": 1.4135802469135805e-05, "grad_norm": 21.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 230, "epoch": 1, "loss": 0.144, "lr": 1.4197530864197532e-05, "grad_norm": 10.0, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 231, "epoch": 1, "loss": 0.0269, "lr": 1.4259259259259259e-05, "grad_norm": 14.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 232, "epoch": 1, "loss": 0.0644, "lr": 1.4320987654320988e-05, "grad_norm": 13.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 233, "epoch": 1, "loss": 0.0285, "lr": 1.4382716049382716e-05, "grad_norm": 10.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 234, "epoch": 1, "loss": 0.1137, "lr": 1.4444444444444446e-05, "grad_norm": 11.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 235, "epoch": 1, "loss": 0.0216, "lr": 1.4506172839506174e-05, "grad_norm": 4.84375, "samples_per_sec": 11.06, "gpu_mem_gb": 6.82} {"step": 236, "epoch": 1, "loss": 0.2057, "lr": 1.4567901234567903e-05, "grad_norm": 21.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 237, "epoch": 1, "loss": 0.0095, "lr": 1.4629629629629631e-05, "grad_norm": 14.5625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 238, "epoch": 1, "loss": 0.5582, "lr": 1.469135802469136e-05, "grad_norm": 22.5, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 239, "epoch": 1, "loss": 0.1993, "lr": 1.4753086419753087e-05, "grad_norm": 22.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 240, "epoch": 1, "loss": 0.0434, "lr": 1.4814814814814815e-05, "grad_norm": 14.125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 241, "epoch": 1, "loss": 0.0252, "lr": 1.4876543209876545e-05, "grad_norm": 12.0625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 242, "epoch": 1, "loss": 0.0044, "lr": 1.4938271604938272e-05, "grad_norm": 9.0625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 243, "epoch": 1, "loss": 0.0067, "lr": 1.5000000000000002e-05, "grad_norm": 16.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 244, "epoch": 1, "loss": 0.2488, "lr": 1.506172839506173e-05, "grad_norm": 16.25, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 245, "epoch": 1, "loss": 0.0259, "lr": 1.5123456790123458e-05, "grad_norm": 14.75, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 246, "epoch": 1, "loss": 0.3467, "lr": 1.5185185185185187e-05, "grad_norm": 20.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 247, "epoch": 1, "loss": 0.1065, "lr": 1.5246913580246915e-05, "grad_norm": 9.125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 248, "epoch": 1, "loss": 0.19, "lr": 1.5308641975308643e-05, "grad_norm": 5.3125, "samples_per_sec": 11.04, "gpu_mem_gb": 6.85} {"step": 249, "epoch": 1, "loss": 0.3271, "lr": 1.537037037037037e-05, "grad_norm": 13.5, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 250, "epoch": 1, "loss": 0.2096, "lr": 1.54320987654321e-05, "grad_norm": 14.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 251, "epoch": 1, "loss": 0.0796, "lr": 1.549382716049383e-05, "grad_norm": 11.625, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 252, "epoch": 1, "loss": 0.3298, "lr": 1.555555555555556e-05, "grad_norm": 8.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 253, "epoch": 1, "loss": 0.3595, "lr": 1.5617283950617285e-05, "grad_norm": 21.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 254, "epoch": 1, "loss": 0.1541, "lr": 1.5679012345679014e-05, "grad_norm": 11.75, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 255, "epoch": 1, "loss": 0.0768, "lr": 1.5740740740740744e-05, "grad_norm": 17.5, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 256, "epoch": 1, "loss": 0.0031, "lr": 1.580246913580247e-05, "grad_norm": 17.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 257, "epoch": 1, "loss": 0.0031, "lr": 1.58641975308642e-05, "grad_norm": 21.25, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 258, "epoch": 1, "loss": 0.1681, "lr": 1.5925925925925926e-05, "grad_norm": 13.875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 259, "epoch": 1, "loss": 0.0768, "lr": 1.5987654320987655e-05, "grad_norm": 22.5, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 260, "epoch": 1, "loss": 0.1235, "lr": 1.6049382716049385e-05, "grad_norm": 16.25, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 261, "epoch": 1, "loss": 0.2399, "lr": 1.6111111111111115e-05, "grad_norm": 13.375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 262, "epoch": 1, "loss": 0.0626, "lr": 1.617283950617284e-05, "grad_norm": 22.5, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 263, "epoch": 1, "loss": 0.4823, "lr": 1.623456790123457e-05, "grad_norm": 20.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 264, "epoch": 1, "loss": 0.2962, "lr": 1.6296296296296297e-05, "grad_norm": 18.5, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 265, "epoch": 1, "loss": 0.0558, "lr": 1.6358024691358026e-05, "grad_norm": 10.1875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 266, "epoch": 1, "loss": 0.1168, "lr": 1.6419753086419752e-05, "grad_norm": 14.5625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 267, "epoch": 1, "loss": 0.0393, "lr": 1.6481481481481482e-05, "grad_norm": 7.15625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 268, "epoch": 1, "loss": 0.3242, "lr": 1.654320987654321e-05, "grad_norm": 11.875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 269, "epoch": 1, "loss": 0.0638, "lr": 1.660493827160494e-05, "grad_norm": 10.75, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 270, "epoch": 1, "loss": 0.1554, "lr": 1.6666666666666667e-05, "grad_norm": 17.125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 271, "epoch": 1, "loss": 0.0309, "lr": 1.6728395061728397e-05, "grad_norm": 6.0, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 272, "epoch": 1, "loss": 0.0381, "lr": 1.6790123456790123e-05, "grad_norm": 5.6875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 273, "epoch": 1, "loss": 0.0343, "lr": 1.6851851851851853e-05, "grad_norm": 6.34375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 274, "epoch": 1, "loss": 0.0352, "lr": 1.6913580246913582e-05, "grad_norm": 11.375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 275, "epoch": 1, "loss": 0.0304, "lr": 1.697530864197531e-05, "grad_norm": 11.1875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 276, "epoch": 1, "loss": 0.3255, "lr": 1.7037037037037038e-05, "grad_norm": 6.34375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 277, "epoch": 1, "loss": 0.0744, "lr": 1.7098765432098768e-05, "grad_norm": 12.125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 278, "epoch": 1, "loss": 0.0219, "lr": 1.7160493827160498e-05, "grad_norm": 4.0625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 279, "epoch": 1, "loss": 0.0027, "lr": 1.7222222222222224e-05, "grad_norm": 11.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 280, "epoch": 1, "loss": 0.1814, "lr": 1.728395061728395e-05, "grad_norm": 12.9375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 281, "epoch": 1, "loss": 0.1749, "lr": 1.734567901234568e-05, "grad_norm": 12.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 282, "epoch": 1, "loss": 0.1631, "lr": 1.740740740740741e-05, "grad_norm": 18.125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 283, "epoch": 1, "loss": 0.1885, "lr": 1.746913580246914e-05, "grad_norm": 11.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 284, "epoch": 1, "loss": 0.1144, "lr": 1.7530864197530865e-05, "grad_norm": 9.0, "samples_per_sec": 8.74, "gpu_mem_gb": 6.83} {"step": 285, "epoch": 1, "loss": 0.6741, "lr": 1.7592592592592595e-05, "grad_norm": 24.375, "samples_per_sec": 11.8, "gpu_mem_gb": 6.83} {"step": 286, "epoch": 1, "loss": 0.0287, "lr": 1.7654320987654324e-05, "grad_norm": 7.5, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 287, "epoch": 1, "loss": 0.0038, "lr": 1.771604938271605e-05, "grad_norm": 8.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 288, "epoch": 1, "loss": 0.1492, "lr": 1.7777777777777777e-05, "grad_norm": 19.75, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 289, "epoch": 1, "loss": 0.0471, "lr": 1.7839506172839506e-05, "grad_norm": 11.1875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 290, "epoch": 1, "loss": 0.4994, "lr": 1.7901234567901236e-05, "grad_norm": 36.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 291, "epoch": 1, "loss": 0.2823, "lr": 1.7962962962962965e-05, "grad_norm": 18.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 292, "epoch": 1, "loss": 0.2385, "lr": 1.802469135802469e-05, "grad_norm": 6.78125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 293, "epoch": 1, "loss": 0.0195, "lr": 1.808641975308642e-05, "grad_norm": 28.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 294, "epoch": 1, "loss": 0.0822, "lr": 1.814814814814815e-05, "grad_norm": 10.0625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 295, "epoch": 1, "loss": 0.1444, "lr": 1.820987654320988e-05, "grad_norm": 23.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 296, "epoch": 1, "loss": 0.2406, "lr": 1.8271604938271607e-05, "grad_norm": 15.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 297, "epoch": 1, "loss": 0.0475, "lr": 1.8333333333333333e-05, "grad_norm": 19.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 298, "epoch": 1, "loss": 0.1577, "lr": 1.8395061728395062e-05, "grad_norm": 16.0, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 299, "epoch": 1, "loss": 0.2626, "lr": 1.8456790123456792e-05, "grad_norm": 20.625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 300, "epoch": 1, "loss": 0.145, "lr": 1.851851851851852e-05, "grad_norm": 10.5625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 301, "epoch": 1, "loss": 0.2188, "lr": 1.8580246913580248e-05, "grad_norm": 16.875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 302, "epoch": 1, "loss": 0.089, "lr": 1.8641975308641977e-05, "grad_norm": 12.8125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 303, "epoch": 1, "loss": 0.0992, "lr": 1.8703703703703707e-05, "grad_norm": 15.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 304, "epoch": 1, "loss": 0.013, "lr": 1.8765432098765433e-05, "grad_norm": 6.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 305, "epoch": 1, "loss": 0.0748, "lr": 1.8827160493827163e-05, "grad_norm": 14.5, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 306, "epoch": 1, "loss": 0.3364, "lr": 1.888888888888889e-05, "grad_norm": 19.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 307, "epoch": 1, "loss": 0.2233, "lr": 1.895061728395062e-05, "grad_norm": 6.1875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 308, "epoch": 1, "loss": 0.0203, "lr": 1.901234567901235e-05, "grad_norm": 12.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 309, "epoch": 1, "loss": 0.0229, "lr": 1.9074074074074075e-05, "grad_norm": 6.1875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 310, "epoch": 1, "loss": 0.2943, "lr": 1.9135802469135804e-05, "grad_norm": 8.875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 311, "epoch": 1, "loss": 0.3459, "lr": 1.9197530864197534e-05, "grad_norm": 8.125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 312, "epoch": 1, "loss": 0.3593, "lr": 1.925925925925926e-05, "grad_norm": 9.6875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 313, "epoch": 1, "loss": 0.0871, "lr": 1.932098765432099e-05, "grad_norm": 6.90625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 314, "epoch": 1, "loss": 0.1663, "lr": 1.9382716049382716e-05, "grad_norm": 10.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 315, "epoch": 1, "loss": 0.002, "lr": 1.9444444444444445e-05, "grad_norm": 6.21875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 316, "epoch": 1, "loss": 0.0163, "lr": 1.9506172839506175e-05, "grad_norm": 9.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 317, "epoch": 1, "loss": 0.0256, "lr": 1.9567901234567905e-05, "grad_norm": 12.5, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 318, "epoch": 1, "loss": 0.0604, "lr": 1.962962962962963e-05, "grad_norm": 9.6875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 319, "epoch": 1, "loss": 0.0045, "lr": 1.969135802469136e-05, "grad_norm": 13.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 320, "epoch": 1, "loss": 0.1872, "lr": 1.9753086419753087e-05, "grad_norm": 9.5, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 321, "epoch": 1, "loss": 0.7212, "lr": 1.9814814814814816e-05, "grad_norm": 14.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 322, "epoch": 1, "loss": 0.0674, "lr": 1.9876543209876546e-05, "grad_norm": 7.90625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 323, "epoch": 1, "loss": 0.1223, "lr": 1.9938271604938272e-05, "grad_norm": 9.4375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 324, "epoch": 1, "loss": 0.3022, "lr": 2e-05, "grad_norm": 15.125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 325, "epoch": 1, "loss": 0.0038, "lr": 1.9999999551374053e-05, "grad_norm": 15.9375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 326, "epoch": 1, "loss": 0.025, "lr": 1.999999820549625e-05, "grad_norm": 11.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 327, "epoch": 1, "loss": 0.1332, "lr": 1.9999995962366713e-05, "grad_norm": 18.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 328, "epoch": 1, "loss": 0.0037, "lr": 1.9999992821985637e-05, "grad_norm": 6.875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 329, "epoch": 1, "loss": 0.0272, "lr": 1.999998878435331e-05, "grad_norm": 22.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 330, "epoch": 1, "loss": 0.0086, "lr": 1.9999983849470097e-05, "grad_norm": 7.9375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 331, "epoch": 1, "loss": 0.3715, "lr": 1.9999978017336438e-05, "grad_norm": 19.0, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 332, "epoch": 1, "loss": 0.0102, "lr": 1.999997128795285e-05, "grad_norm": 5.75, "samples_per_sec": 10.39, "gpu_mem_gb": 6.83} {"step": 333, "epoch": 1, "loss": 0.1721, "lr": 1.9999963661319944e-05, "grad_norm": 11.1875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 334, "epoch": 1, "loss": 0.388, "lr": 1.9999955137438406e-05, "grad_norm": 43.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 335, "epoch": 1, "loss": 0.004, "lr": 1.999994571630899e-05, "grad_norm": 12.0625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 336, "epoch": 1, "loss": 0.0047, "lr": 1.9999935397932555e-05, "grad_norm": 10.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 337, "epoch": 1, "loss": 0.0141, "lr": 1.999992418231002e-05, "grad_norm": 13.8125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 338, "epoch": 1, "loss": 0.0783, "lr": 1.9999912069442388e-05, "grad_norm": 11.5, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 339, "epoch": 1, "loss": 0.0527, "lr": 1.9999899059330753e-05, "grad_norm": 13.25, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 340, "epoch": 1, "loss": 0.576, "lr": 1.9999885151976275e-05, "grad_norm": 24.625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.84} {"step": 341, "epoch": 1, "loss": 0.0265, "lr": 1.999987034738021e-05, "grad_norm": 23.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 342, "epoch": 1, "loss": 0.2171, "lr": 1.9999854645543876e-05, "grad_norm": 7.53125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 343, "epoch": 1, "loss": 0.087, "lr": 1.9999838046468693e-05, "grad_norm": 5.40625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 344, "epoch": 1, "loss": 0.3782, "lr": 1.9999820550156145e-05, "grad_norm": 6.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 345, "epoch": 1, "loss": 0.0502, "lr": 1.99998021566078e-05, "grad_norm": 9.9375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 346, "epoch": 1, "loss": 0.107, "lr": 1.9999782865825314e-05, "grad_norm": 9.125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 347, "epoch": 1, "loss": 0.2806, "lr": 1.999976267781041e-05, "grad_norm": 9.3125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 348, "epoch": 1, "loss": 0.0932, "lr": 1.9999741592564903e-05, "grad_norm": 7.125, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 349, "epoch": 1, "loss": 0.0826, "lr": 1.999971961009069e-05, "grad_norm": 6.59375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 350, "epoch": 1, "loss": 0.095, "lr": 1.9999696730389734e-05, "grad_norm": 11.3125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 351, "epoch": 1, "loss": 0.0225, "lr": 1.9999672953464095e-05, "grad_norm": 10.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 352, "epoch": 1, "loss": 0.0353, "lr": 1.9999648279315903e-05, "grad_norm": 3.9375, "samples_per_sec": 10.36, "gpu_mem_gb": 6.82} {"step": 353, "epoch": 1, "loss": 0.2024, "lr": 1.9999622707947377e-05, "grad_norm": 16.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 354, "epoch": 1, "loss": 0.3427, "lr": 1.9999596239360804e-05, "grad_norm": 7.40625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 355, "epoch": 1, "loss": 0.2929, "lr": 1.9999568873558567e-05, "grad_norm": 20.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 356, "epoch": 1, "loss": 0.145, "lr": 1.999954061054311e-05, "grad_norm": 17.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 357, "epoch": 1, "loss": 0.0676, "lr": 1.999951145031698e-05, "grad_norm": 6.6875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 358, "epoch": 1, "loss": 0.0012, "lr": 1.9999481392882787e-05, "grad_norm": 4.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 359, "epoch": 1, "loss": 0.2748, "lr": 1.9999450438243233e-05, "grad_norm": 12.25, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 360, "epoch": 1, "loss": 0.1511, "lr": 1.9999418586401092e-05, "grad_norm": 14.375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 361, "epoch": 1, "loss": 0.4929, "lr": 1.9999385837359217e-05, "grad_norm": 14.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 362, "epoch": 1, "loss": 0.1729, "lr": 1.9999352191120556e-05, "grad_norm": 12.875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 363, "epoch": 1, "loss": 0.1364, "lr": 1.9999317647688127e-05, "grad_norm": 9.375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 364, "epoch": 1, "loss": 0.5765, "lr": 1.999928220706502e-05, "grad_norm": 16.5, "samples_per_sec": 10.97, "gpu_mem_gb": 6.84} {"step": 365, "epoch": 1, "loss": 0.2165, "lr": 1.9999245869254423e-05, "grad_norm": 11.6875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 366, "epoch": 1, "loss": 0.0075, "lr": 1.9999208634259594e-05, "grad_norm": 11.0625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 367, "epoch": 1, "loss": 0.2387, "lr": 1.9999170502083876e-05, "grad_norm": 11.8125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 368, "epoch": 1, "loss": 0.0085, "lr": 1.9999131472730688e-05, "grad_norm": 10.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 369, "epoch": 1, "loss": 0.092, "lr": 1.9999091546203533e-05, "grad_norm": 8.8125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 370, "epoch": 1, "loss": 0.086, "lr": 1.999905072250599e-05, "grad_norm": 15.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 371, "epoch": 1, "loss": 0.01, "lr": 1.999900900164173e-05, "grad_norm": 9.0625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 372, "epoch": 1, "loss": 0.1498, "lr": 1.999896638361449e-05, "grad_norm": 10.0, "samples_per_sec": 10.54, "gpu_mem_gb": 6.85} {"step": 373, "epoch": 1, "loss": 0.4487, "lr": 1.999892286842809e-05, "grad_norm": 16.875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 374, "epoch": 1, "loss": 0.3275, "lr": 1.9998878456086448e-05, "grad_norm": 11.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 375, "epoch": 1, "loss": 0.3046, "lr": 1.9998833146593535e-05, "grad_norm": 13.3125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 376, "epoch": 1, "loss": 0.3783, "lr": 1.9998786939953425e-05, "grad_norm": 11.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 377, "epoch": 1, "loss": 0.1063, "lr": 1.9998739836170257e-05, "grad_norm": 9.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 378, "epoch": 1, "loss": 0.0192, "lr": 1.9998691835248263e-05, "grad_norm": 6.28125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 379, "epoch": 1, "loss": 0.1363, "lr": 1.999864293719175e-05, "grad_norm": 12.6875, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 380, "epoch": 1, "loss": 0.0618, "lr": 1.9998593142005104e-05, "grad_norm": 9.0, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 381, "epoch": 1, "loss": 0.0159, "lr": 1.9998542449692794e-05, "grad_norm": 14.6875, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 382, "epoch": 1, "loss": 0.0273, "lr": 1.999849086025936e-05, "grad_norm": 7.3125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 383, "epoch": 1, "loss": 0.128, "lr": 1.9998438373709444e-05, "grad_norm": 7.4375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 384, "epoch": 1, "loss": 0.0382, "lr": 1.9998384990047746e-05, "grad_norm": 14.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 385, "epoch": 1, "loss": 0.0853, "lr": 1.999833070927906e-05, "grad_norm": 10.1875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 386, "epoch": 1, "loss": 0.1474, "lr": 1.9998275531408257e-05, "grad_norm": 5.09375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 387, "epoch": 1, "loss": 0.0356, "lr": 1.9998219456440285e-05, "grad_norm": 16.125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 388, "epoch": 1, "loss": 0.0168, "lr": 1.9998162484380176e-05, "grad_norm": 13.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 389, "epoch": 1, "loss": 0.1489, "lr": 1.9998104615233042e-05, "grad_norm": 9.3125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 390, "epoch": 1, "loss": 0.0886, "lr": 1.9998045849004075e-05, "grad_norm": 15.375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 391, "epoch": 1, "loss": 0.0429, "lr": 1.999798618569855e-05, "grad_norm": 17.0, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 392, "epoch": 1, "loss": 0.2997, "lr": 1.9997925625321818e-05, "grad_norm": 18.75, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 393, "epoch": 1, "loss": 0.0131, "lr": 1.9997864167879313e-05, "grad_norm": 6.96875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 394, "epoch": 1, "loss": 0.1242, "lr": 1.999780181337655e-05, "grad_norm": 11.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 395, "epoch": 1, "loss": 0.0069, "lr": 1.9997738561819123e-05, "grad_norm": 10.125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 396, "epoch": 1, "loss": 0.0099, "lr": 1.999767441321271e-05, "grad_norm": 10.625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 397, "epoch": 1, "loss": 0.0054, "lr": 1.9997609367563064e-05, "grad_norm": 8.1875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 398, "epoch": 1, "loss": 0.2672, "lr": 1.999754342487602e-05, "grad_norm": 12.5, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 399, "epoch": 1, "loss": 0.237, "lr": 1.99974765851575e-05, "grad_norm": 9.625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 400, "epoch": 1, "loss": 0.0815, "lr": 1.9997408848413494e-05, "grad_norm": 8.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 401, "epoch": 1, "loss": 0.0428, "lr": 1.9997340214650084e-05, "grad_norm": 12.9375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 402, "epoch": 1, "loss": 0.0939, "lr": 1.999727068387343e-05, "grad_norm": 7.9375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 403, "epoch": 1, "loss": 0.046, "lr": 1.9997200256089765e-05, "grad_norm": 2.421875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 404, "epoch": 1, "loss": 0.0101, "lr": 1.9997128931305413e-05, "grad_norm": 14.5, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 405, "epoch": 1, "loss": 0.0419, "lr": 1.9997056709526776e-05, "grad_norm": 8.4375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 406, "epoch": 1, "loss": 0.2118, "lr": 1.9996983590760326e-05, "grad_norm": 7.3125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.84} {"step": 407, "epoch": 1, "loss": 0.2111, "lr": 1.9996909575012625e-05, "grad_norm": 4.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 408, "epoch": 1, "loss": 0.0033, "lr": 1.9996834662290323e-05, "grad_norm": 5.21875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 409, "epoch": 1, "loss": 0.2671, "lr": 1.9996758852600132e-05, "grad_norm": 7.09375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 410, "epoch": 1, "loss": 0.097, "lr": 1.9996682145948858e-05, "grad_norm": 9.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 411, "epoch": 1, "loss": 0.0007, "lr": 1.9996604542343384e-05, "grad_norm": 7.21875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 412, "epoch": 1, "loss": 0.225, "lr": 1.9996526041790673e-05, "grad_norm": 9.0625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 413, "epoch": 1, "loss": 0.1576, "lr": 1.9996446644297762e-05, "grad_norm": 14.9375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 414, "epoch": 1, "loss": 0.0114, "lr": 1.9996366349871784e-05, "grad_norm": 4.9375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 415, "epoch": 1, "loss": 0.0061, "lr": 1.9996285158519942e-05, "grad_norm": 7.53125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 416, "epoch": 1, "loss": 0.0051, "lr": 1.9996203070249516e-05, "grad_norm": 6.34375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 417, "epoch": 1, "loss": 0.0087, "lr": 1.9996120085067873e-05, "grad_norm": 3.203125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 418, "epoch": 1, "loss": 0.0181, "lr": 1.9996036202982462e-05, "grad_norm": 8.8125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 419, "epoch": 1, "loss": 0.0079, "lr": 1.999595142400081e-05, "grad_norm": 5.71875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 420, "epoch": 1, "loss": 0.1477, "lr": 1.9995865748130518e-05, "grad_norm": 9.9375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 421, "epoch": 1, "loss": 0.002, "lr": 1.9995779175379275e-05, "grad_norm": 8.8125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 422, "epoch": 1, "loss": 0.0709, "lr": 1.9995691705754853e-05, "grad_norm": 5.3125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 423, "epoch": 1, "loss": 0.2299, "lr": 1.9995603339265095e-05, "grad_norm": 6.3125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 424, "epoch": 1, "loss": 0.7518, "lr": 1.9995514075917934e-05, "grad_norm": 17.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 425, "epoch": 1, "loss": 0.0963, "lr": 1.9995423915721374e-05, "grad_norm": 13.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 426, "epoch": 1, "loss": 0.0048, "lr": 1.9995332858683512e-05, "grad_norm": 8.5625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 427, "epoch": 1, "loss": 0.145, "lr": 1.999524090481251e-05, "grad_norm": 10.1875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 428, "epoch": 1, "loss": 0.134, "lr": 1.999514805411663e-05, "grad_norm": 14.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 429, "epoch": 1, "loss": 0.2812, "lr": 1.9995054306604187e-05, "grad_norm": 11.875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 430, "epoch": 1, "loss": 0.1272, "lr": 1.9994959662283605e-05, "grad_norm": 6.34375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 431, "epoch": 1, "loss": 0.0666, "lr": 1.9994864121163373e-05, "grad_norm": 11.25, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 432, "epoch": 1, "loss": 0.0877, "lr": 1.999476768325206e-05, "grad_norm": 8.875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 433, "epoch": 1, "loss": 0.2108, "lr": 1.9994670348558325e-05, "grad_norm": 15.375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 434, "epoch": 1, "loss": 0.1167, "lr": 1.9994572117090897e-05, "grad_norm": 11.25, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 435, "epoch": 1, "loss": 0.2043, "lr": 1.999447298885859e-05, "grad_norm": 8.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 436, "epoch": 1, "loss": 0.0417, "lr": 1.9994372963870297e-05, "grad_norm": 10.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 437, "epoch": 1, "loss": 0.1231, "lr": 1.9994272042134996e-05, "grad_norm": 10.6875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 438, "epoch": 1, "loss": 0.2162, "lr": 1.999417022366174e-05, "grad_norm": 9.0, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 439, "epoch": 1, "loss": 0.3988, "lr": 1.999406750845967e-05, "grad_norm": 12.6875, "samples_per_sec": 11.01, "gpu_mem_gb": 6.84} {"step": 440, "epoch": 1, "loss": 0.0113, "lr": 1.9993963896537995e-05, "grad_norm": 7.3125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 441, "epoch": 1, "loss": 0.3845, "lr": 1.9993859387906016e-05, "grad_norm": 7.9375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 442, "epoch": 1, "loss": 0.0065, "lr": 1.9993753982573106e-05, "grad_norm": 7.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 443, "epoch": 1, "loss": 0.1054, "lr": 1.9993647680548726e-05, "grad_norm": 9.375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 444, "epoch": 1, "loss": 0.1569, "lr": 1.999354048184241e-05, "grad_norm": 10.3125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 445, "epoch": 1, "loss": 0.0162, "lr": 1.999343238646378e-05, "grad_norm": 9.75, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 446, "epoch": 1, "loss": 0.1764, "lr": 1.999332339442254e-05, "grad_norm": 10.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 447, "epoch": 1, "loss": 0.0084, "lr": 1.999321350572846e-05, "grad_norm": 5.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 448, "epoch": 1, "loss": 0.0178, "lr": 1.9993102720391404e-05, "grad_norm": 5.53125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 449, "epoch": 1, "loss": 0.0006, "lr": 1.999299103842131e-05, "grad_norm": 13.1875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 450, "epoch": 1, "loss": 0.0107, "lr": 1.9992878459828203e-05, "grad_norm": 5.5625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 451, "epoch": 1, "loss": 0.0021, "lr": 1.9992764984622178e-05, "grad_norm": 5.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 452, "epoch": 1, "loss": 0.306, "lr": 1.9992650612813422e-05, "grad_norm": 7.1875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 453, "epoch": 1, "loss": 0.0689, "lr": 1.9992535344412193e-05, "grad_norm": 3.34375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 454, "epoch": 1, "loss": 0.0814, "lr": 1.9992419179428836e-05, "grad_norm": 10.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 455, "epoch": 1, "loss": 0.2352, "lr": 1.999230211787378e-05, "grad_norm": 7.75, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 456, "epoch": 1, "loss": 0.0317, "lr": 1.9992184159757513e-05, "grad_norm": 6.03125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 457, "epoch": 1, "loss": 0.0094, "lr": 1.999206530509063e-05, "grad_norm": 3.8125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 458, "epoch": 1, "loss": 0.2119, "lr": 1.9991945553883796e-05, "grad_norm": 5.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 459, "epoch": 1, "loss": 0.0037, "lr": 1.999182490614775e-05, "grad_norm": 6.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 460, "epoch": 1, "loss": 0.0352, "lr": 1.999170336189332e-05, "grad_norm": 3.484375, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 461, "epoch": 1, "loss": 0.0773, "lr": 1.9991580921131412e-05, "grad_norm": 8.6875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 462, "epoch": 1, "loss": 0.3255, "lr": 1.999145758387301e-05, "grad_norm": 6.5, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 463, "epoch": 1, "loss": 0.0123, "lr": 1.9991333350129186e-05, "grad_norm": 6.53125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 464, "epoch": 1, "loss": 0.0117, "lr": 1.999120821991108e-05, "grad_norm": 6.09375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 465, "epoch": 1, "loss": 0.0064, "lr": 1.999108219322992e-05, "grad_norm": 5.15625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 466, "epoch": 1, "loss": 0.0245, "lr": 1.999095527009702e-05, "grad_norm": 1.890625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 467, "epoch": 1, "loss": 0.0029, "lr": 1.999082745052376e-05, "grad_norm": 0.95703125, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 468, "epoch": 1, "loss": 0.2729, "lr": 1.9990698734521614e-05, "grad_norm": 6.96875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 469, "epoch": 1, "loss": 0.0247, "lr": 1.9990569122102127e-05, "grad_norm": 8.75, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 470, "epoch": 1, "loss": 0.1112, "lr": 1.9990438613276934e-05, "grad_norm": 7.5, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 471, "epoch": 1, "loss": 0.0019, "lr": 1.9990307208057743e-05, "grad_norm": 2.640625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 472, "epoch": 1, "loss": 0.0055, "lr": 1.9990174906456343e-05, "grad_norm": 0.87109375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 473, "epoch": 1, "loss": 0.0084, "lr": 1.9990041708484604e-05, "grad_norm": 8.375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 474, "epoch": 1, "loss": 0.1905, "lr": 1.9989907614154478e-05, "grad_norm": 10.25, "samples_per_sec": 10.5, "gpu_mem_gb": 6.84} {"step": 475, "epoch": 1, "loss": 0.0921, "lr": 1.9989772623478e-05, "grad_norm": 10.875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.85} {"step": 476, "epoch": 1, "loss": 0.3566, "lr": 1.9989636736467278e-05, "grad_norm": 10.8125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 477, "epoch": 1, "loss": 0.4564, "lr": 1.9989499953134506e-05, "grad_norm": 10.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 478, "epoch": 1, "loss": 0.0032, "lr": 1.9989362273491954e-05, "grad_norm": 9.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 479, "epoch": 1, "loss": 0.0052, "lr": 1.998922369755198e-05, "grad_norm": 5.84375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 480, "epoch": 1, "loss": 0.001, "lr": 1.9989084225327014e-05, "grad_norm": 8.5, "samples_per_sec": 10.34, "gpu_mem_gb": 6.82} {"step": 481, "epoch": 1, "loss": 0.0059, "lr": 1.9988943856829573e-05, "grad_norm": 3.21875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 482, "epoch": 1, "loss": 0.1901, "lr": 1.9988802592072252e-05, "grad_norm": 7.625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 483, "epoch": 1, "loss": 0.0032, "lr": 1.9988660431067723e-05, "grad_norm": 8.9375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 484, "epoch": 1, "loss": 0.2768, "lr": 1.998851737382874e-05, "grad_norm": 10.625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 485, "epoch": 1, "loss": 0.0288, "lr": 1.9988373420368146e-05, "grad_norm": 6.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 486, "epoch": 1, "loss": 0.2084, "lr": 1.9988228570698853e-05, "grad_norm": 14.375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 487, "epoch": 1, "loss": 0.1663, "lr": 1.9988082824833857e-05, "grad_norm": 9.3125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 488, "epoch": 1, "loss": 0.3257, "lr": 1.9987936182786234e-05, "grad_norm": 9.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 489, "epoch": 1, "loss": 0.0753, "lr": 1.9987788644569144e-05, "grad_norm": 6.53125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 490, "epoch": 1, "loss": 0.1689, "lr": 1.9987640210195825e-05, "grad_norm": 8.875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 491, "epoch": 1, "loss": 0.0043, "lr": 1.9987490879679593e-05, "grad_norm": 3.390625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 492, "epoch": 1, "loss": 0.0082, "lr": 1.9987340653033847e-05, "grad_norm": 4.8125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 493, "epoch": 1, "loss": 0.0018, "lr": 1.9987189530272072e-05, "grad_norm": 8.5, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 494, "epoch": 1, "loss": 0.0172, "lr": 1.9987037511407815e-05, "grad_norm": 10.1875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 495, "epoch": 1, "loss": 0.014, "lr": 1.998688459645473e-05, "grad_norm": 6.34375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 496, "epoch": 1, "loss": 0.1084, "lr": 1.998673078542653e-05, "grad_norm": 12.375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 497, "epoch": 1, "loss": 0.3395, "lr": 1.9986576078337016e-05, "grad_norm": 13.0625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 498, "epoch": 1, "loss": 0.0105, "lr": 1.998642047520007e-05, "grad_norm": 5.125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 499, "epoch": 1, "loss": 0.0026, "lr": 1.9986263976029652e-05, "grad_norm": 2.921875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 500, "epoch": 1, "loss": 0.0002, "lr": 1.9986106580839807e-05, "grad_norm": 6.1875, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 501, "epoch": 1, "loss": 0.1108, "lr": 1.9985948289644654e-05, "grad_norm": 5.09375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 502, "epoch": 1, "loss": 0.0028, "lr": 1.99857891024584e-05, "grad_norm": 7.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 503, "epoch": 1, "loss": 0.0498, "lr": 1.998562901929532e-05, "grad_norm": 11.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 504, "epoch": 1, "loss": 0.1646, "lr": 1.9985468040169785e-05, "grad_norm": 10.75, "samples_per_sec": 10.73, "gpu_mem_gb": 6.84} {"step": 505, "epoch": 1, "loss": 0.0263, "lr": 1.9985306165096242e-05, "grad_norm": 7.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 506, "epoch": 1, "loss": 0.0197, "lr": 1.9985143394089204e-05, "grad_norm": 4.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 507, "epoch": 1, "loss": 0.0012, "lr": 1.9984979727163286e-05, "grad_norm": 7.84375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.85} {"step": 508, "epoch": 1, "loss": 0.0006, "lr": 1.9984815164333163e-05, "grad_norm": 1.65625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 509, "epoch": 1, "loss": 0.0112, "lr": 1.998464970561361e-05, "grad_norm": 3.59375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 510, "epoch": 1, "loss": 0.001, "lr": 1.998448335101947e-05, "grad_norm": 3.890625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 511, "epoch": 1, "loss": 0.0614, "lr": 1.9984316100565665e-05, "grad_norm": 4.46875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 512, "epoch": 1, "loss": 0.0066, "lr": 1.9984147954267206e-05, "grad_norm": 6.71875, "samples_per_sec": 9.02, "gpu_mem_gb": 6.83} {"step": 513, "epoch": 1, "loss": 0.2939, "lr": 1.998397891213918e-05, "grad_norm": 10.0625, "samples_per_sec": 11.89, "gpu_mem_gb": 6.83} {"step": 514, "epoch": 1, "loss": 0.2698, "lr": 1.9983808974196752e-05, "grad_norm": 6.71875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 515, "epoch": 1, "loss": 0.1887, "lr": 1.998363814045517e-05, "grad_norm": 6.21875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 516, "epoch": 1, "loss": 0.0019, "lr": 1.9983466410929764e-05, "grad_norm": 3.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 517, "epoch": 1, "loss": 0.1278, "lr": 1.998329378563594e-05, "grad_norm": 7.28125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 518, "epoch": 1, "loss": 0.0761, "lr": 1.998312026458919e-05, "grad_norm": 7.65625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 519, "epoch": 1, "loss": 0.1957, "lr": 1.998294584780508e-05, "grad_norm": 6.78125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 520, "epoch": 1, "loss": 0.0023, "lr": 1.998277053529926e-05, "grad_norm": 8.3125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 521, "epoch": 1, "loss": 0.0845, "lr": 1.9982594327087462e-05, "grad_norm": 12.5625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 522, "epoch": 1, "loss": 0.4009, "lr": 1.9982417223185497e-05, "grad_norm": 13.4375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 523, "epoch": 1, "loss": 0.0849, "lr": 1.998223922360925e-05, "grad_norm": 6.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 524, "epoch": 1, "loss": 0.0936, "lr": 1.99820603283747e-05, "grad_norm": 9.5625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 525, "epoch": 1, "loss": 0.0525, "lr": 1.9981880537497894e-05, "grad_norm": 6.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 526, "epoch": 1, "loss": 0.0267, "lr": 1.9981699850994964e-05, "grad_norm": 8.5, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 527, "epoch": 1, "loss": 0.1042, "lr": 1.998151826888212e-05, "grad_norm": 5.84375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 528, "epoch": 1, "loss": 0.0007, "lr": 1.9981335791175655e-05, "grad_norm": 7.0, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 529, "epoch": 1, "loss": 0.0021, "lr": 1.9981152417891948e-05, "grad_norm": 7.75, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 530, "epoch": 1, "loss": 0.4298, "lr": 1.9980968149047448e-05, "grad_norm": 7.0, "samples_per_sec": 10.23, "gpu_mem_gb": 6.83} {"step": 531, "epoch": 1, "loss": 0.1598, "lr": 1.9980782984658682e-05, "grad_norm": 11.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 532, "epoch": 1, "loss": 0.0008, "lr": 1.9980596924742276e-05, "grad_norm": 12.5625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 533, "epoch": 1, "loss": 0.1013, "lr": 1.9980409969314917e-05, "grad_norm": 4.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 534, "epoch": 1, "loss": 0.0416, "lr": 1.998022211839338e-05, "grad_norm": 8.625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 535, "epoch": 1, "loss": 0.0052, "lr": 1.9980033371994524e-05, "grad_norm": 6.34375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 536, "epoch": 1, "loss": 0.0047, "lr": 1.9979843730135278e-05, "grad_norm": 4.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 537, "epoch": 1, "loss": 0.0683, "lr": 1.997965319283266e-05, "grad_norm": 9.125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 538, "epoch": 1, "loss": 0.012, "lr": 1.9979461760103773e-05, "grad_norm": 3.59375, "samples_per_sec": 11.07, "gpu_mem_gb": 6.82} {"step": 539, "epoch": 1, "loss": 0.1853, "lr": 1.9979269431965782e-05, "grad_norm": 6.90625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 540, "epoch": 1, "loss": 0.0013, "lr": 1.997907620843595e-05, "grad_norm": 3.296875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 541, "epoch": 1, "loss": 0.0106, "lr": 1.9978882089531616e-05, "grad_norm": 5.4375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 542, "epoch": 1, "loss": 0.1149, "lr": 1.997868707527019e-05, "grad_norm": 15.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 543, "epoch": 1, "loss": 0.0008, "lr": 1.997849116566918e-05, "grad_norm": 0.57421875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 544, "epoch": 1, "loss": 0.2135, "lr": 1.997829436074615e-05, "grad_norm": 12.8125, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 545, "epoch": 1, "loss": 0.2026, "lr": 1.9978096660518772e-05, "grad_norm": 7.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 546, "epoch": 1, "loss": 0.1457, "lr": 1.997789806500478e-05, "grad_norm": 7.4375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 547, "epoch": 1, "loss": 0.0039, "lr": 1.997769857422199e-05, "grad_norm": 4.59375, "samples_per_sec": 10.3, "gpu_mem_gb": 6.83} {"step": 548, "epoch": 1, "loss": 0.478, "lr": 1.9977498188188306e-05, "grad_norm": 12.75, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 549, "epoch": 1, "loss": 0.0061, "lr": 1.9977296906921702e-05, "grad_norm": 7.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 550, "epoch": 1, "loss": 0.0116, "lr": 1.997709473044024e-05, "grad_norm": 8.375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 551, "epoch": 1, "loss": 0.061, "lr": 1.9976891658762064e-05, "grad_norm": 8.4375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 552, "epoch": 1, "loss": 0.0428, "lr": 1.9976687691905394e-05, "grad_norm": 5.59375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 553, "epoch": 1, "loss": 0.0064, "lr": 1.9976482829888528e-05, "grad_norm": 2.078125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 554, "epoch": 1, "loss": 0.0445, "lr": 1.9976277072729845e-05, "grad_norm": 2.296875, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 555, "epoch": 1, "loss": 0.0008, "lr": 1.9976070420447814e-05, "grad_norm": 5.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 556, "epoch": 1, "loss": 0.0201, "lr": 1.997586287306097e-05, "grad_norm": 11.0625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 557, "epoch": 1, "loss": 0.0008, "lr": 1.9975654430587942e-05, "grad_norm": 4.40625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 558, "epoch": 1, "loss": 0.0228, "lr": 1.9975445093047425e-05, "grad_norm": 8.8125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 559, "epoch": 1, "loss": 0.0012, "lr": 1.9975234860458207e-05, "grad_norm": 8.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 560, "epoch": 1, "loss": 0.0013, "lr": 1.997502373283915e-05, "grad_norm": 5.375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 561, "epoch": 1, "loss": 0.1182, "lr": 1.99748117102092e-05, "grad_norm": 9.3125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 562, "epoch": 1, "loss": 0.1868, "lr": 1.997459879258737e-05, "grad_norm": 7.0, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 563, "epoch": 1, "loss": 0.0004, "lr": 1.997438497999278e-05, "grad_norm": 14.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 564, "epoch": 1, "loss": 0.0104, "lr": 1.9974170272444604e-05, "grad_norm": 5.0, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 565, "epoch": 1, "loss": 0.0045, "lr": 1.997395466996211e-05, "grad_norm": 3.9375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 566, "epoch": 1, "loss": 0.1351, "lr": 1.9973738172564642e-05, "grad_norm": 8.625, "samples_per_sec": 10.53, "gpu_mem_gb": 6.84} {"step": 567, "epoch": 1, "loss": 0.0023, "lr": 1.997352078027162e-05, "grad_norm": 9.25, "samples_per_sec": 10.6, "gpu_mem_gb": 6.85} {"step": 568, "epoch": 1, "loss": 0.0729, "lr": 1.997330249310256e-05, "grad_norm": 7.90625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 569, "epoch": 1, "loss": 0.0028, "lr": 1.997308331107704e-05, "grad_norm": 1.7109375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 570, "epoch": 1, "loss": 0.1933, "lr": 1.9972863234214732e-05, "grad_norm": 8.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 571, "epoch": 1, "loss": 0.1285, "lr": 1.997264226253538e-05, "grad_norm": 8.25, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 572, "epoch": 1, "loss": 0.0314, "lr": 1.9972420396058806e-05, "grad_norm": 2.109375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 573, "epoch": 1, "loss": 0.4735, "lr": 1.9972197634804922e-05, "grad_norm": 10.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 574, "epoch": 1, "loss": 0.1764, "lr": 1.9971973978793717e-05, "grad_norm": 6.71875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 575, "epoch": 1, "loss": 0.0589, "lr": 1.9971749428045255e-05, "grad_norm": 6.5, "samples_per_sec": 10.79, "gpu_mem_gb": 6.84} {"step": 576, "epoch": 1, "loss": 0.1891, "lr": 1.9971523982579682e-05, "grad_norm": 30.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 577, "epoch": 1, "loss": 0.0014, "lr": 1.9971297642417233e-05, "grad_norm": 2.4375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 578, "epoch": 1, "loss": 0.0921, "lr": 1.997107040757821e-05, "grad_norm": 6.3125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 579, "epoch": 1, "loss": 0.394, "lr": 1.9970842278083004e-05, "grad_norm": 11.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 580, "epoch": 1, "loss": 0.1502, "lr": 1.9970613253952086e-05, "grad_norm": 6.28125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 581, "epoch": 1, "loss": 0.0135, "lr": 1.9970383335206002e-05, "grad_norm": 6.0, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 582, "epoch": 1, "loss": 0.0338, "lr": 1.9970152521865384e-05, "grad_norm": 8.0625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 583, "epoch": 1, "loss": 0.4359, "lr": 1.9969920813950935e-05, "grad_norm": 9.0625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 584, "epoch": 1, "loss": 0.1602, "lr": 1.9969688211483456e-05, "grad_norm": 6.53125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 585, "epoch": 1, "loss": 0.1429, "lr": 1.996945471448381e-05, "grad_norm": 4.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 586, "epoch": 1, "loss": 0.1016, "lr": 1.9969220322972948e-05, "grad_norm": 5.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 587, "epoch": 1, "loss": 0.0845, "lr": 1.9968985036971908e-05, "grad_norm": 10.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 588, "epoch": 1, "loss": 0.1963, "lr": 1.996874885650179e-05, "grad_norm": 10.0, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 589, "epoch": 1, "loss": 0.1135, "lr": 1.9968511781583793e-05, "grad_norm": 10.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 590, "epoch": 1, "loss": 0.1671, "lr": 1.9968273812239185e-05, "grad_norm": 6.9375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 591, "epoch": 1, "loss": 0.0801, "lr": 1.996803494848932e-05, "grad_norm": 4.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 592, "epoch": 1, "loss": 0.0009, "lr": 1.9967795190355627e-05, "grad_norm": 3.296875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 593, "epoch": 1, "loss": 0.0015, "lr": 1.9967554537859622e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 594, "epoch": 1, "loss": 0.0207, "lr": 1.9967312991022897e-05, "grad_norm": 2.09375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 595, "epoch": 1, "loss": 0.1424, "lr": 1.9967070549867123e-05, "grad_norm": 3.96875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 596, "epoch": 1, "loss": 0.6318, "lr": 1.996682721441405e-05, "grad_norm": 11.9375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 597, "epoch": 1, "loss": 0.194, "lr": 1.9966582984685522e-05, "grad_norm": 6.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 598, "epoch": 1, "loss": 0.0467, "lr": 1.9966337860703446e-05, "grad_norm": 3.53125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 599, "epoch": 1, "loss": 0.033, "lr": 1.9966091842489815e-05, "grad_norm": 8.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 600, "epoch": 1, "loss": 0.2707, "lr": 1.99658449300667e-05, "grad_norm": 6.75, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 601, "epoch": 1, "loss": 0.2218, "lr": 1.996559712345626e-05, "grad_norm": 8.5625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 602, "epoch": 1, "loss": 0.0034, "lr": 1.996534842268073e-05, "grad_norm": 7.34375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.85} {"step": 603, "epoch": 1, "loss": 0.0066, "lr": 1.9965098827762423e-05, "grad_norm": 6.03125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 604, "epoch": 1, "loss": 0.0143, "lr": 1.9964848338723736e-05, "grad_norm": 6.6875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 605, "epoch": 1, "loss": 0.0218, "lr": 1.996459695558714e-05, "grad_norm": 6.03125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 606, "epoch": 1, "loss": 0.0168, "lr": 1.9964344678375194e-05, "grad_norm": 3.390625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 607, "epoch": 1, "loss": 0.0029, "lr": 1.996409150711053e-05, "grad_norm": 3.859375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 608, "epoch": 1, "loss": 0.0023, "lr": 1.9963837441815866e-05, "grad_norm": 6.46875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 609, "epoch": 1, "loss": 0.1352, "lr": 1.9963582482514003e-05, "grad_norm": 12.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 610, "epoch": 1, "loss": 0.0069, "lr": 1.9963326629227807e-05, "grad_norm": 4.34375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 611, "epoch": 1, "loss": 0.0128, "lr": 1.9963069881980243e-05, "grad_norm": 4.9375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 612, "epoch": 1, "loss": 0.3572, "lr": 1.9962812240794344e-05, "grad_norm": 7.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 613, "epoch": 1, "loss": 0.3903, "lr": 1.996255370569323e-05, "grad_norm": 12.75, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 614, "epoch": 1, "loss": 0.0524, "lr": 1.9962294276700094e-05, "grad_norm": 8.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 615, "epoch": 1, "loss": 0.007, "lr": 1.9962033953838215e-05, "grad_norm": 4.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 616, "epoch": 1, "loss": 0.0456, "lr": 1.996177273713095e-05, "grad_norm": 3.65625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 617, "epoch": 1, "loss": 0.0011, "lr": 1.9961510626601736e-05, "grad_norm": 7.4375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 618, "epoch": 1, "loss": 0.0109, "lr": 1.9961247622274093e-05, "grad_norm": 6.5625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 619, "epoch": 1, "loss": 0.0032, "lr": 1.9960983724171622e-05, "grad_norm": 8.0625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 620, "epoch": 1, "loss": 0.0928, "lr": 1.9960718932317994e-05, "grad_norm": 8.5, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 621, "epoch": 1, "loss": 0.065, "lr": 1.9960453246736972e-05, "grad_norm": 8.6875, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 622, "epoch": 1, "loss": 0.0062, "lr": 1.9960186667452395e-05, "grad_norm": 4.375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 623, "epoch": 1, "loss": 0.0029, "lr": 1.9959919194488177e-05, "grad_norm": 3.90625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 624, "epoch": 1, "loss": 0.1079, "lr": 1.9959650827868323e-05, "grad_norm": 7.6875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 625, "epoch": 1, "loss": 0.0512, "lr": 1.9959381567616912e-05, "grad_norm": 6.0, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 626, "epoch": 1, "loss": 0.2289, "lr": 1.99591114137581e-05, "grad_norm": 8.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 627, "epoch": 1, "loss": 0.0077, "lr": 1.9958840366316127e-05, "grad_norm": 3.0625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 628, "epoch": 1, "loss": 0.1154, "lr": 1.9958568425315316e-05, "grad_norm": 9.5625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 629, "epoch": 1, "loss": 0.2276, "lr": 1.995829559078006e-05, "grad_norm": 4.875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 630, "epoch": 1, "loss": 0.1502, "lr": 1.995802186273485e-05, "grad_norm": 5.96875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 631, "epoch": 1, "loss": 0.1443, "lr": 1.995774724120424e-05, "grad_norm": 40.5, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 632, "epoch": 1, "loss": 0.0854, "lr": 1.9957471726212866e-05, "grad_norm": 3.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 633, "epoch": 1, "loss": 0.0482, "lr": 1.9957195317785453e-05, "grad_norm": 9.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 634, "epoch": 1, "loss": 0.0014, "lr": 1.9956918015946806e-05, "grad_norm": 5.46875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 635, "epoch": 1, "loss": 0.1169, "lr": 1.99566398207218e-05, "grad_norm": 7.9375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 636, "epoch": 1, "loss": 0.0526, "lr": 1.99563607321354e-05, "grad_norm": 2.984375, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 637, "epoch": 1, "loss": 0.1382, "lr": 1.9956080750212642e-05, "grad_norm": 5.71875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 638, "epoch": 1, "loss": 0.0295, "lr": 1.9955799874978655e-05, "grad_norm": 3.5, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 639, "epoch": 1, "loss": 0.0348, "lr": 1.9955518106458633e-05, "grad_norm": 8.1875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 640, "epoch": 1, "loss": 0.0049, "lr": 1.9955235444677865e-05, "grad_norm": 3.1875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 641, "epoch": 1, "loss": 0.0711, "lr": 1.9954951889661704e-05, "grad_norm": 7.15625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 642, "epoch": 1, "loss": 0.1897, "lr": 1.99546674414356e-05, "grad_norm": 4.25, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 643, "epoch": 1, "loss": 0.0145, "lr": 1.9954382100025074e-05, "grad_norm": 6.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 644, "epoch": 1, "loss": 0.0074, "lr": 1.9954095865455726e-05, "grad_norm": 2.6875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 645, "epoch": 1, "loss": 0.001, "lr": 1.9953808737753236e-05, "grad_norm": 4.125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 646, "epoch": 1, "loss": 0.1315, "lr": 1.9953520716943373e-05, "grad_norm": 7.15625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 647, "epoch": 1, "loss": 0.0932, "lr": 1.9953231803051977e-05, "grad_norm": 5.3125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 648, "epoch": 1, "loss": 0.028, "lr": 1.9952941996104968e-05, "grad_norm": 4.34375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 649, "epoch": 1, "loss": 0.1068, "lr": 1.995265129612835e-05, "grad_norm": 6.15625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 650, "epoch": 1, "loss": 0.177, "lr": 1.995235970314821e-05, "grad_norm": 6.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 651, "epoch": 1, "loss": 0.117, "lr": 1.9952067217190708e-05, "grad_norm": 5.3125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 652, "epoch": 1, "loss": 0.2837, "lr": 1.9951773838282084e-05, "grad_norm": 8.5, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 653, "epoch": 1, "loss": 0.1409, "lr": 1.9951479566448672e-05, "grad_norm": 10.6875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 654, "epoch": 1, "loss": 0.2415, "lr": 1.9951184401716867e-05, "grad_norm": 9.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 655, "epoch": 1, "loss": 0.2814, "lr": 1.995088834411315e-05, "grad_norm": 8.1875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 656, "epoch": 1, "loss": 0.319, "lr": 1.9950591393664095e-05, "grad_norm": 14.25, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 657, "epoch": 1, "loss": 0.0067, "lr": 1.995029355039634e-05, "grad_norm": 7.84375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 658, "epoch": 1, "loss": 0.1084, "lr": 1.9949994814336605e-05, "grad_norm": 4.75, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 659, "epoch": 1, "loss": 0.1037, "lr": 1.99496951855117e-05, "grad_norm": 8.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 660, "epoch": 1, "loss": 0.0031, "lr": 1.994939466394851e-05, "grad_norm": 6.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 661, "epoch": 1, "loss": 0.3669, "lr": 1.9949093249674e-05, "grad_norm": 8.25, "samples_per_sec": 10.56, "gpu_mem_gb": 6.85} {"step": 662, "epoch": 1, "loss": 0.0881, "lr": 1.9948790942715206e-05, "grad_norm": 4.6875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 663, "epoch": 1, "loss": 0.0547, "lr": 1.9948487743099262e-05, "grad_norm": 5.8125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 664, "epoch": 1, "loss": 0.1197, "lr": 1.9948183650853368e-05, "grad_norm": 7.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 665, "epoch": 1, "loss": 0.0144, "lr": 1.994787866600481e-05, "grad_norm": 8.5625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 666, "epoch": 1, "loss": 0.3063, "lr": 1.994757278858095e-05, "grad_norm": 6.1875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 667, "epoch": 1, "loss": 0.2185, "lr": 1.9947266018609238e-05, "grad_norm": 9.5, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 668, "epoch": 1, "loss": 0.139, "lr": 1.9946958356117194e-05, "grad_norm": 7.6875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 669, "epoch": 1, "loss": 0.2357, "lr": 1.994664980113243e-05, "grad_norm": 5.71875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 670, "epoch": 1, "loss": 0.0023, "lr": 1.994634035368262e-05, "grad_norm": 5.3125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 671, "epoch": 1, "loss": 0.0009, "lr": 1.994603001379554e-05, "grad_norm": 2.015625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 672, "epoch": 1, "loss": 0.0018, "lr": 1.9945718781499032e-05, "grad_norm": 7.125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 673, "epoch": 1, "loss": 0.0047, "lr": 1.9945406656821016e-05, "grad_norm": 3.625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 674, "epoch": 1, "loss": 0.4575, "lr": 1.9945093639789505e-05, "grad_norm": 11.125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.84} {"step": 675, "epoch": 1, "loss": 0.0996, "lr": 1.9944779730432586e-05, "grad_norm": 6.71875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 676, "epoch": 1, "loss": 0.0015, "lr": 1.9944464928778413e-05, "grad_norm": 4.59375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 677, "epoch": 1, "loss": 0.0012, "lr": 1.994414923485524e-05, "grad_norm": 6.3125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 678, "epoch": 1, "loss": 0.4826, "lr": 1.9943832648691392e-05, "grad_norm": 11.5, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 679, "epoch": 1, "loss": 0.0516, "lr": 1.9943515170315274e-05, "grad_norm": 6.84375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 680, "epoch": 1, "loss": 0.001, "lr": 1.9943196799755375e-05, "grad_norm": 7.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 681, "epoch": 1, "loss": 0.1161, "lr": 1.9942877537040256e-05, "grad_norm": 5.40625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 682, "epoch": 1, "loss": 0.1766, "lr": 1.9942557382198566e-05, "grad_norm": 3.671875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 683, "epoch": 1, "loss": 0.467, "lr": 1.9942236335259024e-05, "grad_norm": 7.53125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 684, "epoch": 1, "loss": 0.0137, "lr": 1.9941914396250447e-05, "grad_norm": 3.171875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 685, "epoch": 1, "loss": 0.1762, "lr": 1.9941591565201712e-05, "grad_norm": 12.625, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 686, "epoch": 1, "loss": 0.0162, "lr": 1.994126784214179e-05, "grad_norm": 2.96875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 687, "epoch": 1, "loss": 0.047, "lr": 1.9940943227099726e-05, "grad_norm": 4.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 688, "epoch": 1, "loss": 0.1063, "lr": 1.994061772010465e-05, "grad_norm": 5.40625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 689, "epoch": 1, "loss": 0.0028, "lr": 1.994029132118576e-05, "grad_norm": 8.0, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 690, "epoch": 1, "loss": 0.0396, "lr": 1.993996403037235e-05, "grad_norm": 5.53125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 691, "epoch": 1, "loss": 0.1913, "lr": 1.993963584769378e-05, "grad_norm": 6.375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 692, "epoch": 1, "loss": 0.1776, "lr": 1.9939306773179498e-05, "grad_norm": 6.25, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 693, "epoch": 1, "loss": 0.3104, "lr": 1.993897680685903e-05, "grad_norm": 20.375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 694, "epoch": 1, "loss": 0.1925, "lr": 1.9938645948761986e-05, "grad_norm": 5.40625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 695, "epoch": 1, "loss": 0.1361, "lr": 1.993831419891805e-05, "grad_norm": 5.28125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 696, "epoch": 1, "loss": 0.1344, "lr": 1.9937981557356986e-05, "grad_norm": 11.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 697, "epoch": 1, "loss": 0.0042, "lr": 1.9937648024108643e-05, "grad_norm": 5.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 698, "epoch": 1, "loss": 0.1013, "lr": 1.993731359920295e-05, "grad_norm": 8.4375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 699, "epoch": 1, "loss": 0.0063, "lr": 1.9936978282669906e-05, "grad_norm": 6.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 700, "epoch": 1, "loss": 0.1317, "lr": 1.9936642074539604e-05, "grad_norm": 3.796875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 701, "epoch": 1, "loss": 0.1147, "lr": 1.9936304974842208e-05, "grad_norm": 5.59375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.84} {"step": 702, "epoch": 1, "loss": 0.0148, "lr": 1.993596698360796e-05, "grad_norm": 2.140625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 703, "epoch": 1, "loss": 0.0021, "lr": 1.9935628100867193e-05, "grad_norm": 3.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 704, "epoch": 1, "loss": 0.1349, "lr": 1.9935288326650314e-05, "grad_norm": 4.6875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 705, "epoch": 1, "loss": 0.1791, "lr": 1.99349476609878e-05, "grad_norm": 4.09375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 706, "epoch": 1, "loss": 0.0009, "lr": 1.993460610391023e-05, "grad_norm": 5.25, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 707, "epoch": 1, "loss": 0.0847, "lr": 1.993426365544824e-05, "grad_norm": 9.5625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 708, "epoch": 1, "loss": 0.0289, "lr": 1.9933920315632557e-05, "grad_norm": 5.9375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 709, "epoch": 1, "loss": 0.1001, "lr": 1.9933576084493996e-05, "grad_norm": 5.6875, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 710, "epoch": 1, "loss": 0.4213, "lr": 1.9933230962063433e-05, "grad_norm": 5.65625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 711, "epoch": 1, "loss": 0.0036, "lr": 1.993288494837184e-05, "grad_norm": 3.296875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 712, "epoch": 1, "loss": 0.0083, "lr": 1.9932538043450263e-05, "grad_norm": 5.125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.85} {"step": 713, "epoch": 1, "loss": 0.0057, "lr": 1.9932190247329824e-05, "grad_norm": 2.9375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 714, "epoch": 1, "loss": 0.0395, "lr": 1.9931841560041735e-05, "grad_norm": 2.8125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 715, "epoch": 1, "loss": 0.2508, "lr": 1.9931491981617276e-05, "grad_norm": 7.65625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 716, "epoch": 1, "loss": 0.0043, "lr": 1.9931141512087814e-05, "grad_norm": 5.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 717, "epoch": 1, "loss": 0.0017, "lr": 1.9930790151484804e-05, "grad_norm": 3.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 718, "epoch": 1, "loss": 0.1622, "lr": 1.993043789983976e-05, "grad_norm": 6.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 719, "epoch": 1, "loss": 0.0091, "lr": 1.993008475718429e-05, "grad_norm": 1.4765625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 720, "epoch": 1, "loss": 0.0092, "lr": 1.9929730723550084e-05, "grad_norm": 3.9375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 721, "epoch": 1, "loss": 0.0183, "lr": 1.9929375798968907e-05, "grad_norm": 6.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 722, "epoch": 1, "loss": 0.2769, "lr": 1.9929019983472605e-05, "grad_norm": 5.46875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 723, "epoch": 1, "loss": 0.0052, "lr": 1.99286632770931e-05, "grad_norm": 4.84375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 724, "epoch": 1, "loss": 0.0184, "lr": 1.99283056798624e-05, "grad_norm": 5.34375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 725, "epoch": 1, "loss": 0.0016, "lr": 1.992794719181259e-05, "grad_norm": 0.5390625, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 726, "epoch": 1, "loss": 0.0029, "lr": 1.992758781297584e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 727, "epoch": 1, "loss": 0.1024, "lr": 1.9927227543384387e-05, "grad_norm": 7.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 728, "epoch": 1, "loss": 0.0057, "lr": 1.9926866383070563e-05, "grad_norm": 6.15625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 729, "epoch": 1, "loss": 0.1338, "lr": 1.992650433206677e-05, "grad_norm": 3.109375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.85} {"step": 730, "epoch": 1, "loss": 0.0135, "lr": 1.9926141390405493e-05, "grad_norm": 1.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 731, "epoch": 1, "loss": 0.3005, "lr": 1.9925777558119294e-05, "grad_norm": 6.75, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 732, "epoch": 1, "loss": 0.1426, "lr": 1.9925412835240826e-05, "grad_norm": 7.25, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 733, "epoch": 1, "loss": 0.001, "lr": 1.992504722180281e-05, "grad_norm": 3.015625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 734, "epoch": 1, "loss": 0.2175, "lr": 1.9924680717838045e-05, "grad_norm": 6.28125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 735, "epoch": 1, "loss": 0.0006, "lr": 1.9924313323379426e-05, "grad_norm": 6.09375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 736, "epoch": 1, "loss": 0.0276, "lr": 1.9923945038459908e-05, "grad_norm": 7.78125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 737, "epoch": 1, "loss": 0.1777, "lr": 1.9923575863112545e-05, "grad_norm": 6.6875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 738, "epoch": 1, "loss": 0.0026, "lr": 1.992320579737045e-05, "grad_norm": 6.34375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 739, "epoch": 1, "loss": 0.0069, "lr": 1.992283484126684e-05, "grad_norm": 5.15625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 740, "epoch": 1, "loss": 0.0027, "lr": 1.992246299483499e-05, "grad_norm": 3.140625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 741, "epoch": 1, "loss": 0.0012, "lr": 1.9922090258108265e-05, "grad_norm": 3.78125, "samples_per_sec": 8.85, "gpu_mem_gb": 6.82} {"step": 742, "epoch": 1, "loss": 0.0022, "lr": 1.992171663112011e-05, "grad_norm": 5.6875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.85} {"step": 743, "epoch": 1, "loss": 0.0088, "lr": 1.9921342113904048e-05, "grad_norm": 5.59375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 744, "epoch": 1, "loss": 0.0007, "lr": 1.9920966706493686e-05, "grad_norm": 3.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 745, "epoch": 1, "loss": 0.0024, "lr": 1.9920590408922705e-05, "grad_norm": 5.5625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 746, "epoch": 1, "loss": 0.0619, "lr": 1.9920213221224865e-05, "grad_norm": 9.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 747, "epoch": 1, "loss": 0.3509, "lr": 1.9919835143434016e-05, "grad_norm": 5.03125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 748, "epoch": 1, "loss": 0.2791, "lr": 1.9919456175584076e-05, "grad_norm": 7.65625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 749, "epoch": 1, "loss": 0.0927, "lr": 1.9919076317709053e-05, "grad_norm": 6.0, "samples_per_sec": 10.64, "gpu_mem_gb": 6.84} {"step": 750, "epoch": 1, "loss": 0.179, "lr": 1.9918695569843024e-05, "grad_norm": 4.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 751, "epoch": 1, "loss": 0.0143, "lr": 1.9918313932020155e-05, "grad_norm": 9.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 752, "epoch": 1, "loss": 0.0071, "lr": 1.991793140427469e-05, "grad_norm": 6.3125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 753, "epoch": 1, "loss": 0.0065, "lr": 1.9917547986640947e-05, "grad_norm": 3.8125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.85} {"step": 754, "epoch": 1, "loss": 0.0035, "lr": 1.991716367915333e-05, "grad_norm": 5.28125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 755, "epoch": 1, "loss": 0.0686, "lr": 1.9916778481846322e-05, "grad_norm": 4.75, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 756, "epoch": 1, "loss": 0.1568, "lr": 1.9916392394754483e-05, "grad_norm": 4.78125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 757, "epoch": 1, "loss": 0.0027, "lr": 1.9916005417912457e-05, "grad_norm": 4.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 758, "epoch": 1, "loss": 0.0022, "lr": 1.9915617551354965e-05, "grad_norm": 3.828125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 759, "epoch": 1, "loss": 0.074, "lr": 1.991522879511681e-05, "grad_norm": 5.03125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 760, "epoch": 1, "loss": 0.0058, "lr": 1.9914839149232868e-05, "grad_norm": 13.9375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.85} {"step": 761, "epoch": 1, "loss": 0.0212, "lr": 1.9914448613738107e-05, "grad_norm": 5.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 762, "epoch": 1, "loss": 0.0127, "lr": 1.9914057188667558e-05, "grad_norm": 8.25, "samples_per_sec": 10.36, "gpu_mem_gb": 6.82} {"step": 763, "epoch": 1, "loss": 0.1194, "lr": 1.9913664874056355e-05, "grad_norm": 6.03125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 764, "epoch": 1, "loss": 0.0962, "lr": 1.9913271669939688e-05, "grad_norm": 9.375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 765, "epoch": 1, "loss": 0.0666, "lr": 1.9912877576352843e-05, "grad_norm": 4.40625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 766, "epoch": 1, "loss": 0.0066, "lr": 1.9912482593331174e-05, "grad_norm": 5.96875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 767, "epoch": 1, "loss": 0.0032, "lr": 1.9912086720910128e-05, "grad_norm": 2.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 768, "epoch": 1, "loss": 0.0272, "lr": 1.991168995912522e-05, "grad_norm": 3.40625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 769, "epoch": 1, "loss": 0.0092, "lr": 1.9911292308012054e-05, "grad_norm": 5.96875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 770, "epoch": 1, "loss": 0.2888, "lr": 1.9910893767606302e-05, "grad_norm": 10.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 771, "epoch": 1, "loss": 0.0608, "lr": 1.991049433794373e-05, "grad_norm": 2.546875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 772, "epoch": 1, "loss": 0.0055, "lr": 1.9910094019060174e-05, "grad_norm": 3.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 773, "epoch": 1, "loss": 0.0072, "lr": 1.990969281099155e-05, "grad_norm": 4.15625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 774, "epoch": 1, "loss": 0.1342, "lr": 1.9909290713773863e-05, "grad_norm": 5.21875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 775, "epoch": 1, "loss": 0.0291, "lr": 1.990888772744319e-05, "grad_norm": 8.0625, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 776, "epoch": 1, "loss": 0.0416, "lr": 1.9908483852035684e-05, "grad_norm": 5.78125, "samples_per_sec": 10.24, "gpu_mem_gb": 6.83} {"step": 777, "epoch": 1, "loss": 0.0079, "lr": 1.9908079087587583e-05, "grad_norm": 2.265625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 778, "epoch": 1, "loss": 0.0004, "lr": 1.990767343413521e-05, "grad_norm": 3.015625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 779, "epoch": 1, "loss": 0.0026, "lr": 1.990726689171496e-05, "grad_norm": 0.74609375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 780, "epoch": 1, "loss": 0.0028, "lr": 1.9906859460363307e-05, "grad_norm": 2.0625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 781, "epoch": 1, "loss": 0.0286, "lr": 1.9906451140116813e-05, "grad_norm": 8.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 782, "epoch": 1, "loss": 0.0043, "lr": 1.9906041931012115e-05, "grad_norm": 4.25, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 783, "epoch": 1, "loss": 0.0207, "lr": 1.9905631833085922e-05, "grad_norm": 8.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 784, "epoch": 1, "loss": 0.0094, "lr": 1.990522084637503e-05, "grad_norm": 2.796875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 785, "epoch": 1, "loss": 0.0545, "lr": 1.990480897091633e-05, "grad_norm": 6.25, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 786, "epoch": 1, "loss": 0.0249, "lr": 1.990439620674676e-05, "grad_norm": 4.75, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 787, "epoch": 1, "loss": 0.044, "lr": 1.9903982553903364e-05, "grad_norm": 4.65625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 788, "epoch": 1, "loss": 0.1692, "lr": 1.990356801242326e-05, "grad_norm": 3.59375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 789, "epoch": 1, "loss": 0.001, "lr": 1.9903152582343633e-05, "grad_norm": 7.4375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 790, "epoch": 1, "loss": 0.0101, "lr": 1.990273626370176e-05, "grad_norm": 4.4375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 791, "epoch": 1, "loss": 0.0276, "lr": 1.9902319056535003e-05, "grad_norm": 3.875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 792, "epoch": 1, "loss": 0.0409, "lr": 1.9901900960880792e-05, "grad_norm": 2.796875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 793, "epoch": 1, "loss": 0.0964, "lr": 1.9901481976776637e-05, "grad_norm": 6.90625, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 794, "epoch": 1, "loss": 0.0937, "lr": 1.9901062104260135e-05, "grad_norm": 7.5, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 795, "epoch": 1, "loss": 0.2435, "lr": 1.990064134336896e-05, "grad_norm": 5.75, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 796, "epoch": 1, "loss": 0.0326, "lr": 1.990021969414086e-05, "grad_norm": 3.875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 797, "epoch": 1, "loss": 0.0144, "lr": 1.9899797156613672e-05, "grad_norm": 5.4375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 798, "epoch": 1, "loss": 0.0351, "lr": 1.9899373730825306e-05, "grad_norm": 3.140625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 799, "epoch": 1, "loss": 0.0034, "lr": 1.9898949416813757e-05, "grad_norm": 5.28125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 800, "epoch": 1, "loss": 0.3615, "lr": 1.9898524214617093e-05, "grad_norm": 6.40625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 801, "epoch": 1, "loss": 0.0243, "lr": 1.9898098124273465e-05, "grad_norm": 5.65625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 802, "epoch": 1, "loss": 0.007, "lr": 1.989767114582111e-05, "grad_norm": 6.53125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 803, "epoch": 1, "loss": 0.1053, "lr": 1.9897243279298334e-05, "grad_norm": 4.4375, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 804, "epoch": 1, "loss": 0.1771, "lr": 1.989681452474353e-05, "grad_norm": 4.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 805, "epoch": 1, "loss": 0.0223, "lr": 1.9896384882195163e-05, "grad_norm": 5.71875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 806, "epoch": 1, "loss": 0.2407, "lr": 1.9895954351691788e-05, "grad_norm": 5.84375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 807, "epoch": 1, "loss": 0.0007, "lr": 1.9895522933272028e-05, "grad_norm": 3.984375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 808, "epoch": 1, "loss": 0.1646, "lr": 1.9895090626974602e-05, "grad_norm": 4.28125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 809, "epoch": 1, "loss": 0.0699, "lr": 1.9894657432838288e-05, "grad_norm": 4.25, "samples_per_sec": 10.52, "gpu_mem_gb": 6.85} {"step": 810, "epoch": 1, "loss": 0.0018, "lr": 1.9894223350901965e-05, "grad_norm": 2.859375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 811, "epoch": 1, "loss": 0.0003, "lr": 1.9893788381204572e-05, "grad_norm": 1.8828125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 812, "epoch": 1, "loss": 0.009, "lr": 1.9893352523785144e-05, "grad_norm": 8.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 813, "epoch": 1, "loss": 0.1248, "lr": 1.989291577868278e-05, "grad_norm": 7.53125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 814, "epoch": 1, "loss": 0.0461, "lr": 1.9892478145936675e-05, "grad_norm": 4.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 815, "epoch": 1, "loss": 0.0044, "lr": 1.9892039625586095e-05, "grad_norm": 1.59375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 816, "epoch": 1, "loss": 0.0023, "lr": 1.9891600217670382e-05, "grad_norm": 5.0625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 817, "epoch": 1, "loss": 0.0008, "lr": 1.9891159922228963e-05, "grad_norm": 4.5, "samples_per_sec": 10.72, "gpu_mem_gb": 6.85} {"step": 818, "epoch": 1, "loss": 0.1371, "lr": 1.9890718739301346e-05, "grad_norm": 2.65625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 819, "epoch": 1, "loss": 0.1004, "lr": 1.9890276668927112e-05, "grad_norm": 3.265625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 820, "epoch": 1, "loss": 0.2207, "lr": 1.988983371114593e-05, "grad_norm": 8.625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 821, "epoch": 1, "loss": 0.062, "lr": 1.988938986599754e-05, "grad_norm": 6.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 822, "epoch": 1, "loss": 0.1712, "lr": 1.9888945133521776e-05, "grad_norm": 6.40625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 823, "epoch": 1, "loss": 0.1581, "lr": 1.988849951375853e-05, "grad_norm": 8.0, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 824, "epoch": 1, "loss": 0.306, "lr": 1.988805300674779e-05, "grad_norm": 10.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 825, "epoch": 1, "loss": 0.0079, "lr": 1.9887605612529622e-05, "grad_norm": 7.53125, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 826, "epoch": 1, "loss": 0.2782, "lr": 1.9887157331144164e-05, "grad_norm": 11.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 827, "epoch": 1, "loss": 0.0001, "lr": 1.988670816263164e-05, "grad_norm": 1.1796875, "samples_per_sec": 10.45, "gpu_mem_gb": 6.82} {"step": 828, "epoch": 1, "loss": 0.4653, "lr": 1.9886258107032352e-05, "grad_norm": 11.625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 829, "epoch": 1, "loss": 0.0008, "lr": 1.988580716438668e-05, "grad_norm": 5.09375, "samples_per_sec": 10.36, "gpu_mem_gb": 6.85} {"step": 830, "epoch": 1, "loss": 0.0006, "lr": 1.9885355334735082e-05, "grad_norm": 7.0625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 831, "epoch": 1, "loss": 0.0472, "lr": 1.9884902618118108e-05, "grad_norm": 4.5625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 832, "epoch": 1, "loss": 0.0557, "lr": 1.988444901457637e-05, "grad_norm": 3.640625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 833, "epoch": 1, "loss": 0.0019, "lr": 1.9883994524150573e-05, "grad_norm": 3.265625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 834, "epoch": 1, "loss": 0.2018, "lr": 1.988353914688149e-05, "grad_norm": 5.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 835, "epoch": 1, "loss": 0.1107, "lr": 1.9883082882809983e-05, "grad_norm": 10.1875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 836, "epoch": 1, "loss": 0.0049, "lr": 1.988262573197699e-05, "grad_norm": 6.0, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 837, "epoch": 1, "loss": 0.0013, "lr": 1.988216769442353e-05, "grad_norm": 4.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 838, "epoch": 1, "loss": 0.0315, "lr": 1.9881708770190703e-05, "grad_norm": 4.53125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 839, "epoch": 1, "loss": 0.0039, "lr": 1.988124895931968e-05, "grad_norm": 4.71875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 840, "epoch": 1, "loss": 0.0722, "lr": 1.9880788261851716e-05, "grad_norm": 7.0625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 841, "epoch": 1, "loss": 0.2815, "lr": 1.988032667782816e-05, "grad_norm": 6.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 842, "epoch": 1, "loss": 0.0337, "lr": 1.9879864207290413e-05, "grad_norm": 9.8125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 843, "epoch": 1, "loss": 0.0034, "lr": 1.987940085027998e-05, "grad_norm": 7.34375, "samples_per_sec": 10.2, "gpu_mem_gb": 6.83} {"step": 844, "epoch": 1, "loss": 0.2523, "lr": 1.987893660683843e-05, "grad_norm": 6.53125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 845, "epoch": 1, "loss": 0.0999, "lr": 1.9878471477007424e-05, "grad_norm": 6.21875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 846, "epoch": 1, "loss": 0.0059, "lr": 1.987800546082869e-05, "grad_norm": 7.96875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 847, "epoch": 1, "loss": 0.0058, "lr": 1.9877538558344042e-05, "grad_norm": 3.28125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 848, "epoch": 1, "loss": 0.0324, "lr": 1.9877070769595374e-05, "grad_norm": 3.453125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 849, "epoch": 1, "loss": 0.0187, "lr": 1.9876602094624657e-05, "grad_norm": 3.234375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 850, "epoch": 1, "loss": 0.1342, "lr": 1.9876132533473946e-05, "grad_norm": 3.546875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 851, "epoch": 1, "loss": 0.0816, "lr": 1.987566208618537e-05, "grad_norm": 9.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 852, "epoch": 1, "loss": 0.31, "lr": 1.987519075280114e-05, "grad_norm": 5.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 853, "epoch": 1, "loss": 0.0595, "lr": 1.987471853336355e-05, "grad_norm": 9.1875, "samples_per_sec": 10.49, "gpu_mem_gb": 6.85} {"step": 854, "epoch": 1, "loss": 0.3174, "lr": 1.9874245427914967e-05, "grad_norm": 7.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 855, "epoch": 1, "loss": 0.0464, "lr": 1.987377143649784e-05, "grad_norm": 12.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 856, "epoch": 1, "loss": 0.0539, "lr": 1.98732965591547e-05, "grad_norm": 5.03125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 857, "epoch": 1, "loss": 0.0094, "lr": 1.987282079592815e-05, "grad_norm": 3.984375, "samples_per_sec": 10.35, "gpu_mem_gb": 6.85} {"step": 858, "epoch": 1, "loss": 0.0014, "lr": 1.9872344146860886e-05, "grad_norm": 3.375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 859, "epoch": 1, "loss": 0.2112, "lr": 1.987186661199567e-05, "grad_norm": 16.25, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 860, "epoch": 1, "loss": 0.0791, "lr": 1.987138819137535e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 861, "epoch": 1, "loss": 0.0051, "lr": 1.987090888504285e-05, "grad_norm": 2.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 862, "epoch": 1, "loss": 0.0059, "lr": 1.9870428693041184e-05, "grad_norm": 2.046875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 863, "epoch": 1, "loss": 0.2505, "lr": 1.986994761541343e-05, "grad_norm": 7.75, "samples_per_sec": 10.41, "gpu_mem_gb": 6.84} {"step": 864, "epoch": 1, "loss": 0.3001, "lr": 1.9869465652202756e-05, "grad_norm": 6.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 865, "epoch": 1, "loss": 0.0019, "lr": 1.9868982803452404e-05, "grad_norm": 7.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 866, "epoch": 1, "loss": 0.0018, "lr": 1.9868499069205697e-05, "grad_norm": 5.625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 867, "epoch": 1, "loss": 0.0012, "lr": 1.9868014449506045e-05, "grad_norm": 1.71875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 868, "epoch": 1, "loss": 0.1046, "lr": 1.986752894439692e-05, "grad_norm": 5.03125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.84} {"step": 869, "epoch": 1, "loss": 0.0045, "lr": 1.9867042553921888e-05, "grad_norm": 3.546875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 870, "epoch": 1, "loss": 0.0535, "lr": 1.9866555278124597e-05, "grad_norm": 7.59375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 871, "epoch": 1, "loss": 0.0426, "lr": 1.986606711704876e-05, "grad_norm": 6.46875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 872, "epoch": 1, "loss": 0.0867, "lr": 1.986557807073818e-05, "grad_norm": 4.3125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 873, "epoch": 1, "loss": 0.1658, "lr": 1.986508813923674e-05, "grad_norm": 16.5, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 874, "epoch": 1, "loss": 0.0719, "lr": 1.9864597322588396e-05, "grad_norm": 29.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 875, "epoch": 1, "loss": 0.0422, "lr": 1.9864105620837182e-05, "grad_norm": 6.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 876, "epoch": 1, "loss": 0.0947, "lr": 1.9863613034027224e-05, "grad_norm": 5.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 877, "epoch": 1, "loss": 0.0135, "lr": 1.9863119562202718e-05, "grad_norm": 7.34375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 878, "epoch": 1, "loss": 0.0322, "lr": 1.9862625205407936e-05, "grad_norm": 5.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 879, "epoch": 1, "loss": 0.0007, "lr": 1.986212996368724e-05, "grad_norm": 1.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 880, "epoch": 1, "loss": 0.1609, "lr": 1.986163383708506e-05, "grad_norm": 4.875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 881, "epoch": 1, "loss": 0.093, "lr": 1.9861136825645918e-05, "grad_norm": 4.6875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 882, "epoch": 1, "loss": 0.0761, "lr": 1.98606389294144e-05, "grad_norm": 5.1875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 883, "epoch": 1, "loss": 0.1272, "lr": 1.986014014843519e-05, "grad_norm": 8.8125, "samples_per_sec": 10.3, "gpu_mem_gb": 6.85} {"step": 884, "epoch": 1, "loss": 0.0163, "lr": 1.985964048275303e-05, "grad_norm": 1.375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 885, "epoch": 1, "loss": 0.1255, "lr": 1.985913993241276e-05, "grad_norm": 6.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 886, "epoch": 1, "loss": 0.0466, "lr": 1.985863849745929e-05, "grad_norm": 6.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 887, "epoch": 1, "loss": 0.1907, "lr": 1.9858136177937615e-05, "grad_norm": 9.5, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 888, "epoch": 1, "loss": 0.0054, "lr": 1.9857632973892796e-05, "grad_norm": 3.09375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 889, "epoch": 1, "loss": 0.0001, "lr": 1.9857128885369997e-05, "grad_norm": 4.53125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 890, "epoch": 1, "loss": 0.0131, "lr": 1.9856623912414433e-05, "grad_norm": 11.0625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 891, "epoch": 1, "loss": 0.2284, "lr": 1.985611805507142e-05, "grad_norm": 5.90625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 892, "epoch": 1, "loss": 0.0095, "lr": 1.9855611313386348e-05, "grad_norm": 6.75, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 893, "epoch": 1, "loss": 0.048, "lr": 1.9855103687404683e-05, "grad_norm": 5.625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 894, "epoch": 1, "loss": 0.1002, "lr": 1.9854595177171968e-05, "grad_norm": 4.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 895, "epoch": 1, "loss": 0.0022, "lr": 1.9854085782733835e-05, "grad_norm": 5.625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 896, "epoch": 1, "loss": 0.0281, "lr": 1.9853575504135986e-05, "grad_norm": 7.25, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 897, "epoch": 1, "loss": 0.0003, "lr": 1.985306434142421e-05, "grad_norm": 4.125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 898, "epoch": 1, "loss": 0.0613, "lr": 1.985255229464436e-05, "grad_norm": 8.4375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 899, "epoch": 1, "loss": 0.2333, "lr": 1.985203936384239e-05, "grad_norm": 5.46875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 900, "epoch": 1, "loss": 0.0025, "lr": 1.9851525549064324e-05, "grad_norm": 3.03125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 901, "epoch": 1, "loss": 0.0026, "lr": 1.9851010850356255e-05, "grad_norm": 7.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 902, "epoch": 1, "loss": 0.17, "lr": 1.9850495267764373e-05, "grad_norm": 8.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 903, "epoch": 1, "loss": 0.0616, "lr": 1.9849978801334935e-05, "grad_norm": 6.09375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 904, "epoch": 1, "loss": 0.1713, "lr": 1.984946145111428e-05, "grad_norm": 7.8125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 905, "epoch": 1, "loss": 0.0286, "lr": 1.984894321714883e-05, "grad_norm": 1.8984375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 906, "epoch": 1, "loss": 0.0014, "lr": 1.984842409948508e-05, "grad_norm": 2.921875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 907, "epoch": 1, "loss": 0.0035, "lr": 1.9847904098169612e-05, "grad_norm": 6.21875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 908, "epoch": 1, "loss": 0.0004, "lr": 1.9847383213249082e-05, "grad_norm": 5.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 909, "epoch": 1, "loss": 0.0049, "lr": 1.9846861444770226e-05, "grad_norm": 3.46875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 910, "epoch": 1, "loss": 0.3496, "lr": 1.984633879277986e-05, "grad_norm": 6.65625, "samples_per_sec": 10.37, "gpu_mem_gb": 6.84} {"step": 911, "epoch": 1, "loss": 0.0469, "lr": 1.984581525732488e-05, "grad_norm": 4.4375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 912, "epoch": 1, "loss": 0.0013, "lr": 1.9845290838452257e-05, "grad_norm": 6.875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 913, "epoch": 1, "loss": 0.025, "lr": 1.9844765536209045e-05, "grad_norm": 3.09375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 914, "epoch": 1, "loss": 0.3153, "lr": 1.9844239350642382e-05, "grad_norm": 5.3125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 915, "epoch": 1, "loss": 0.236, "lr": 1.9843712281799473e-05, "grad_norm": 4.75, "samples_per_sec": 10.51, "gpu_mem_gb": 6.84} {"step": 916, "epoch": 1, "loss": 0.3777, "lr": 1.984318432972762e-05, "grad_norm": 9.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 917, "epoch": 1, "loss": 0.2822, "lr": 1.9842655494474178e-05, "grad_norm": 9.5625, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 918, "epoch": 1, "loss": 0.0995, "lr": 1.984212577608661e-05, "grad_norm": 4.65625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 919, "epoch": 1, "loss": 0.0277, "lr": 1.984159517461244e-05, "grad_norm": 5.03125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 920, "epoch": 1, "loss": 0.1459, "lr": 1.9841063690099278e-05, "grad_norm": 4.375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 921, "epoch": 1, "loss": 0.0044, "lr": 1.9840531322594808e-05, "grad_norm": 4.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 922, "epoch": 1, "loss": 0.173, "lr": 1.98399980721468e-05, "grad_norm": 6.375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 923, "epoch": 1, "loss": 0.0054, "lr": 1.9839463938803103e-05, "grad_norm": 5.03125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 924, "epoch": 1, "loss": 0.0093, "lr": 1.9838928922611634e-05, "grad_norm": 5.09375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 925, "epoch": 1, "loss": 0.0975, "lr": 1.9838393023620402e-05, "grad_norm": 6.3125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 926, "epoch": 1, "loss": 0.1374, "lr": 1.9837856241877493e-05, "grad_norm": 9.9375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 927, "epoch": 1, "loss": 0.0145, "lr": 1.9837318577431068e-05, "grad_norm": 4.0, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 928, "epoch": 1, "loss": 0.006, "lr": 1.9836780030329365e-05, "grad_norm": 5.6875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 929, "epoch": 1, "loss": 0.0507, "lr": 1.983624060062071e-05, "grad_norm": 4.40625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 930, "epoch": 1, "loss": 0.002, "lr": 1.9835700288353502e-05, "grad_norm": 4.34375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 931, "epoch": 1, "loss": 0.0009, "lr": 1.983515909357622e-05, "grad_norm": 3.109375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 932, "epoch": 1, "loss": 0.0113, "lr": 1.9834617016337424e-05, "grad_norm": 4.21875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 933, "epoch": 1, "loss": 0.2492, "lr": 1.983407405668575e-05, "grad_norm": 6.125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.84} {"step": 934, "epoch": 1, "loss": 0.0033, "lr": 1.9833530214669917e-05, "grad_norm": 2.640625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 935, "epoch": 1, "loss": 0.0175, "lr": 1.9832985490338724e-05, "grad_norm": 2.84375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 936, "epoch": 1, "loss": 0.0078, "lr": 1.9832439883741038e-05, "grad_norm": 5.78125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 937, "epoch": 1, "loss": 0.0486, "lr": 1.9831893394925825e-05, "grad_norm": 6.46875, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 938, "epoch": 1, "loss": 0.1356, "lr": 1.983134602394211e-05, "grad_norm": 3.15625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 939, "epoch": 1, "loss": 0.0731, "lr": 1.9830797770839008e-05, "grad_norm": 4.625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 940, "epoch": 1, "loss": 0.0696, "lr": 1.9830248635665718e-05, "grad_norm": 5.6875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 941, "epoch": 1, "loss": 0.0395, "lr": 1.9829698618471497e-05, "grad_norm": 3.78125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 942, "epoch": 1, "loss": 0.0074, "lr": 1.9829147719305706e-05, "grad_norm": 5.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 943, "epoch": 1, "loss": 0.0031, "lr": 1.9828595938217777e-05, "grad_norm": 3.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 944, "epoch": 1, "loss": 0.0026, "lr": 1.982804327525721e-05, "grad_norm": 5.46875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 945, "epoch": 1, "loss": 0.0084, "lr": 1.9827489730473597e-05, "grad_norm": 1.8515625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 946, "epoch": 1, "loss": 0.0172, "lr": 1.9826935303916605e-05, "grad_norm": 7.8125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 947, "epoch": 1, "loss": 0.1173, "lr": 1.9826379995635977e-05, "grad_norm": 5.5625, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 948, "epoch": 1, "loss": 0.1496, "lr": 1.9825823805681543e-05, "grad_norm": 5.40625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 949, "epoch": 1, "loss": 0.0011, "lr": 1.9825266734103203e-05, "grad_norm": 3.390625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 950, "epoch": 1, "loss": 0.2232, "lr": 1.9824708780950942e-05, "grad_norm": 7.09375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 951, "epoch": 1, "loss": 0.1278, "lr": 1.9824149946274827e-05, "grad_norm": 3.453125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 952, "epoch": 1, "loss": 0.0732, "lr": 1.9823590230124995e-05, "grad_norm": 4.34375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 953, "epoch": 1, "loss": 0.0534, "lr": 1.982302963255166e-05, "grad_norm": 4.9375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 954, "epoch": 1, "loss": 0.0019, "lr": 1.9822468153605137e-05, "grad_norm": 6.78125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 955, "epoch": 1, "loss": 0.2542, "lr": 1.9821905793335793e-05, "grad_norm": 6.46875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 956, "epoch": 1, "loss": 0.0999, "lr": 1.982134255179409e-05, "grad_norm": 4.90625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 957, "epoch": 1, "loss": 0.1247, "lr": 1.982077842903056e-05, "grad_norm": 4.09375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 958, "epoch": 1, "loss": 0.0062, "lr": 1.982021342509583e-05, "grad_norm": 3.03125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 959, "epoch": 1, "loss": 0.0379, "lr": 1.9819647540040585e-05, "grad_norm": 10.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 960, "epoch": 1, "loss": 0.0061, "lr": 1.9819080773915603e-05, "grad_norm": 1.859375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 961, "epoch": 1, "loss": 0.3088, "lr": 1.9818513126771734e-05, "grad_norm": 10.625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 962, "epoch": 1, "loss": 0.0055, "lr": 1.9817944598659918e-05, "grad_norm": 4.1875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 963, "epoch": 1, "loss": 0.0025, "lr": 1.9817375189631157e-05, "grad_norm": 9.0625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 964, "epoch": 1, "loss": 0.0656, "lr": 1.981680489973655e-05, "grad_norm": 5.75, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 965, "epoch": 1, "loss": 0.0454, "lr": 1.981623372902726e-05, "grad_norm": 5.1875, "samples_per_sec": 8.73, "gpu_mem_gb": 6.85} {"step": 966, "epoch": 1, "loss": 0.0228, "lr": 1.9815661677554537e-05, "grad_norm": 3.515625, "samples_per_sec": 11.41, "gpu_mem_gb": 6.85} {"step": 967, "epoch": 1, "loss": 0.0582, "lr": 1.9815088745369708e-05, "grad_norm": 10.6875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 968, "epoch": 1, "loss": 0.0948, "lr": 1.981451493252418e-05, "grad_norm": 8.1875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 969, "epoch": 1, "loss": 0.0446, "lr": 1.981394023906944e-05, "grad_norm": 4.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 970, "epoch": 1, "loss": 0.0026, "lr": 1.981336466505705e-05, "grad_norm": 4.6875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 971, "epoch": 1, "loss": 0.0401, "lr": 1.9812788210538657e-05, "grad_norm": 5.71875, "samples_per_sec": 10.45, "gpu_mem_gb": 6.82} {"step": 972, "epoch": 1, "loss": 0.0511, "lr": 1.981221087556598e-05, "grad_norm": 4.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 973, "epoch": 1, "loss": 0.0734, "lr": 1.981163266019082e-05, "grad_norm": 4.75, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 974, "epoch": 1, "loss": 0.1565, "lr": 1.9811053564465062e-05, "grad_norm": 7.59375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 975, "epoch": 1, "loss": 0.0101, "lr": 1.9810473588440662e-05, "grad_norm": 2.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 976, "epoch": 1, "loss": 0.0017, "lr": 1.980989273216966e-05, "grad_norm": 3.625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 977, "epoch": 1, "loss": 0.1917, "lr": 1.980931099570417e-05, "grad_norm": 7.25, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 978, "epoch": 1, "loss": 0.0009, "lr": 1.980872837909639e-05, "grad_norm": 4.84375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 979, "epoch": 1, "loss": 0.0368, "lr": 1.98081448823986e-05, "grad_norm": 3.4375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 980, "epoch": 1, "loss": 0.0615, "lr": 1.980756050566315e-05, "grad_norm": 5.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 981, "epoch": 1, "loss": 0.1905, "lr": 1.9806975248942472e-05, "grad_norm": 3.78125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 982, "epoch": 1, "loss": 0.0864, "lr": 1.9806389112289078e-05, "grad_norm": 8.8125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 983, "epoch": 1, "loss": 0.006, "lr": 1.980580209575557e-05, "grad_norm": 4.9375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 984, "epoch": 1, "loss": 0.1541, "lr": 1.98052141993946e-05, "grad_norm": 6.84375, "samples_per_sec": 10.37, "gpu_mem_gb": 6.82} {"step": 985, "epoch": 1, "loss": 0.2445, "lr": 1.9804625423258933e-05, "grad_norm": 4.96875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 986, "epoch": 1, "loss": 0.0045, "lr": 1.980403576740139e-05, "grad_norm": 4.71875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 987, "epoch": 1, "loss": 0.0101, "lr": 1.9803445231874877e-05, "grad_norm": 6.46875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 988, "epoch": 1, "loss": 0.0062, "lr": 1.9802853816732384e-05, "grad_norm": 3.15625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 989, "epoch": 1, "loss": 0.0857, "lr": 1.980226152202697e-05, "grad_norm": 12.0625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 990, "epoch": 1, "loss": 0.174, "lr": 1.9801668347811786e-05, "grad_norm": 4.6875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 991, "epoch": 1, "loss": 0.0003, "lr": 1.980107429414005e-05, "grad_norm": 2.875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 992, "epoch": 1, "loss": 0.0556, "lr": 1.980047936106506e-05, "grad_norm": 6.90625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 993, "epoch": 1, "loss": 0.0067, "lr": 1.9799883548640208e-05, "grad_norm": 2.765625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 994, "epoch": 1, "loss": 0.1205, "lr": 1.9799286856918944e-05, "grad_norm": 7.84375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 995, "epoch": 1, "loss": 0.0088, "lr": 1.979868928595481e-05, "grad_norm": 3.1875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 996, "epoch": 1, "loss": 0.045, "lr": 1.9798090835801418e-05, "grad_norm": 3.34375, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 997, "epoch": 1, "loss": 0.0019, "lr": 1.979749150651247e-05, "grad_norm": 4.25, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 998, "epoch": 1, "loss": 0.0016, "lr": 1.979689129814174e-05, "grad_norm": 2.65625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 999, "epoch": 1, "loss": 0.0019, "lr": 1.979629021074308e-05, "grad_norm": 5.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1000, "epoch": 1, "loss": 0.0394, "lr": 1.9795688244370423e-05, "grad_norm": 9.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1001, "epoch": 1, "loss": 0.0856, "lr": 1.979508539907778e-05, "grad_norm": 10.875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1002, "epoch": 1, "loss": 0.1628, "lr": 1.979448167491924e-05, "grad_norm": 5.375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.84} {"step": 1003, "epoch": 1, "loss": 0.0063, "lr": 1.9793877071948978e-05, "grad_norm": 3.921875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1004, "epoch": 1, "loss": 0.0472, "lr": 1.979327159022124e-05, "grad_norm": 3.015625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 1005, "epoch": 1, "loss": 0.0331, "lr": 1.9792665229790347e-05, "grad_norm": 7.96875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 1006, "epoch": 1, "loss": 0.0357, "lr": 1.9792057990710717e-05, "grad_norm": 2.546875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1007, "epoch": 1, "loss": 0.1143, "lr": 1.979144987303682e-05, "grad_norm": 4.9375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1008, "epoch": 1, "loss": 0.3629, "lr": 1.979084087682323e-05, "grad_norm": 6.8125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1009, "epoch": 1, "loss": 0.0002, "lr": 1.9790231002124586e-05, "grad_norm": 4.875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1010, "epoch": 1, "loss": 0.0634, "lr": 1.978962024899561e-05, "grad_norm": 3.453125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1011, "epoch": 1, "loss": 0.0767, "lr": 1.9789008617491096e-05, "grad_norm": 5.09375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1012, "epoch": 1, "loss": 0.0468, "lr": 1.9788396107665932e-05, "grad_norm": 5.5, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1013, "epoch": 1, "loss": 0.0886, "lr": 1.978778271957507e-05, "grad_norm": 5.15625, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 1014, "epoch": 1, "loss": 0.1203, "lr": 1.9787168453273546e-05, "grad_norm": 6.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1015, "epoch": 1, "loss": 0.0002, "lr": 1.978655330881648e-05, "grad_norm": 4.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1016, "epoch": 1, "loss": 0.1515, "lr": 1.978593728625906e-05, "grad_norm": 4.0625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1017, "epoch": 1, "loss": 0.0009, "lr": 1.978532038565656e-05, "grad_norm": 5.71875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1018, "epoch": 1, "loss": 0.0055, "lr": 1.9784702607064335e-05, "grad_norm": 5.78125, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 1019, "epoch": 1, "loss": 0.0032, "lr": 1.9784083950537818e-05, "grad_norm": 2.515625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 1020, "epoch": 1, "loss": 0.1609, "lr": 1.9783464416132507e-05, "grad_norm": 7.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1021, "epoch": 1, "loss": 0.1848, "lr": 1.9782844003904002e-05, "grad_norm": 11.9375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 1022, "epoch": 1, "loss": 0.0204, "lr": 1.9782222713907956e-05, "grad_norm": 5.71875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1023, "epoch": 1, "loss": 0.1083, "lr": 1.9781600546200126e-05, "grad_norm": 6.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1024, "epoch": 1, "loss": 0.0477, "lr": 1.9780977500836334e-05, "grad_norm": 8.0, "samples_per_sec": 10.35, "gpu_mem_gb": 6.82} {"step": 1025, "epoch": 1, "loss": 0.1707, "lr": 1.978035357787248e-05, "grad_norm": 6.71875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1026, "epoch": 1, "loss": 0.0075, "lr": 1.9779728777364548e-05, "grad_norm": 2.65625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1027, "epoch": 1, "loss": 0.0393, "lr": 1.9779103099368596e-05, "grad_norm": 5.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1028, "epoch": 1, "loss": 0.0032, "lr": 1.9778476543940762e-05, "grad_norm": 5.96875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1029, "epoch": 1, "loss": 0.0062, "lr": 1.9777849111137265e-05, "grad_norm": 5.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1030, "epoch": 1, "loss": 0.0028, "lr": 1.977722080101441e-05, "grad_norm": 5.5625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1031, "epoch": 1, "loss": 0.0171, "lr": 1.9776591613628558e-05, "grad_norm": 5.1875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 1032, "epoch": 1, "loss": 0.0105, "lr": 1.977596154903617e-05, "grad_norm": 6.09375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 1033, "epoch": 1, "loss": 0.0292, "lr": 1.9775330607293782e-05, "grad_norm": 3.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 1034, "epoch": 1, "loss": 0.2039, "lr": 1.9774698788458e-05, "grad_norm": 6.5625, "samples_per_sec": 10.27, "gpu_mem_gb": 6.83} {"step": 1035, "epoch": 1, "loss": 0.0261, "lr": 1.9774066092585517e-05, "grad_norm": 2.8125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1036, "epoch": 1, "loss": 0.0219, "lr": 1.9773432519733098e-05, "grad_norm": 2.875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1037, "epoch": 1, "loss": 0.0014, "lr": 1.9772798069957596e-05, "grad_norm": 4.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1038, "epoch": 1, "loss": 0.1638, "lr": 1.977216274331593e-05, "grad_norm": 5.75, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 1039, "epoch": 1, "loss": 0.0645, "lr": 1.9771526539865112e-05, "grad_norm": 5.4375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 1040, "epoch": 1, "loss": 0.0445, "lr": 1.9770889459662223e-05, "grad_norm": 3.328125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 1041, "epoch": 1, "loss": 0.0034, "lr": 1.977025150276442e-05, "grad_norm": 4.34375, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 1042, "epoch": 1, "loss": 0.0393, "lr": 1.9769612669228952e-05, "grad_norm": 3.53125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 1043, "epoch": 1, "loss": 0.0018, "lr": 1.9768972959113136e-05, "grad_norm": 3.984375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1044, "epoch": 1, "loss": 0.1009, "lr": 1.976833237247437e-05, "grad_norm": 3.671875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1045, "epoch": 1, "loss": 0.0096, "lr": 1.9767690909370127e-05, "grad_norm": 3.671875, "samples_per_sec": 10.31, "gpu_mem_gb": 6.83} {"step": 1046, "epoch": 1, "loss": 0.0067, "lr": 1.9767048569857963e-05, "grad_norm": 7.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1047, "epoch": 1, "loss": 0.0211, "lr": 1.9766405353995517e-05, "grad_norm": 5.78125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1048, "epoch": 1, "loss": 0.0671, "lr": 1.9765761261840496e-05, "grad_norm": 6.46875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 1049, "epoch": 1, "loss": 0.0083, "lr": 1.97651162934507e-05, "grad_norm": 5.46875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 1050, "epoch": 1, "loss": 0.0013, "lr": 1.9764470448883987e-05, "grad_norm": 4.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1051, "epoch": 1, "loss": 0.0053, "lr": 1.9763823728198313e-05, "grad_norm": 6.0625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1052, "epoch": 1, "loss": 0.1651, "lr": 1.9763176131451706e-05, "grad_norm": 4.53125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1053, "epoch": 1, "loss": 0.1188, "lr": 1.9762527658702266e-05, "grad_norm": 6.53125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1054, "epoch": 1, "loss": 0.0018, "lr": 1.9761878310008185e-05, "grad_norm": 6.28125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1055, "epoch": 1, "loss": 0.1762, "lr": 1.9761228085427718e-05, "grad_norm": 6.625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1056, "epoch": 1, "loss": 0.3448, "lr": 1.976057698501921e-05, "grad_norm": 5.90625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.84} {"step": 1057, "epoch": 1, "loss": 0.0577, "lr": 1.9759925008841085e-05, "grad_norm": 3.65625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1058, "epoch": 1, "loss": 0.0067, "lr": 1.9759272156951836e-05, "grad_norm": 8.1875, "samples_per_sec": 10.34, "gpu_mem_gb": 6.82} {"step": 1059, "epoch": 1, "loss": 0.0009, "lr": 1.975861842941004e-05, "grad_norm": 1.3359375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1060, "epoch": 1, "loss": 0.013, "lr": 1.9757963826274357e-05, "grad_norm": 7.375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1061, "epoch": 1, "loss": 0.0162, "lr": 1.975730834760352e-05, "grad_norm": 2.953125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.85} {"step": 1062, "epoch": 1, "loss": 0.0486, "lr": 1.975665199345634e-05, "grad_norm": 1.8125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 1063, "epoch": 1, "loss": 0.3635, "lr": 1.9755994763891712e-05, "grad_norm": 7.40625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1064, "epoch": 1, "loss": 0.005, "lr": 1.9755336658968606e-05, "grad_norm": 2.28125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1065, "epoch": 1, "loss": 0.1099, "lr": 1.9754677678746064e-05, "grad_norm": 3.046875, "samples_per_sec": 10.39, "gpu_mem_gb": 6.85} {"step": 1066, "epoch": 1, "loss": 0.0061, "lr": 1.9754017823283215e-05, "grad_norm": 1.953125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1067, "epoch": 1, "loss": 0.0026, "lr": 1.9753357092639274e-05, "grad_norm": 2.46875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1068, "epoch": 1, "loss": 0.0007, "lr": 1.9752695486873516e-05, "grad_norm": 2.078125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1069, "epoch": 1, "loss": 0.0005, "lr": 1.9752033006045305e-05, "grad_norm": 2.875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1070, "epoch": 1, "loss": 0.0004, "lr": 1.9751369650214085e-05, "grad_norm": 1.640625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1071, "epoch": 1, "loss": 0.0053, "lr": 1.9750705419439376e-05, "grad_norm": 1.890625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1072, "epoch": 1, "loss": 0.028, "lr": 1.975004031378077e-05, "grad_norm": 3.53125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1073, "epoch": 1, "loss": 0.0037, "lr": 1.974937433329795e-05, "grad_norm": 2.375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 1074, "epoch": 1, "loss": 0.3217, "lr": 1.974870747805067e-05, "grad_norm": 5.03125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.84} {"step": 1075, "epoch": 1, "loss": 0.0427, "lr": 1.9748039748098762e-05, "grad_norm": 3.828125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1076, "epoch": 1, "loss": 0.0407, "lr": 1.974737114350214e-05, "grad_norm": 6.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1077, "epoch": 1, "loss": 0.1477, "lr": 1.9746701664320796e-05, "grad_norm": 23.625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1078, "epoch": 1, "loss": 0.1786, "lr": 1.9746031310614793e-05, "grad_norm": 3.75, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1079, "epoch": 1, "loss": 0.0008, "lr": 1.9745360082444285e-05, "grad_norm": 5.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1080, "epoch": 1, "loss": 0.0001, "lr": 1.9744687979869494e-05, "grad_norm": 0.29296875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1081, "epoch": 1, "loss": 0.2418, "lr": 1.9744015002950728e-05, "grad_norm": 7.375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1082, "epoch": 1, "loss": 0.0043, "lr": 1.9743341151748368e-05, "grad_norm": 6.84375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.82} {"step": 1083, "epoch": 1, "loss": 0.009, "lr": 1.9742666426322877e-05, "grad_norm": 4.9375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1084, "epoch": 1, "loss": 0.896, "lr": 1.9741990826734793e-05, "grad_norm": 11.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1085, "epoch": 1, "loss": 0.0002, "lr": 1.9741314353044734e-05, "grad_norm": 3.609375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1086, "epoch": 1, "loss": 0.067, "lr": 1.9740637005313397e-05, "grad_norm": 9.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1087, "epoch": 1, "loss": 0.2003, "lr": 1.973995878360156e-05, "grad_norm": 9.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1088, "epoch": 1, "loss": 0.0012, "lr": 1.973927968797007e-05, "grad_norm": 5.875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 1089, "epoch": 1, "loss": 0.0005, "lr": 1.9738599718479866e-05, "grad_norm": 0.1416015625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 1090, "epoch": 1, "loss": 0.0575, "lr": 1.9737918875191957e-05, "grad_norm": 10.125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1091, "epoch": 1, "loss": 0.0569, "lr": 1.973723715816743e-05, "grad_norm": 3.390625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1092, "epoch": 1, "loss": 0.044, "lr": 1.973655456746745e-05, "grad_norm": 4.84375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1093, "epoch": 1, "loss": 0.0055, "lr": 1.973587110315327e-05, "grad_norm": 2.859375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1094, "epoch": 1, "loss": 0.0008, "lr": 1.9735186765286206e-05, "grad_norm": 10.0625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 1095, "epoch": 1, "loss": 0.1061, "lr": 1.9734501553927667e-05, "grad_norm": 4.03125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1096, "epoch": 1, "loss": 0.0008, "lr": 1.9733815469139123e-05, "grad_norm": 5.3125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 1097, "epoch": 1, "loss": 0.0042, "lr": 1.973312851098215e-05, "grad_norm": 6.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1098, "epoch": 1, "loss": 0.002, "lr": 1.973244067951837e-05, "grad_norm": 5.71875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 1099, "epoch": 1, "loss": 0.0018, "lr": 1.9731751974809503e-05, "grad_norm": 5.90625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1100, "epoch": 1, "loss": 0.1786, "lr": 1.9731062396917346e-05, "grad_norm": 6.96875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1101, "epoch": 1, "loss": 0.001, "lr": 1.9730371945903773e-05, "grad_norm": 3.828125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1102, "epoch": 1, "loss": 0.2539, "lr": 1.972968062183073e-05, "grad_norm": 7.03125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 1103, "epoch": 1, "loss": 0.0002, "lr": 1.972898842476025e-05, "grad_norm": 5.0625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1104, "epoch": 1, "loss": 0.4997, "lr": 1.9728295354754436e-05, "grad_norm": 7.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1105, "epoch": 1, "loss": 0.0146, "lr": 1.972760141187548e-05, "grad_norm": 5.65625, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 1106, "epoch": 1, "loss": 0.0017, "lr": 1.972690659618564e-05, "grad_norm": 5.84375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.85} {"step": 1107, "epoch": 1, "loss": 0.0104, "lr": 1.9726210907747264e-05, "grad_norm": 10.125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1108, "epoch": 1, "loss": 0.2507, "lr": 1.9725514346622766e-05, "grad_norm": 6.8125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 1109, "epoch": 1, "loss": 0.0904, "lr": 1.9724816912874653e-05, "grad_norm": 9.5625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1110, "epoch": 1, "loss": 0.0061, "lr": 1.97241186065655e-05, "grad_norm": 1.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1111, "epoch": 1, "loss": 0.002, "lr": 1.972341942775796e-05, "grad_norm": 2.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1112, "epoch": 1, "loss": 0.0091, "lr": 1.9722719376514767e-05, "grad_norm": 4.875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 1113, "epoch": 1, "loss": 0.0023, "lr": 1.9722018452898733e-05, "grad_norm": 5.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1114, "epoch": 1, "loss": 0.0566, "lr": 1.972131665697275e-05, "grad_norm": 7.03125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1115, "epoch": 1, "loss": 0.1744, "lr": 1.972061398879979e-05, "grad_norm": 2.828125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 1116, "epoch": 1, "loss": 0.0206, "lr": 1.9719910448442893e-05, "grad_norm": 3.890625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1117, "epoch": 1, "loss": 0.0553, "lr": 1.971920603596519e-05, "grad_norm": 4.78125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1118, "epoch": 1, "loss": 0.13, "lr": 1.9718500751429885e-05, "grad_norm": 7.28125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1119, "epoch": 1, "loss": 0.0038, "lr": 1.971779459490025e-05, "grad_norm": 4.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1120, "epoch": 1, "loss": 0.1213, "lr": 1.971708756643966e-05, "grad_norm": 5.65625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 1121, "epoch": 1, "loss": 0.0234, "lr": 1.9716379666111544e-05, "grad_norm": 6.25, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1122, "epoch": 1, "loss": 0.0007, "lr": 1.9715670893979416e-05, "grad_norm": 3.8125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1123, "epoch": 1, "loss": 0.0096, "lr": 1.9714961250106878e-05, "grad_norm": 4.78125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1124, "epoch": 1, "loss": 0.1581, "lr": 1.97142507345576e-05, "grad_norm": 9.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 1125, "epoch": 1, "loss": 0.0095, "lr": 1.971353934739533e-05, "grad_norm": 7.03125, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 1126, "epoch": 1, "loss": 0.2834, "lr": 1.9712827088683905e-05, "grad_norm": 6.53125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1127, "epoch": 1, "loss": 0.0829, "lr": 1.9712113958487226e-05, "grad_norm": 3.921875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1128, "epoch": 1, "loss": 0.3502, "lr": 1.9711399956869278e-05, "grad_norm": 7.8125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1129, "epoch": 1, "loss": 0.054, "lr": 1.971068508389413e-05, "grad_norm": 2.703125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 1130, "epoch": 1, "loss": 0.0371, "lr": 1.9709969339625922e-05, "grad_norm": 3.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1131, "epoch": 1, "loss": 0.0065, "lr": 1.970925272412887e-05, "grad_norm": 0.58984375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1132, "epoch": 1, "loss": 0.0005, "lr": 1.9708535237467278e-05, "grad_norm": 3.734375, "samples_per_sec": 10.25, "gpu_mem_gb": 6.83} {"step": 1133, "epoch": 1, "loss": 0.0004, "lr": 1.9707816879705524e-05, "grad_norm": 2.5625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1134, "epoch": 1, "loss": 0.05, "lr": 1.9707097650908057e-05, "grad_norm": 2.984375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1135, "epoch": 1, "loss": 0.0549, "lr": 1.9706377551139414e-05, "grad_norm": 3.515625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 1136, "epoch": 1, "loss": 0.1972, "lr": 1.9705656580464204e-05, "grad_norm": 6.6875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1137, "epoch": 1, "loss": 0.0002, "lr": 1.970493473894712e-05, "grad_norm": 6.21875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1138, "epoch": 1, "loss": 0.2325, "lr": 1.970421202665292e-05, "grad_norm": 9.6875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 1139, "epoch": 1, "loss": 0.1542, "lr": 1.9703488443646465e-05, "grad_norm": 4.53125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1140, "epoch": 1, "loss": 0.135, "lr": 1.970276398999266e-05, "grad_norm": 4.84375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 1141, "epoch": 1, "loss": 0.1123, "lr": 1.9702038665756522e-05, "grad_norm": 7.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1142, "epoch": 1, "loss": 0.1328, "lr": 1.9701312471003123e-05, "grad_norm": 9.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1143, "epoch": 1, "loss": 0.0295, "lr": 1.9700585405797623e-05, "grad_norm": 5.90625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 1144, "epoch": 1, "loss": 0.0001, "lr": 1.9699857470205258e-05, "grad_norm": 3.796875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1145, "epoch": 1, "loss": 0.0004, "lr": 1.9699128664291344e-05, "grad_norm": 4.34375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1146, "epoch": 1, "loss": 0.0003, "lr": 1.969839898812127e-05, "grad_norm": 3.828125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1147, "epoch": 1, "loss": 0.0056, "lr": 1.9697668441760508e-05, "grad_norm": 2.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1148, "epoch": 1, "loss": 0.1254, "lr": 1.9696937025274608e-05, "grad_norm": 4.625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1149, "epoch": 1, "loss": 0.1719, "lr": 1.969620473872919e-05, "grad_norm": 8.625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1150, "epoch": 1, "loss": 0.0697, "lr": 1.9695471582189972e-05, "grad_norm": 4.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1151, "epoch": 1, "loss": 0.0748, "lr": 1.9694737555722722e-05, "grad_norm": 6.21875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1152, "epoch": 1, "loss": 0.3767, "lr": 1.9694002659393306e-05, "grad_norm": 6.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1153, "epoch": 1, "loss": 0.003, "lr": 1.9693266893267665e-05, "grad_norm": 2.640625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1154, "epoch": 1, "loss": 0.0426, "lr": 1.9692530257411812e-05, "grad_norm": 3.125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1155, "epoch": 1, "loss": 0.0017, "lr": 1.9691792751891847e-05, "grad_norm": 4.1875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1156, "epoch": 1, "loss": 0.0026, "lr": 1.9691054376773936e-05, "grad_norm": 1.5078125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1157, "epoch": 1, "loss": 0.13, "lr": 1.9690315132124337e-05, "grad_norm": 9.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 1158, "epoch": 1, "loss": 0.002, "lr": 1.9689575018009374e-05, "grad_norm": 5.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1159, "epoch": 1, "loss": 0.0005, "lr": 1.9688834034495454e-05, "grad_norm": 1.796875, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 1160, "epoch": 1, "loss": 0.0003, "lr": 1.9688092181649065e-05, "grad_norm": 1.9453125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1161, "epoch": 1, "loss": 0.0011, "lr": 1.9687349459536768e-05, "grad_norm": 3.859375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 1162, "epoch": 1, "loss": 0.2327, "lr": 1.9686605868225203e-05, "grad_norm": 4.34375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1163, "epoch": 1, "loss": 0.0687, "lr": 1.968586140778109e-05, "grad_norm": 4.09375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1164, "epoch": 1, "loss": 0.1242, "lr": 1.9685116078271224e-05, "grad_norm": 6.6875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1165, "epoch": 1, "loss": 0.2363, "lr": 1.9684369879762486e-05, "grad_norm": 6.53125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1166, "epoch": 1, "loss": 0.1244, "lr": 1.968362281232182e-05, "grad_norm": 5.03125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1167, "epoch": 1, "loss": 0.0006, "lr": 1.9682874876016263e-05, "grad_norm": 5.28125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1168, "epoch": 1, "loss": 0.2412, "lr": 1.9682126070912918e-05, "grad_norm": 5.59375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 1169, "epoch": 1, "loss": 0.0289, "lr": 1.968137639707898e-05, "grad_norm": 1.7578125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1170, "epoch": 1, "loss": 0.0387, "lr": 1.968062585458171e-05, "grad_norm": 3.609375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1171, "epoch": 1, "loss": 0.01, "lr": 1.9679874443488447e-05, "grad_norm": 3.578125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1172, "epoch": 1, "loss": 0.0034, "lr": 1.9679122163866614e-05, "grad_norm": 0.4375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 1173, "epoch": 1, "loss": 0.0028, "lr": 1.967836901578371e-05, "grad_norm": 2.453125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 1174, "epoch": 1, "loss": 0.0022, "lr": 1.967761499930731e-05, "grad_norm": 3.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1175, "epoch": 1, "loss": 0.0004, "lr": 1.9676860114505073e-05, "grad_norm": 5.125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 1176, "epoch": 1, "loss": 0.0188, "lr": 1.9676104361444724e-05, "grad_norm": 0.9921875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 1177, "epoch": 1, "loss": 0.004, "lr": 1.9675347740194076e-05, "grad_norm": 8.0625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 1178, "epoch": 1, "loss": 0.5679, "lr": 1.9674590250821017e-05, "grad_norm": 9.5, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1179, "epoch": 1, "loss": 0.005, "lr": 1.967383189339352e-05, "grad_norm": 6.09375, "samples_per_sec": 9.89, "gpu_mem_gb": 6.83} {"step": 1180, "epoch": 1, "loss": 0.0007, "lr": 1.9673072667979615e-05, "grad_norm": 2.015625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1181, "epoch": 1, "loss": 0.209, "lr": 1.9672312574647434e-05, "grad_norm": 3.609375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1182, "epoch": 1, "loss": 0.0758, "lr": 1.967155161346517e-05, "grad_norm": 9.875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 1183, "epoch": 1, "loss": 0.1703, "lr": 1.9670789784501104e-05, "grad_norm": 4.53125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1184, "epoch": 1, "loss": 0.0184, "lr": 1.9670027087823595e-05, "grad_norm": 6.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1185, "epoch": 1, "loss": 0.0006, "lr": 1.9669263523501066e-05, "grad_norm": 6.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1186, "epoch": 1, "loss": 0.2362, "lr": 1.966849909160204e-05, "grad_norm": 7.90625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.84} {"step": 1187, "epoch": 1, "loss": 0.1456, "lr": 1.9667733792195097e-05, "grad_norm": 7.15625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1188, "epoch": 1, "loss": 0.1835, "lr": 1.9666967625348907e-05, "grad_norm": 7.53125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1189, "epoch": 1, "loss": 0.0518, "lr": 1.9666200591132213e-05, "grad_norm": 4.59375, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 1190, "epoch": 1, "loss": 0.1402, "lr": 1.9665432689613836e-05, "grad_norm": 4.71875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 1191, "epoch": 1, "loss": 0.0002, "lr": 1.9664663920862684e-05, "grad_norm": 6.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1192, "epoch": 1, "loss": 0.059, "lr": 1.9663894284947728e-05, "grad_norm": 3.84375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1193, "epoch": 1, "loss": 0.3595, "lr": 1.966312378193802e-05, "grad_norm": 5.5625, "samples_per_sec": 8.93, "gpu_mem_gb": 6.83} {"step": 1194, "epoch": 1, "loss": 0.002, "lr": 1.9662352411902702e-05, "grad_norm": 3.859375, "samples_per_sec": 11.79, "gpu_mem_gb": 6.85} {"step": 1195, "epoch": 1, "loss": 0.3192, "lr": 1.9661580174910982e-05, "grad_norm": 4.53125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1196, "epoch": 1, "loss": 0.0475, "lr": 1.9660807071032145e-05, "grad_norm": 3.859375, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 1197, "epoch": 1, "loss": 0.0033, "lr": 1.9660033100335564e-05, "grad_norm": 3.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1198, "epoch": 1, "loss": 0.0439, "lr": 1.9659258262890683e-05, "grad_norm": 3.859375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1199, "epoch": 1, "loss": 0.1297, "lr": 1.9658482558767024e-05, "grad_norm": 4.71875, "samples_per_sec": 10.35, "gpu_mem_gb": 6.83} {"step": 1200, "epoch": 1, "loss": 0.0128, "lr": 1.9657705988034182e-05, "grad_norm": 3.671875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1201, "epoch": 1, "loss": 0.068, "lr": 1.9656928550761844e-05, "grad_norm": 4.90625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1202, "epoch": 1, "loss": 0.0006, "lr": 1.9656150247019757e-05, "grad_norm": 3.578125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1203, "epoch": 1, "loss": 0.0928, "lr": 1.965537107687776e-05, "grad_norm": 6.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 1204, "epoch": 1, "loss": 0.1098, "lr": 1.965459104040576e-05, "grad_norm": 7.9375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1205, "epoch": 1, "loss": 0.0005, "lr": 1.965381013767375e-05, "grad_norm": 2.328125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1206, "epoch": 1, "loss": 0.0145, "lr": 1.9653028368751795e-05, "grad_norm": 8.0, "samples_per_sec": 10.31, "gpu_mem_gb": 6.83} {"step": 1207, "epoch": 1, "loss": 0.098, "lr": 1.9652245733710037e-05, "grad_norm": 7.46875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1208, "epoch": 1, "loss": 0.0426, "lr": 1.9651462232618703e-05, "grad_norm": 6.0, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1209, "epoch": 1, "loss": 0.0063, "lr": 1.965067786554809e-05, "grad_norm": 2.09375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1210, "epoch": 1, "loss": 0.0128, "lr": 1.9649892632568573e-05, "grad_norm": 5.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1211, "epoch": 1, "loss": 0.1015, "lr": 1.9649106533750613e-05, "grad_norm": 5.625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1212, "epoch": 1, "loss": 0.1539, "lr": 1.964831956916474e-05, "grad_norm": 3.890625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.85} {"step": 1213, "epoch": 1, "loss": 0.238, "lr": 1.964753173888156e-05, "grad_norm": 5.71875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1214, "epoch": 1, "loss": 0.0074, "lr": 1.964674304297177e-05, "grad_norm": 0.71484375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1215, "epoch": 1, "loss": 0.0268, "lr": 1.9645953481506127e-05, "grad_norm": 7.40625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 1216, "epoch": 1, "loss": 0.1067, "lr": 1.9645163054555483e-05, "grad_norm": 6.96875, "samples_per_sec": 10.43, "gpu_mem_gb": 6.85} {"step": 1217, "epoch": 1, "loss": 0.067, "lr": 1.964437176219075e-05, "grad_norm": 5.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1218, "epoch": 1, "loss": 0.0686, "lr": 1.9643579604482937e-05, "grad_norm": 3.328125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1219, "epoch": 1, "loss": 0.3337, "lr": 1.9642786581503114e-05, "grad_norm": 9.9375, "samples_per_sec": 10.32, "gpu_mem_gb": 6.83} {"step": 1220, "epoch": 1, "loss": 0.03, "lr": 1.9641992693322438e-05, "grad_norm": 3.8125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 1221, "epoch": 1, "loss": 0.0032, "lr": 1.9641197940012136e-05, "grad_norm": 3.0625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1222, "epoch": 1, "loss": 0.0823, "lr": 1.9640402321643524e-05, "grad_norm": 6.96875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1223, "epoch": 1, "loss": 0.0005, "lr": 1.9639605838287984e-05, "grad_norm": 6.8125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 1224, "epoch": 1, "loss": 0.0026, "lr": 1.9638808490016984e-05, "grad_norm": 4.375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1225, "epoch": 1, "loss": 0.0877, "lr": 1.9638010276902065e-05, "grad_norm": 2.71875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1226, "epoch": 1, "loss": 0.1444, "lr": 1.9637211199014842e-05, "grad_norm": 6.9375, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 1227, "epoch": 1, "loss": 0.0616, "lr": 1.963641125642702e-05, "grad_norm": 3.40625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1228, "epoch": 1, "loss": 0.0003, "lr": 1.963561044921037e-05, "grad_norm": 3.359375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1229, "epoch": 1, "loss": 0.0012, "lr": 1.963480877743675e-05, "grad_norm": 6.71875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 1230, "epoch": 1, "loss": 0.0048, "lr": 1.9634006241178078e-05, "grad_norm": 1.125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1231, "epoch": 1, "loss": 0.08, "lr": 1.9633202840506375e-05, "grad_norm": 3.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1232, "epoch": 1, "loss": 0.0377, "lr": 1.9632398575493717e-05, "grad_norm": 7.34375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1233, "epoch": 1, "loss": 0.133, "lr": 1.963159344621227e-05, "grad_norm": 3.765625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1234, "epoch": 1, "loss": 0.044, "lr": 1.963078745273428e-05, "grad_norm": 5.40625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1235, "epoch": 1, "loss": 0.0079, "lr": 1.9629980595132056e-05, "grad_norm": 3.71875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1236, "epoch": 1, "loss": 0.1322, "lr": 1.9629172873477995e-05, "grad_norm": 11.25, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1237, "epoch": 1, "loss": 0.0145, "lr": 1.9628364287844572e-05, "grad_norm": 2.65625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 1238, "epoch": 1, "loss": 0.0314, "lr": 1.9627554838304343e-05, "grad_norm": 7.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1239, "epoch": 1, "loss": 0.0023, "lr": 1.962674452492993e-05, "grad_norm": 2.671875, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 1240, "epoch": 1, "loss": 0.0002, "lr": 1.9625933347794034e-05, "grad_norm": 2.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1241, "epoch": 1, "loss": 0.0316, "lr": 1.9625121306969447e-05, "grad_norm": 5.09375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1242, "epoch": 1, "loss": 0.2339, "lr": 1.9624308402529023e-05, "grad_norm": 7.71875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1243, "epoch": 1, "loss": 0.0007, "lr": 1.9623494634545706e-05, "grad_norm": 5.53125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1244, "epoch": 1, "loss": 0.0996, "lr": 1.9622680003092503e-05, "grad_norm": 4.15625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1245, "epoch": 1, "loss": 0.0382, "lr": 1.9621864508242517e-05, "grad_norm": 2.453125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1246, "epoch": 1, "loss": 0.0427, "lr": 1.9621048150068914e-05, "grad_norm": 4.71875, "samples_per_sec": 10.21, "gpu_mem_gb": 6.85} {"step": 1247, "epoch": 1, "loss": 0.0492, "lr": 1.962023092864494e-05, "grad_norm": 4.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1248, "epoch": 1, "loss": 0.0006, "lr": 1.961941284404392e-05, "grad_norm": 2.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1249, "epoch": 1, "loss": 0.2503, "lr": 1.9618593896339263e-05, "grad_norm": 3.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.84} {"step": 1250, "epoch": 1, "loss": 0.0054, "lr": 1.961777408560444e-05, "grad_norm": 2.03125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1251, "epoch": 1, "loss": 0.214, "lr": 1.9616953411913017e-05, "grad_norm": 6.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1252, "epoch": 1, "loss": 0.0069, "lr": 1.9616131875338625e-05, "grad_norm": 1.8515625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1253, "epoch": 1, "loss": 0.0002, "lr": 1.9615309475954975e-05, "grad_norm": 3.328125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1254, "epoch": 1, "loss": 0.0261, "lr": 1.961448621383586e-05, "grad_norm": 2.828125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1255, "epoch": 1, "loss": 0.1937, "lr": 1.9613662089055148e-05, "grad_norm": 3.515625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1256, "epoch": 1, "loss": 0.0108, "lr": 1.961283710168678e-05, "grad_norm": 8.0, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1257, "epoch": 1, "loss": 0.0102, "lr": 1.9612011251804784e-05, "grad_norm": 4.4375, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 1258, "epoch": 1, "loss": 0.0007, "lr": 1.9611184539483252e-05, "grad_norm": 0.2197265625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 1259, "epoch": 1, "loss": 0.1898, "lr": 1.9610356964796366e-05, "grad_norm": 7.6875, "samples_per_sec": 10.37, "gpu_mem_gb": 6.83} {"step": 1260, "epoch": 1, "loss": 0.0445, "lr": 1.960952852781838e-05, "grad_norm": 6.15625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1261, "epoch": 1, "loss": 0.0474, "lr": 1.9608699228623623e-05, "grad_norm": 5.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1262, "epoch": 1, "loss": 0.0004, "lr": 1.9607869067286506e-05, "grad_norm": 1.453125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1263, "epoch": 1, "loss": 0.0117, "lr": 1.960703804388152e-05, "grad_norm": 2.59375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1264, "epoch": 1, "loss": 0.0007, "lr": 1.9606206158483217e-05, "grad_norm": 4.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1265, "epoch": 1, "loss": 0.1679, "lr": 1.960537341116625e-05, "grad_norm": 4.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1266, "epoch": 1, "loss": 0.0228, "lr": 1.960453980200533e-05, "grad_norm": 3.25, "samples_per_sec": 10.21, "gpu_mem_gb": 6.83} {"step": 1267, "epoch": 1, "loss": 0.0977, "lr": 1.9603705331075255e-05, "grad_norm": 9.9375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1268, "epoch": 1, "loss": 0.0963, "lr": 1.9602869998450898e-05, "grad_norm": 3.546875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 1269, "epoch": 1, "loss": 0.0023, "lr": 1.9602033804207212e-05, "grad_norm": 3.703125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1270, "epoch": 1, "loss": 0.0013, "lr": 1.960119674841922e-05, "grad_norm": 5.9375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1271, "epoch": 1, "loss": 0.0025, "lr": 1.960035883116203e-05, "grad_norm": 7.40625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1272, "epoch": 1, "loss": 0.2069, "lr": 1.9599520052510824e-05, "grad_norm": 6.1875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 1273, "epoch": 1, "loss": 0.054, "lr": 1.959868041254086e-05, "grad_norm": 3.25, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83} {"step": 1274, "epoch": 1, "loss": 0.0783, "lr": 1.9597839911327475e-05, "grad_norm": 3.78125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 1275, "epoch": 1, "loss": 0.0043, "lr": 1.959699854894609e-05, "grad_norm": 6.28125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1276, "epoch": 1, "loss": 0.1524, "lr": 1.9596156325472183e-05, "grad_norm": 3.78125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1277, "epoch": 1, "loss": 0.0004, "lr": 1.9595313240981332e-05, "grad_norm": 2.09375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1278, "epoch": 1, "loss": 0.0277, "lr": 1.9594469295549184e-05, "grad_norm": 2.328125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1279, "epoch": 1, "loss": 0.227, "lr": 1.959362448925146e-05, "grad_norm": 4.9375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 1280, "epoch": 1, "loss": 0.0534, "lr": 1.9592778822163955e-05, "grad_norm": 3.15625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 1281, "epoch": 1, "loss": 0.0835, "lr": 1.9591932294362552e-05, "grad_norm": 2.328125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1282, "epoch": 1, "loss": 0.0065, "lr": 1.9591084905923207e-05, "grad_norm": 7.15625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1283, "epoch": 1, "loss": 0.0532, "lr": 1.959023665692195e-05, "grad_norm": 1.7734375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 1284, "epoch": 1, "loss": 0.2143, "lr": 1.958938754743489e-05, "grad_norm": 4.59375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1285, "epoch": 1, "loss": 0.2526, "lr": 1.9588537577538216e-05, "grad_norm": 18.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 1286, "epoch": 1, "loss": 0.016, "lr": 1.9587686747308184e-05, "grad_norm": 3.34375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 1287, "epoch": 1, "loss": 0.0674, "lr": 1.9586835056821148e-05, "grad_norm": 3.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1288, "epoch": 1, "loss": 0.2331, "lr": 1.9585982506153515e-05, "grad_norm": 7.03125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1289, "epoch": 1, "loss": 0.0228, "lr": 1.9585129095381784e-05, "grad_norm": 8.0625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1290, "epoch": 1, "loss": 0.0104, "lr": 1.958427482458253e-05, "grad_norm": 2.671875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1291, "epoch": 1, "loss": 0.0629, "lr": 1.9583419693832397e-05, "grad_norm": 3.078125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1292, "epoch": 1, "loss": 0.0451, "lr": 1.9582563703208115e-05, "grad_norm": 7.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1293, "epoch": 1, "loss": 0.0003, "lr": 1.9581706852786492e-05, "grad_norm": 2.234375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 1294, "epoch": 1, "loss": 0.0004, "lr": 1.95808491426444e-05, "grad_norm": 6.5625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1295, "epoch": 1, "loss": 0.0575, "lr": 1.9579990572858802e-05, "grad_norm": 5.28125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 1296, "epoch": 1, "loss": 0.1537, "lr": 1.9579131143506736e-05, "grad_norm": 5.03125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1297, "epoch": 1, "loss": 0.0983, "lr": 1.957827085466531e-05, "grad_norm": 7.9375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.84} {"step": 1298, "epoch": 1, "loss": 0.0731, "lr": 1.9577409706411717e-05, "grad_norm": 4.28125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 1299, "epoch": 1, "loss": 0.0571, "lr": 1.9576547698823222e-05, "grad_norm": 4.53125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1300, "epoch": 1, "loss": 0.1238, "lr": 1.957568483197717e-05, "grad_norm": 7.28125, "samples_per_sec": 10.37, "gpu_mem_gb": 6.83} {"step": 1301, "epoch": 1, "loss": 0.0049, "lr": 1.9574821105950976e-05, "grad_norm": 5.125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1302, "epoch": 1, "loss": 0.1755, "lr": 1.9573956520822147e-05, "grad_norm": 4.59375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1303, "epoch": 1, "loss": 0.0232, "lr": 1.9573091076668255e-05, "grad_norm": 3.203125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 1304, "epoch": 1, "loss": 0.0003, "lr": 1.9572224773566945e-05, "grad_norm": 5.71875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 1305, "epoch": 1, "loss": 0.0148, "lr": 1.9571357611595955e-05, "grad_norm": 5.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1306, "epoch": 1, "loss": 0.3209, "lr": 1.957048959083309e-05, "grad_norm": 6.4375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 1307, "epoch": 1, "loss": 0.0003, "lr": 1.956962071135623e-05, "grad_norm": 0.6953125, "samples_per_sec": 10.32, "gpu_mem_gb": 6.82} {"step": 1308, "epoch": 1, "loss": 0.0088, "lr": 1.956875097324334e-05, "grad_norm": 2.21875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1309, "epoch": 1, "loss": 0.0433, "lr": 1.9567880376572456e-05, "grad_norm": 4.28125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 1310, "epoch": 1, "loss": 0.0004, "lr": 1.9567008921421684e-05, "grad_norm": 0.263671875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 1311, "epoch": 1, "loss": 0.1622, "lr": 1.9566136607869224e-05, "grad_norm": 5.28125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 1312, "epoch": 1, "loss": 0.0052, "lr": 1.956526343599335e-05, "grad_norm": 2.796875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 1313, "epoch": 1, "loss": 0.0059, "lr": 1.9564389405872396e-05, "grad_norm": 3.703125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 1314, "epoch": 1, "loss": 0.0003, "lr": 1.9563514517584786e-05, "grad_norm": 5.90625, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 1315, "epoch": 1, "loss": 0.0526, "lr": 1.9562638771209026e-05, "grad_norm": 4.875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1316, "epoch": 1, "loss": 0.0001, "lr": 1.956176216682369e-05, "grad_norm": 7.71875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1317, "epoch": 1, "loss": 0.0005, "lr": 1.9560884704507427e-05, "grad_norm": 3.375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1318, "epoch": 1, "loss": 0.2873, "lr": 1.9560006384338973e-05, "grad_norm": 9.625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 1319, "epoch": 1, "loss": 0.0007, "lr": 1.9559127206397136e-05, "grad_norm": 9.125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1320, "epoch": 1, "loss": 0.1464, "lr": 1.9558247170760796e-05, "grad_norm": 3.515625, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 1321, "epoch": 1, "loss": 0.0522, "lr": 1.9557366277508916e-05, "grad_norm": 2.953125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1322, "epoch": 1, "loss": 0.3166, "lr": 1.9556484526720534e-05, "grad_norm": 8.0, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1323, "epoch": 1, "loss": 0.1007, "lr": 1.9555601918474764e-05, "grad_norm": 8.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1324, "epoch": 1, "loss": 0.0424, "lr": 1.9554718452850804e-05, "grad_norm": 3.65625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 1325, "epoch": 1, "loss": 0.0847, "lr": 1.9553834129927917e-05, "grad_norm": 4.71875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1326, "epoch": 1, "loss": 0.0476, "lr": 1.9552948949785452e-05, "grad_norm": 2.84375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1327, "epoch": 1, "loss": 0.5266, "lr": 1.9552062912502832e-05, "grad_norm": 10.125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 1328, "epoch": 1, "loss": 0.0038, "lr": 1.9551176018159554e-05, "grad_norm": 7.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1329, "epoch": 1, "loss": 0.0452, "lr": 1.9550288266835194e-05, "grad_norm": 7.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1330, "epoch": 1, "loss": 0.1459, "lr": 1.9549399658609415e-05, "grad_norm": 10.375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1331, "epoch": 1, "loss": 0.0224, "lr": 1.9548510193561938e-05, "grad_norm": 5.21875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1332, "epoch": 1, "loss": 0.1503, "lr": 1.9547619871772575e-05, "grad_norm": 3.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1333, "epoch": 1, "loss": 0.2741, "lr": 1.9546728693321204e-05, "grad_norm": 6.375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 1334, "epoch": 1, "loss": 0.0131, "lr": 1.9545836658287793e-05, "grad_norm": 3.421875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1335, "epoch": 1, "loss": 0.0013, "lr": 1.954494376675238e-05, "grad_norm": 5.40625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1336, "epoch": 1, "loss": 0.0024, "lr": 1.9544050018795076e-05, "grad_norm": 4.28125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1337, "epoch": 1, "loss": 0.001, "lr": 1.9543155414496074e-05, "grad_norm": 3.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1338, "epoch": 1, "loss": 0.0015, "lr": 1.9542259953935643e-05, "grad_norm": 2.8125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 1339, "epoch": 1, "loss": 0.1009, "lr": 1.9541363637194132e-05, "grad_norm": 6.9375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1340, "epoch": 1, "loss": 0.0046, "lr": 1.9540466464351956e-05, "grad_norm": 4.53125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1341, "epoch": 1, "loss": 0.1409, "lr": 1.9539568435489617e-05, "grad_norm": 2.78125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 1342, "epoch": 1, "loss": 0.2289, "lr": 1.953866955068769e-05, "grad_norm": 5.65625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1343, "epoch": 1, "loss": 0.2046, "lr": 1.9537769810026833e-05, "grad_norm": 6.3125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1344, "epoch": 1, "loss": 0.004, "lr": 1.9536869213587768e-05, "grad_norm": 4.59375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1345, "epoch": 1, "loss": 0.0062, "lr": 1.9535967761451307e-05, "grad_norm": 11.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 1346, "epoch": 1, "loss": 0.0078, "lr": 1.953506545369833e-05, "grad_norm": 6.71875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1347, "epoch": 1, "loss": 0.0111, "lr": 1.9534162290409796e-05, "grad_norm": 5.28125, "samples_per_sec": 10.29, "gpu_mem_gb": 6.83} {"step": 1348, "epoch": 1, "loss": 0.0069, "lr": 1.9533258271666743e-05, "grad_norm": 2.09375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 1349, "epoch": 1, "loss": 0.091, "lr": 1.9532353397550287e-05, "grad_norm": 6.78125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1350, "epoch": 1, "loss": 0.0015, "lr": 1.953144766814161e-05, "grad_norm": 6.84375, "samples_per_sec": 10.26, "gpu_mem_gb": 6.82} {"step": 1351, "epoch": 1, "loss": 0.2979, "lr": 1.9530541083521986e-05, "grad_norm": 7.09375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 1352, "epoch": 1, "loss": 0.0419, "lr": 1.9529633643772755e-05, "grad_norm": 3.9375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1353, "epoch": 1, "loss": 0.0009, "lr": 1.952872534897534e-05, "grad_norm": 5.15625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1354, "epoch": 1, "loss": 0.1984, "lr": 1.9527816199211236e-05, "grad_norm": 7.34375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 1355, "epoch": 1, "loss": 0.1713, "lr": 1.9526906194562015e-05, "grad_norm": 5.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1356, "epoch": 1, "loss": 0.0126, "lr": 1.9525995335109333e-05, "grad_norm": 4.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1357, "epoch": 1, "loss": 0.0006, "lr": 1.9525083620934914e-05, "grad_norm": 3.84375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 1358, "epoch": 1, "loss": 0.0061, "lr": 1.952417105212056e-05, "grad_norm": 0.66796875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1359, "epoch": 1, "loss": 0.0662, "lr": 1.9523257628748148e-05, "grad_norm": 4.125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1360, "epoch": 1, "loss": 0.0097, "lr": 1.9522343350899643e-05, "grad_norm": 3.0, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 1361, "epoch": 1, "loss": 0.0594, "lr": 1.9521428218657075e-05, "grad_norm": 3.09375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 1362, "epoch": 1, "loss": 0.0755, "lr": 1.9520512232102552e-05, "grad_norm": 5.875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 1363, "epoch": 1, "loss": 0.0247, "lr": 1.9519595391318266e-05, "grad_norm": 8.4375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1364, "epoch": 1, "loss": 0.008, "lr": 1.951867769638648e-05, "grad_norm": 4.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 1365, "epoch": 1, "loss": 0.1242, "lr": 1.9517759147389533e-05, "grad_norm": 9.3125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1366, "epoch": 1, "loss": 0.2044, "lr": 1.951683974440984e-05, "grad_norm": 5.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 1367, "epoch": 1, "loss": 0.0233, "lr": 1.9515919487529893e-05, "grad_norm": 4.03125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1368, "epoch": 1, "loss": 0.3045, "lr": 1.951499837683227e-05, "grad_norm": 8.4375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1369, "epoch": 1, "loss": 0.0681, "lr": 1.9514076412399615e-05, "grad_norm": 7.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1370, "epoch": 1, "loss": 0.0009, "lr": 1.951315359431465e-05, "grad_norm": 16.0, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1371, "epoch": 1, "loss": 0.0107, "lr": 1.951222992266017e-05, "grad_norm": 2.515625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1372, "epoch": 1, "loss": 0.0171, "lr": 1.951130539751906e-05, "grad_norm": 6.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1373, "epoch": 1, "loss": 0.0377, "lr": 1.9510380018974267e-05, "grad_norm": 5.625, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 1374, "epoch": 1, "loss": 0.0005, "lr": 1.9509453787108827e-05, "grad_norm": 2.5, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 1375, "epoch": 1, "loss": 0.0259, "lr": 1.9508526702005842e-05, "grad_norm": 4.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 1376, "epoch": 1, "loss": 0.0553, "lr": 1.9507598763748494e-05, "grad_norm": 4.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1377, "epoch": 1, "loss": 0.0054, "lr": 1.9506669972420047e-05, "grad_norm": 2.03125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 1378, "epoch": 1, "loss": 0.0033, "lr": 1.950574032810383e-05, "grad_norm": 7.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1379, "epoch": 1, "loss": 0.0363, "lr": 1.9504809830883266e-05, "grad_norm": 7.625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1380, "epoch": 1, "loss": 0.1696, "lr": 1.9503878480841832e-05, "grad_norm": 6.9375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1381, "epoch": 1, "loss": 0.1857, "lr": 1.95029462780631e-05, "grad_norm": 5.59375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 1382, "epoch": 1, "loss": 0.0194, "lr": 1.9502013222630714e-05, "grad_norm": 2.609375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1383, "epoch": 1, "loss": 0.0031, "lr": 1.950107931462839e-05, "grad_norm": 3.21875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1384, "epoch": 1, "loss": 0.0175, "lr": 1.950014455413992e-05, "grad_norm": 1.828125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 1385, "epoch": 1, "loss": 0.0186, "lr": 1.949920894124918e-05, "grad_norm": 2.078125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1386, "epoch": 1, "loss": 0.0757, "lr": 1.9498272476040115e-05, "grad_norm": 6.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1387, "epoch": 1, "loss": 0.0449, "lr": 1.9497335158596755e-05, "grad_norm": 7.03125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 1388, "epoch": 1, "loss": 0.0856, "lr": 1.9496396989003195e-05, "grad_norm": 5.09375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 1389, "epoch": 1, "loss": 0.0348, "lr": 1.9495457967343615e-05, "grad_norm": 2.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1390, "epoch": 1, "loss": 0.1268, "lr": 1.9494518093702268e-05, "grad_norm": 9.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 1391, "epoch": 1, "loss": 0.027, "lr": 1.9493577368163483e-05, "grad_norm": 5.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1392, "epoch": 1, "loss": 0.2495, "lr": 1.949263579081167e-05, "grad_norm": 6.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.84} {"step": 1393, "epoch": 1, "loss": 0.0181, "lr": 1.949169336173131e-05, "grad_norm": 2.46875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 1394, "epoch": 1, "loss": 0.0301, "lr": 1.9490750081006965e-05, "grad_norm": 3.390625, "samples_per_sec": 10.27, "gpu_mem_gb": 6.83} {"step": 1395, "epoch": 1, "loss": 0.0007, "lr": 1.948980594872327e-05, "grad_norm": 9.875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1396, "epoch": 1, "loss": 0.0002, "lr": 1.9488860964964934e-05, "grad_norm": 6.5625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1397, "epoch": 1, "loss": 0.0237, "lr": 1.948791512981675e-05, "grad_norm": 2.234375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1398, "epoch": 1, "loss": 0.0028, "lr": 1.9486968443363585e-05, "grad_norm": 3.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1399, "epoch": 1, "loss": 0.0018, "lr": 1.9486020905690373e-05, "grad_norm": 0.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 1400, "epoch": 1, "loss": 0.0478, "lr": 1.9485072516882138e-05, "grad_norm": 7.28125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1401, "epoch": 1, "loss": 0.0833, "lr": 1.9484123277023974e-05, "grad_norm": 6.6875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.84} {"step": 1402, "epoch": 1, "loss": 0.0169, "lr": 1.948317318620105e-05, "grad_norm": 6.375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1403, "epoch": 1, "loss": 0.116, "lr": 1.9482222244498613e-05, "grad_norm": 4.15625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1404, "epoch": 1, "loss": 0.117, "lr": 1.948127045200199e-05, "grad_norm": 8.5, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1405, "epoch": 1, "loss": 0.004, "lr": 1.9480317808796573e-05, "grad_norm": 5.15625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 1406, "epoch": 1, "loss": 0.004, "lr": 1.9479364314967846e-05, "grad_norm": 3.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1407, "epoch": 1, "loss": 0.0019, "lr": 1.947840997060136e-05, "grad_norm": 0.33203125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1408, "epoch": 1, "loss": 0.0086, "lr": 1.947745477578274e-05, "grad_norm": 8.3125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1409, "epoch": 1, "loss": 0.0009, "lr": 1.947649873059769e-05, "grad_norm": 4.125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 1410, "epoch": 1, "loss": 0.0417, "lr": 1.9475541835132e-05, "grad_norm": 3.0625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 1411, "epoch": 1, "loss": 0.016, "lr": 1.9474584089471522e-05, "grad_norm": 6.4375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1412, "epoch": 1, "loss": 0.0001, "lr": 1.9473625493702187e-05, "grad_norm": 5.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1413, "epoch": 1, "loss": 0.103, "lr": 1.947266604791001e-05, "grad_norm": 4.78125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1414, "epoch": 1, "loss": 0.0254, "lr": 1.947170575218108e-05, "grad_norm": 2.703125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 1415, "epoch": 1, "loss": 0.0029, "lr": 1.9470744606601548e-05, "grad_norm": 7.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1416, "epoch": 1, "loss": 0.2818, "lr": 1.9469782611257662e-05, "grad_norm": 7.3125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1417, "epoch": 1, "loss": 0.0232, "lr": 1.946881976623574e-05, "grad_norm": 4.96875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1418, "epoch": 1, "loss": 0.0008, "lr": 1.9467856071622165e-05, "grad_norm": 7.8125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1419, "epoch": 1, "loss": 0.0411, "lr": 1.9466891527503412e-05, "grad_norm": 4.03125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 1420, "epoch": 1, "loss": 0.0038, "lr": 1.9465926133966017e-05, "grad_norm": 2.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1421, "epoch": 1, "loss": 0.0036, "lr": 1.9464959891096606e-05, "grad_norm": 5.34375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 1422, "epoch": 1, "loss": 0.3801, "lr": 1.9463992798981876e-05, "grad_norm": 5.0, "samples_per_sec": 8.99, "gpu_mem_gb": 6.83} {"step": 1423, "epoch": 1, "loss": 0.0015, "lr": 1.94630248577086e-05, "grad_norm": 4.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1424, "epoch": 1, "loss": 0.0998, "lr": 1.946205606736362e-05, "grad_norm": 6.71875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 1425, "epoch": 1, "loss": 0.412, "lr": 1.9461086428033864e-05, "grad_norm": 5.375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1426, "epoch": 1, "loss": 0.2943, "lr": 1.946011593980634e-05, "grad_norm": 11.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1427, "epoch": 1, "loss": 0.0005, "lr": 1.9459144602768115e-05, "grad_norm": 13.5, "samples_per_sec": 10.31, "gpu_mem_gb": 6.82} {"step": 1428, "epoch": 1, "loss": 0.0569, "lr": 1.9458172417006347e-05, "grad_norm": 3.609375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 1429, "epoch": 1, "loss": 0.0933, "lr": 1.945719938260827e-05, "grad_norm": 6.0, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1430, "epoch": 1, "loss": 0.0017, "lr": 1.945622549966118e-05, "grad_norm": 8.1875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 1431, "epoch": 1, "loss": 0.0041, "lr": 1.9455250768252466e-05, "grad_norm": 4.5625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 1432, "epoch": 1, "loss": 0.0632, "lr": 1.9454275188469584e-05, "grad_norm": 5.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1433, "epoch": 1, "loss": 0.0013, "lr": 1.945329876040007e-05, "grad_norm": 1.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1434, "epoch": 1, "loss": 0.0028, "lr": 1.9452321484131528e-05, "grad_norm": 5.65625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1435, "epoch": 1, "loss": 0.0144, "lr": 1.9451343359751653e-05, "grad_norm": 2.90625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1436, "epoch": 1, "loss": 0.0388, "lr": 1.94503643873482e-05, "grad_norm": 3.515625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1437, "epoch": 1, "loss": 0.3187, "lr": 1.944938456700901e-05, "grad_norm": 6.15625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1438, "epoch": 1, "loss": 0.1149, "lr": 1.9448403898822002e-05, "grad_norm": 4.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 1439, "epoch": 1, "loss": 0.0029, "lr": 1.9447422382875158e-05, "grad_norm": 2.453125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1440, "epoch": 1, "loss": 0.332, "lr": 1.9446440019256556e-05, "grad_norm": 9.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1441, "epoch": 1, "loss": 0.004, "lr": 1.9445456808054322e-05, "grad_norm": 8.25, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 1442, "epoch": 1, "loss": 0.0096, "lr": 1.9444472749356692e-05, "grad_norm": 3.09375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1443, "epoch": 1, "loss": 0.0117, "lr": 1.9443487843251954e-05, "grad_norm": 8.75, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1444, "epoch": 1, "loss": 0.0005, "lr": 1.944250208982848e-05, "grad_norm": 3.15625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 1445, "epoch": 1, "loss": 0.0017, "lr": 1.9441515489174708e-05, "grad_norm": 3.453125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 1446, "epoch": 1, "loss": 0.002, "lr": 1.9440528041379173e-05, "grad_norm": 1.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 1447, "epoch": 1, "loss": 0.0139, "lr": 1.943953974653047e-05, "grad_norm": 1.0703125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1448, "epoch": 1, "loss": 0.159, "lr": 1.9438550604717276e-05, "grad_norm": 5.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 1449, "epoch": 1, "loss": 0.1985, "lr": 1.9437560616028335e-05, "grad_norm": 4.28125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1450, "epoch": 1, "loss": 0.0108, "lr": 1.943656978055248e-05, "grad_norm": 12.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1451, "epoch": 1, "loss": 0.0005, "lr": 1.9435578098378612e-05, "grad_norm": 3.828125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 1452, "epoch": 1, "loss": 0.1229, "lr": 1.943458556959571e-05, "grad_norm": 4.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1453, "epoch": 1, "loss": 0.0011, "lr": 1.943359219429283e-05, "grad_norm": 9.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1454, "epoch": 1, "loss": 0.0044, "lr": 1.9432597972559096e-05, "grad_norm": 0.85546875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1455, "epoch": 1, "loss": 0.0028, "lr": 1.9431602904483727e-05, "grad_norm": 2.484375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1456, "epoch": 1, "loss": 0.1109, "lr": 1.9430606990156e-05, "grad_norm": 3.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.84} {"step": 1457, "epoch": 1, "loss": 0.26, "lr": 1.9429610229665264e-05, "grad_norm": 5.90625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1458, "epoch": 1, "loss": 0.0035, "lr": 1.9428612623100972e-05, "grad_norm": 6.34375, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 1459, "epoch": 1, "loss": 0.046, "lr": 1.9427614170552617e-05, "grad_norm": 8.5625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.84} {"step": 1460, "epoch": 1, "loss": 0.0028, "lr": 1.9426614872109798e-05, "grad_norm": 4.09375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1461, "epoch": 1, "loss": 0.0049, "lr": 1.942561472786217e-05, "grad_norm": 8.5625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 1462, "epoch": 1, "loss": 0.3086, "lr": 1.9424613737899473e-05, "grad_norm": 5.21875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1463, "epoch": 1, "loss": 0.1155, "lr": 1.9423611902311522e-05, "grad_norm": 8.5625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 1464, "epoch": 1, "loss": 0.007, "lr": 1.9422609221188208e-05, "grad_norm": 2.515625, "samples_per_sec": 10.53, "gpu_mem_gb": 6.85} {"step": 1465, "epoch": 1, "loss": 0.141, "lr": 1.9421605694619493e-05, "grad_norm": 4.65625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1466, "epoch": 1, "loss": 0.0015, "lr": 1.9420601322695417e-05, "grad_norm": 5.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1467, "epoch": 1, "loss": 0.0962, "lr": 1.9419596105506107e-05, "grad_norm": 5.8125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1468, "epoch": 1, "loss": 0.0006, "lr": 1.941859004314175e-05, "grad_norm": 2.703125, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 1469, "epoch": 1, "loss": 0.094, "lr": 1.9417583135692613e-05, "grad_norm": 4.8125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1470, "epoch": 1, "loss": 0.0063, "lr": 1.9416575383249047e-05, "grad_norm": 4.40625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1471, "epoch": 1, "loss": 0.1407, "lr": 1.9415566785901463e-05, "grad_norm": 5.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1472, "epoch": 1, "loss": 0.4718, "lr": 1.9414557343740367e-05, "grad_norm": 6.46875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1473, "epoch": 1, "loss": 0.1021, "lr": 1.941354705685633e-05, "grad_norm": 2.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1474, "epoch": 1, "loss": 0.0023, "lr": 1.9412535925339998e-05, "grad_norm": 4.9375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 1475, "epoch": 1, "loss": 0.103, "lr": 1.9411523949282094e-05, "grad_norm": 4.8125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1476, "epoch": 1, "loss": 0.0005, "lr": 1.941051112877342e-05, "grad_norm": 1.375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1477, "epoch": 1, "loss": 0.0337, "lr": 1.940949746390485e-05, "grad_norm": 12.875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 1478, "epoch": 1, "loss": 0.0402, "lr": 1.9408482954767335e-05, "grad_norm": 4.0625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 1479, "epoch": 1, "loss": 0.0072, "lr": 1.9407467601451902e-05, "grad_norm": 3.625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1480, "epoch": 1, "loss": 0.0335, "lr": 1.940645140404966e-05, "grad_norm": 4.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1481, "epoch": 1, "loss": 0.0155, "lr": 1.9405434362651778e-05, "grad_norm": 3.828125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.85} {"step": 1482, "epoch": 1, "loss": 0.2837, "lr": 1.9404416477349514e-05, "grad_norm": 8.0625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1483, "epoch": 1, "loss": 0.2281, "lr": 1.94033977482342e-05, "grad_norm": 4.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 1484, "epoch": 1, "loss": 0.0148, "lr": 1.940237817539724e-05, "grad_norm": 4.96875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 1485, "epoch": 1, "loss": 0.0036, "lr": 1.9401357758930116e-05, "grad_norm": 1.0546875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1486, "epoch": 1, "loss": 0.0015, "lr": 1.9400336498924384e-05, "grad_norm": 4.09375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1487, "epoch": 1, "loss": 0.0011, "lr": 1.9399314395471678e-05, "grad_norm": 3.9375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1488, "epoch": 1, "loss": 0.1054, "lr": 1.9398291448663702e-05, "grad_norm": 6.90625, "samples_per_sec": 10.24, "gpu_mem_gb": 6.82} {"step": 1489, "epoch": 1, "loss": 0.0065, "lr": 1.9397267658592248e-05, "grad_norm": 7.03125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 1490, "epoch": 1, "loss": 0.0829, "lr": 1.9396243025349172e-05, "grad_norm": 2.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 1491, "epoch": 1, "loss": 0.0053, "lr": 1.9395217549026404e-05, "grad_norm": 1.9453125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1492, "epoch": 1, "loss": 0.1614, "lr": 1.9394191229715962e-05, "grad_norm": 7.09375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1493, "epoch": 1, "loss": 0.0057, "lr": 1.9393164067509933e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1494, "epoch": 1, "loss": 0.0005, "lr": 1.9392136062500473e-05, "grad_norm": 7.03125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 1495, "epoch": 1, "loss": 0.1666, "lr": 1.9391107214779828e-05, "grad_norm": 3.09375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.85} {"step": 1496, "epoch": 1, "loss": 0.072, "lr": 1.9390077524440307e-05, "grad_norm": 3.796875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1497, "epoch": 1, "loss": 0.0006, "lr": 1.9389046991574298e-05, "grad_norm": 5.8125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1498, "epoch": 1, "loss": 0.0008, "lr": 1.9388015616274266e-05, "grad_norm": 3.171875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 1499, "epoch": 1, "loss": 0.0009, "lr": 1.9386983398632756e-05, "grad_norm": 4.09375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 1500, "epoch": 1, "loss": 0.1445, "lr": 1.9385950338742377e-05, "grad_norm": 8.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1501, "epoch": 1, "loss": 0.0567, "lr": 1.9384916436695827e-05, "grad_norm": 4.28125, "samples_per_sec": 10.52, "gpu_mem_gb": 6.85} {"step": 1502, "epoch": 1, "loss": 0.0006, "lr": 1.938388169258587e-05, "grad_norm": 1.25, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 1503, "epoch": 1, "loss": 0.1375, "lr": 1.938284610650535e-05, "grad_norm": 4.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1504, "epoch": 1, "loss": 0.0008, "lr": 1.9381809678547184e-05, "grad_norm": 7.5625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1505, "epoch": 1, "loss": 0.1289, "lr": 1.938077240880436e-05, "grad_norm": 11.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1506, "epoch": 1, "loss": 0.0169, "lr": 1.9379734297369957e-05, "grad_norm": 1.6953125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1507, "epoch": 1, "loss": 0.0001, "lr": 1.937869534433712e-05, "grad_norm": 4.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1508, "epoch": 1, "loss": 0.0635, "lr": 1.937765554979906e-05, "grad_norm": 2.8125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1509, "epoch": 1, "loss": 0.0138, "lr": 1.9376614913849076e-05, "grad_norm": 3.984375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 1510, "epoch": 1, "loss": 0.0026, "lr": 1.9375573436580545e-05, "grad_norm": 6.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1511, "epoch": 1, "loss": 0.0875, "lr": 1.9374531118086907e-05, "grad_norm": 4.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1512, "epoch": 1, "loss": 0.0753, "lr": 1.9373487958461688e-05, "grad_norm": 4.90625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 1513, "epoch": 1, "loss": 0.1153, "lr": 1.9372443957798486e-05, "grad_norm": 5.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 1514, "epoch": 1, "loss": 0.1756, "lr": 1.9371399116190968e-05, "grad_norm": 4.28125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1515, "epoch": 1, "loss": 0.0003, "lr": 1.9370353433732892e-05, "grad_norm": 3.65625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 1516, "epoch": 1, "loss": 0.0042, "lr": 1.9369306910518078e-05, "grad_norm": 3.328125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1517, "epoch": 1, "loss": 0.3287, "lr": 1.936825954664042e-05, "grad_norm": 7.9375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1518, "epoch": 1, "loss": 0.0002, "lr": 1.9367211342193904e-05, "grad_norm": 0.60546875, "samples_per_sec": 10.29, "gpu_mem_gb": 6.83} {"step": 1519, "epoch": 1, "loss": 0.0937, "lr": 1.936616229727257e-05, "grad_norm": 2.796875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1520, "epoch": 1, "loss": 0.0756, "lr": 1.936511241197055e-05, "grad_norm": 6.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1521, "epoch": 1, "loss": 0.0013, "lr": 1.9364061686382042e-05, "grad_norm": 3.421875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1522, "epoch": 1, "loss": 0.0244, "lr": 1.9363010120601325e-05, "grad_norm": 5.53125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1523, "epoch": 1, "loss": 0.007, "lr": 1.9361957714722746e-05, "grad_norm": 3.640625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1524, "epoch": 1, "loss": 0.0753, "lr": 1.936090446884074e-05, "grad_norm": 1.859375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 1525, "epoch": 1, "loss": 0.0016, "lr": 1.9359850383049804e-05, "grad_norm": 4.34375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 1526, "epoch": 1, "loss": 0.055, "lr": 1.9358795457444515e-05, "grad_norm": 5.03125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 1527, "epoch": 1, "loss": 0.148, "lr": 1.935773969211953e-05, "grad_norm": 4.625, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 1528, "epoch": 1, "loss": 0.0133, "lr": 1.935668308716958e-05, "grad_norm": 5.0625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 1529, "epoch": 1, "loss": 0.0107, "lr": 1.9355625642689467e-05, "grad_norm": 3.6875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1530, "epoch": 1, "loss": 0.1233, "lr": 1.9354567358774065e-05, "grad_norm": 4.71875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1531, "epoch": 1, "loss": 0.0362, "lr": 1.9353508235518337e-05, "grad_norm": 4.46875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1532, "epoch": 1, "loss": 0.0312, "lr": 1.9352448273017302e-05, "grad_norm": 5.25, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 1533, "epoch": 1, "loss": 0.0025, "lr": 1.935138747136608e-05, "grad_norm": 4.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1534, "epoch": 1, "loss": 0.0041, "lr": 1.9350325830659843e-05, "grad_norm": 1.1953125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1535, "epoch": 1, "loss": 0.047, "lr": 1.9349263350993845e-05, "grad_norm": 4.3125, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 1536, "epoch": 1, "loss": 0.006, "lr": 1.934820003246342e-05, "grad_norm": 3.875, "samples_per_sec": 10.43, "gpu_mem_gb": 6.82} {"step": 1537, "epoch": 1, "loss": 0.0056, "lr": 1.9347135875163974e-05, "grad_norm": 2.90625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1538, "epoch": 1, "loss": 0.2775, "lr": 1.9346070879190992e-05, "grad_norm": 7.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1539, "epoch": 1, "loss": 0.4568, "lr": 1.9345005044640028e-05, "grad_norm": 6.125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1540, "epoch": 1, "loss": 0.009, "lr": 1.9343938371606714e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1541, "epoch": 1, "loss": 0.002, "lr": 1.9342870860186758e-05, "grad_norm": 2.5, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 1542, "epoch": 1, "loss": 0.0108, "lr": 1.934180251047594e-05, "grad_norm": 1.9375, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 1543, "epoch": 1, "loss": 0.2859, "lr": 1.9340733322570122e-05, "grad_norm": 5.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1544, "epoch": 1, "loss": 0.0469, "lr": 1.9339663296565235e-05, "grad_norm": 6.53125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1545, "epoch": 1, "loss": 0.0508, "lr": 1.933859243255729e-05, "grad_norm": 5.125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1546, "epoch": 1, "loss": 0.1152, "lr": 1.9337520730642367e-05, "grad_norm": 4.4375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1547, "epoch": 1, "loss": 0.0216, "lr": 1.9336448190916627e-05, "grad_norm": 3.25, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 1548, "epoch": 1, "loss": 0.0697, "lr": 1.93353748134763e-05, "grad_norm": 4.875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 1549, "epoch": 1, "loss": 0.1157, "lr": 1.93343005984177e-05, "grad_norm": 4.5, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 1550, "epoch": 1, "loss": 0.0705, "lr": 1.9333225545837208e-05, "grad_norm": 5.625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1551, "epoch": 1, "loss": 0.0608, "lr": 1.9332149655831287e-05, "grad_norm": 7.5, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1552, "epoch": 1, "loss": 0.0003, "lr": 1.9331072928496465e-05, "grad_norm": 3.203125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1553, "epoch": 1, "loss": 0.0015, "lr": 1.932999536392936e-05, "grad_norm": 3.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1554, "epoch": 1, "loss": 0.5602, "lr": 1.9328916962226644e-05, "grad_norm": 9.875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 1555, "epoch": 1, "loss": 0.2042, "lr": 1.932783772348509e-05, "grad_norm": 5.71875, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 1556, "epoch": 1, "loss": 0.0082, "lr": 1.932675764780153e-05, "grad_norm": 5.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1557, "epoch": 1, "loss": 0.0022, "lr": 1.9325676735272864e-05, "grad_norm": 2.609375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 1558, "epoch": 1, "loss": 0.0013, "lr": 1.932459498599609e-05, "grad_norm": 4.25, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 1559, "epoch": 1, "loss": 0.0241, "lr": 1.9323512400068262e-05, "grad_norm": 6.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1560, "epoch": 1, "loss": 0.0122, "lr": 1.9322428977586515e-05, "grad_norm": 2.40625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1561, "epoch": 1, "loss": 0.2313, "lr": 1.932134471864806e-05, "grad_norm": 4.28125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1562, "epoch": 1, "loss": 0.1846, "lr": 1.9320259623350187e-05, "grad_norm": 7.4375, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 1563, "epoch": 1, "loss": 0.0266, "lr": 1.931917369179025e-05, "grad_norm": 8.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1564, "epoch": 1, "loss": 0.0017, "lr": 1.9318086924065683e-05, "grad_norm": 10.0, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 1565, "epoch": 1, "loss": 0.0707, "lr": 1.9316999320274005e-05, "grad_norm": 5.4375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1566, "epoch": 1, "loss": 0.0017, "lr": 1.9315910880512792e-05, "grad_norm": 6.625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1567, "epoch": 1, "loss": 0.2327, "lr": 1.931482160487971e-05, "grad_norm": 4.78125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1568, "epoch": 1, "loss": 0.0597, "lr": 1.9313731493472494e-05, "grad_norm": 4.71875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 1569, "epoch": 1, "loss": 0.0583, "lr": 1.9312640546388955e-05, "grad_norm": 4.90625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1570, "epoch": 1, "loss": 0.0016, "lr": 1.9311548763726973e-05, "grad_norm": 3.65625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1571, "epoch": 1, "loss": 0.0418, "lr": 1.9310456145584517e-05, "grad_norm": 3.640625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1572, "epoch": 1, "loss": 0.1861, "lr": 1.9309362692059617e-05, "grad_norm": 6.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1573, "epoch": 1, "loss": 0.1548, "lr": 1.9308268403250382e-05, "grad_norm": 5.09375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1574, "epoch": 1, "loss": 0.0644, "lr": 1.9307173279255002e-05, "grad_norm": 7.6875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1575, "epoch": 1, "loss": 0.1068, "lr": 1.9306077320171734e-05, "grad_norm": 7.1875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1576, "epoch": 1, "loss": 0.0467, "lr": 1.9304980526098914e-05, "grad_norm": 4.09375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1577, "epoch": 1, "loss": 0.1518, "lr": 1.9303882897134954e-05, "grad_norm": 6.40625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1578, "epoch": 1, "loss": 0.0026, "lr": 1.9302784433378333e-05, "grad_norm": 2.640625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 1579, "epoch": 1, "loss": 0.046, "lr": 1.9301685134927616e-05, "grad_norm": 5.34375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 1580, "epoch": 1, "loss": 0.0449, "lr": 1.9300585001881438e-05, "grad_norm": 5.1875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1581, "epoch": 1, "loss": 0.0016, "lr": 1.9299484034338506e-05, "grad_norm": 22.75, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1582, "epoch": 1, "loss": 0.1037, "lr": 1.9298382232397606e-05, "grad_norm": 3.515625, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1583, "epoch": 1, "loss": 0.1498, "lr": 1.9297279596157596e-05, "grad_norm": 3.359375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1584, "epoch": 1, "loss": 0.0263, "lr": 1.9296176125717414e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1585, "epoch": 1, "loss": 0.0028, "lr": 1.9295071821176065e-05, "grad_norm": 4.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1586, "epoch": 1, "loss": 0.165, "lr": 1.9293966682632637e-05, "grad_norm": 7.71875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1587, "epoch": 1, "loss": 0.0048, "lr": 1.9292860710186284e-05, "grad_norm": 5.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1588, "epoch": 1, "loss": 0.0028, "lr": 1.929175390393624e-05, "grad_norm": 3.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1589, "epoch": 1, "loss": 0.0017, "lr": 1.9290646263981816e-05, "grad_norm": 2.0625, "samples_per_sec": 10.31, "gpu_mem_gb": 6.83} {"step": 1590, "epoch": 1, "loss": 0.0026, "lr": 1.9289537790422395e-05, "grad_norm": 2.8125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1591, "epoch": 1, "loss": 0.1786, "lr": 1.9288428483357437e-05, "grad_norm": 4.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 1592, "epoch": 1, "loss": 0.0267, "lr": 1.928731834288647e-05, "grad_norm": 1.671875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1593, "epoch": 1, "loss": 0.0002, "lr": 1.92862073691091e-05, "grad_norm": 8.6875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1594, "epoch": 1, "loss": 0.0616, "lr": 1.9285095562125015e-05, "grad_norm": 3.703125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1595, "epoch": 1, "loss": 0.0062, "lr": 1.9283982922033972e-05, "grad_norm": 6.21875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 1596, "epoch": 1, "loss": 0.0015, "lr": 1.92828694489358e-05, "grad_norm": 4.59375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 1597, "epoch": 1, "loss": 0.102, "lr": 1.928175514293041e-05, "grad_norm": 7.5, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 1598, "epoch": 1, "loss": 0.0574, "lr": 1.9280640004117775e-05, "grad_norm": 4.90625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1599, "epoch": 1, "loss": 0.0489, "lr": 1.9279524032597958e-05, "grad_norm": 8.875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1600, "epoch": 1, "loss": 0.1477, "lr": 1.9278407228471085e-05, "grad_norm": 7.28125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1601, "epoch": 1, "loss": 0.2591, "lr": 1.9277289591837366e-05, "grad_norm": 7.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1602, "epoch": 1, "loss": 0.0042, "lr": 1.927617112279708e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.39, "gpu_mem_gb": 6.85} {"step": 1603, "epoch": 1, "loss": 0.0003, "lr": 1.9275051821450584e-05, "grad_norm": 3.09375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1604, "epoch": 1, "loss": 0.0327, "lr": 1.92739316878983e-05, "grad_norm": 3.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1605, "epoch": 1, "loss": 0.1896, "lr": 1.927281072224074e-05, "grad_norm": 5.96875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1606, "epoch": 1, "loss": 0.1681, "lr": 1.9271688924578476e-05, "grad_norm": 6.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1607, "epoch": 1, "loss": 0.0248, "lr": 1.927056629501217e-05, "grad_norm": 5.125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1608, "epoch": 1, "loss": 0.0654, "lr": 1.926944283364254e-05, "grad_norm": 2.90625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1609, "epoch": 1, "loss": 0.0019, "lr": 1.9268318540570395e-05, "grad_norm": 6.3125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1610, "epoch": 1, "loss": 0.0294, "lr": 1.9267193415896614e-05, "grad_norm": 4.09375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 1611, "epoch": 1, "loss": 0.1195, "lr": 1.926606745972215e-05, "grad_norm": 3.765625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 1612, "epoch": 1, "loss": 0.005, "lr": 1.9264940672148018e-05, "grad_norm": 2.96875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1613, "epoch": 1, "loss": 0.0395, "lr": 1.9263813053275332e-05, "grad_norm": 4.65625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1614, "epoch": 1, "loss": 0.0002, "lr": 1.9262684603205264e-05, "grad_norm": 0.4609375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1615, "epoch": 1, "loss": 0.0162, "lr": 1.9261555322039062e-05, "grad_norm": 4.5625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 1616, "epoch": 1, "loss": 0.0464, "lr": 1.9260425209878052e-05, "grad_norm": 7.5, "samples_per_sec": 10.33, "gpu_mem_gb": 6.85} {"step": 1617, "epoch": 1, "loss": 0.0105, "lr": 1.925929426682364e-05, "grad_norm": 9.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1618, "epoch": 1, "loss": 0.1934, "lr": 1.9258162492977286e-05, "grad_norm": 7.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 1619, "epoch": 1, "loss": 0.0898, "lr": 1.925702988844055e-05, "grad_norm": 6.375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 1620, "epoch": 1, "loss": 0.1504, "lr": 1.9255896453315054e-05, "grad_norm": 4.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1621, "epoch": 1, "loss": 0.0341, "lr": 1.9254762187702494e-05, "grad_norm": 7.09375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1622, "epoch": 1, "loss": 0.0033, "lr": 1.9253627091704637e-05, "grad_norm": 2.515625, "samples_per_sec": 10.33, "gpu_mem_gb": 6.82} {"step": 1623, "epoch": 1, "loss": 0.0036, "lr": 1.925249116542334e-05, "grad_norm": 5.6875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.85} {"step": 1624, "epoch": 1, "loss": 0.0003, "lr": 1.9251354408960515e-05, "grad_norm": 4.09375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1625, "epoch": 1, "loss": 0.0028, "lr": 1.9250216822418164e-05, "grad_norm": 2.09375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1626, "epoch": 1, "loss": 0.116, "lr": 1.9249078405898353e-05, "grad_norm": 4.1875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1627, "epoch": 1, "loss": 0.178, "lr": 1.9247939159503232e-05, "grad_norm": 3.4375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.84} {"step": 1628, "epoch": 1, "loss": 0.0005, "lr": 1.9246799083335012e-05, "grad_norm": 5.6875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1629, "epoch": 1, "loss": 0.1018, "lr": 1.924565817749599e-05, "grad_norm": 6.53125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1630, "epoch": 1, "loss": 0.0009, "lr": 1.9244516442088537e-05, "grad_norm": 5.125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 1631, "epoch": 1, "loss": 0.1739, "lr": 1.92433738772151e-05, "grad_norm": 7.40625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1632, "epoch": 1, "loss": 0.2634, "lr": 1.924223048297818e-05, "grad_norm": 3.65625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 1633, "epoch": 1, "loss": 0.0648, "lr": 1.924108625948038e-05, "grad_norm": 2.265625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1634, "epoch": 1, "loss": 0.1534, "lr": 1.9239941206824362e-05, "grad_norm": 4.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1635, "epoch": 1, "loss": 0.003, "lr": 1.9238795325112867e-05, "grad_norm": 6.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1636, "epoch": 1, "loss": 0.001, "lr": 1.9237648614448713e-05, "grad_norm": 2.171875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 1637, "epoch": 1, "loss": 0.1175, "lr": 1.9236501074934786e-05, "grad_norm": 4.625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 1638, "epoch": 1, "loss": 0.0015, "lr": 1.9235352706674046e-05, "grad_norm": 2.203125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 1639, "epoch": 1, "loss": 0.0217, "lr": 1.9234203509769535e-05, "grad_norm": 7.6875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1640, "epoch": 1, "loss": 0.0009, "lr": 1.923305348432436e-05, "grad_norm": 5.625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1641, "epoch": 1, "loss": 0.0934, "lr": 1.9231902630441713e-05, "grad_norm": 5.1875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 1642, "epoch": 1, "loss": 0.0657, "lr": 1.923075094822485e-05, "grad_norm": 4.125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1643, "epoch": 1, "loss": 0.0801, "lr": 1.922959843777711e-05, "grad_norm": 4.4375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 1644, "epoch": 1, "loss": 0.2205, "lr": 1.92284450992019e-05, "grad_norm": 7.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.84} {"step": 1645, "epoch": 1, "loss": 0.0107, "lr": 1.9227290932602703e-05, "grad_norm": 2.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1646, "epoch": 1, "loss": 0.0034, "lr": 1.9226135938083078e-05, "grad_norm": 1.7890625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 1647, "epoch": 1, "loss": 0.0588, "lr": 1.9224980115746658e-05, "grad_norm": 6.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1648, "epoch": 1, "loss": 0.0154, "lr": 1.9223823465697146e-05, "grad_norm": 4.21875, "samples_per_sec": 8.88, "gpu_mem_gb": 6.83} {"step": 1649, "epoch": 1, "loss": 0.0024, "lr": 1.9222665988038324e-05, "grad_norm": 8.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1650, "epoch": 1, "loss": 0.234, "lr": 1.9221507682874047e-05, "grad_norm": 6.625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1651, "epoch": 1, "loss": 0.1882, "lr": 1.9220348550308247e-05, "grad_norm": 7.4375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1652, "epoch": 1, "loss": 0.1691, "lr": 1.9219188590444924e-05, "grad_norm": 7.125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1653, "epoch": 1, "loss": 0.1721, "lr": 1.9218027803388157e-05, "grad_norm": 4.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1654, "epoch": 1, "loss": 0.2672, "lr": 1.9216866189242095e-05, "grad_norm": 4.65625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1655, "epoch": 1, "loss": 0.1344, "lr": 1.921570374811097e-05, "grad_norm": 5.6875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 1656, "epoch": 1, "loss": 0.0807, "lr": 1.921454048009908e-05, "grad_norm": 4.59375, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 1657, "epoch": 1, "loss": 0.0293, "lr": 1.9213376385310794e-05, "grad_norm": 5.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1658, "epoch": 1, "loss": 0.081, "lr": 1.921221146385057e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1659, "epoch": 1, "loss": 0.0325, "lr": 1.921104571582292e-05, "grad_norm": 6.03125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 1660, "epoch": 1, "loss": 0.0567, "lr": 1.9209879141332448e-05, "grad_norm": 4.375, "samples_per_sec": 11.02, "gpu_mem_gb": 6.85} {"step": 1661, "epoch": 1, "loss": 0.001, "lr": 1.9208711740483824e-05, "grad_norm": 7.84375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1662, "epoch": 1, "loss": 0.0042, "lr": 1.9207543513381793e-05, "grad_norm": 4.65625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1663, "epoch": 1, "loss": 0.3368, "lr": 1.9206374460131175e-05, "grad_norm": 6.65625, "samples_per_sec": 10.19, "gpu_mem_gb": 6.83} {"step": 1664, "epoch": 1, "loss": 0.0041, "lr": 1.9205204580836866e-05, "grad_norm": 1.6640625, "samples_per_sec": 11.03, "gpu_mem_gb": 6.83} {"step": 1665, "epoch": 1, "loss": 0.1951, "lr": 1.9204033875603824e-05, "grad_norm": 4.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1666, "epoch": 1, "loss": 0.0767, "lr": 1.9202862344537104e-05, "grad_norm": 3.28125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1667, "epoch": 1, "loss": 0.0278, "lr": 1.9201689987741812e-05, "grad_norm": 4.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 1668, "epoch": 1, "loss": 0.0006, "lr": 1.920051680532314e-05, "grad_norm": 4.25, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1669, "epoch": 1, "loss": 0.0072, "lr": 1.9199342797386357e-05, "grad_norm": 6.28125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1670, "epoch": 1, "loss": 0.4164, "lr": 1.919816796403679e-05, "grad_norm": 11.0, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 1671, "epoch": 1, "loss": 0.1025, "lr": 1.9196992305379868e-05, "grad_norm": 3.40625, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 1672, "epoch": 1, "loss": 0.0166, "lr": 1.919581582152106e-05, "grad_norm": 2.390625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 1673, "epoch": 1, "loss": 0.0494, "lr": 1.9194638512565937e-05, "grad_norm": 4.46875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1674, "epoch": 1, "loss": 0.0187, "lr": 1.9193460378620128e-05, "grad_norm": 3.6875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1675, "epoch": 1, "loss": 0.0205, "lr": 1.9192281419789346e-05, "grad_norm": 4.90625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 1676, "epoch": 1, "loss": 0.0592, "lr": 1.9191101636179368e-05, "grad_norm": 4.78125, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 1677, "epoch": 1, "loss": 0.1344, "lr": 1.9189921027896056e-05, "grad_norm": 5.875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 1678, "epoch": 1, "loss": 0.2404, "lr": 1.9188739595045333e-05, "grad_norm": 5.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1679, "epoch": 1, "loss": 0.0902, "lr": 1.918755733773321e-05, "grad_norm": 7.71875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1680, "epoch": 1, "loss": 0.1392, "lr": 1.9186374256065763e-05, "grad_norm": 3.609375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1681, "epoch": 1, "loss": 0.0168, "lr": 1.9185190350149146e-05, "grad_norm": 2.78125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1682, "epoch": 1, "loss": 0.0065, "lr": 1.918400562008958e-05, "grad_norm": 1.171875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1683, "epoch": 1, "loss": 0.032, "lr": 1.9182820065993367e-05, "grad_norm": 3.015625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 1684, "epoch": 1, "loss": 0.3449, "lr": 1.918163368796689e-05, "grad_norm": 8.75, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 1685, "epoch": 1, "loss": 0.0119, "lr": 1.918044648611658e-05, "grad_norm": 5.375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1686, "epoch": 1, "loss": 0.1098, "lr": 1.9179258460548974e-05, "grad_norm": 3.90625, "samples_per_sec": 10.3, "gpu_mem_gb": 6.83} {"step": 1687, "epoch": 1, "loss": 0.1768, "lr": 1.917806961137066e-05, "grad_norm": 6.71875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1688, "epoch": 1, "loss": 0.0121, "lr": 1.917687993868831e-05, "grad_norm": 4.28125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1689, "epoch": 1, "loss": 0.003, "lr": 1.9175689442608667e-05, "grad_norm": 2.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1690, "epoch": 1, "loss": 0.0679, "lr": 1.9174498123238552e-05, "grad_norm": 3.09375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1691, "epoch": 1, "loss": 0.0212, "lr": 1.917330598068485e-05, "grad_norm": 2.015625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1692, "epoch": 1, "loss": 0.0043, "lr": 1.917211301505453e-05, "grad_norm": 6.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1693, "epoch": 1, "loss": 0.0292, "lr": 1.9170919226454633e-05, "grad_norm": 4.15625, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 1694, "epoch": 1, "loss": 0.087, "lr": 1.916972461499227e-05, "grad_norm": 6.21875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1695, "epoch": 1, "loss": 0.0012, "lr": 1.9168529180774623e-05, "grad_norm": 4.78125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1696, "epoch": 1, "loss": 0.0308, "lr": 1.916733292390896e-05, "grad_norm": 7.0625, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 1697, "epoch": 1, "loss": 0.0151, "lr": 1.916613584450261e-05, "grad_norm": 4.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1698, "epoch": 1, "loss": 0.152, "lr": 1.9164937942662985e-05, "grad_norm": 5.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1699, "epoch": 1, "loss": 0.0847, "lr": 1.9163739218497564e-05, "grad_norm": 3.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1700, "epoch": 1, "loss": 0.001, "lr": 1.9162539672113903e-05, "grad_norm": 0.408203125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1701, "epoch": 1, "loss": 0.0355, "lr": 1.9161339303619637e-05, "grad_norm": 3.84375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 1702, "epoch": 1, "loss": 0.0184, "lr": 1.9160138113122463e-05, "grad_norm": 3.265625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1703, "epoch": 1, "loss": 0.0037, "lr": 1.915893610073016e-05, "grad_norm": 5.125, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 1704, "epoch": 1, "loss": 0.1492, "lr": 1.9157733266550577e-05, "grad_norm": 5.90625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 1705, "epoch": 1, "loss": 0.0023, "lr": 1.915652961069164e-05, "grad_norm": 10.125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 1706, "epoch": 1, "loss": 0.2145, "lr": 1.915532513326135e-05, "grad_norm": 3.4375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 1707, "epoch": 1, "loss": 0.0025, "lr": 1.9154119834367774e-05, "grad_norm": 3.5, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1708, "epoch": 1, "loss": 0.0014, "lr": 1.915291371411906e-05, "grad_norm": 6.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 1709, "epoch": 1, "loss": 0.0332, "lr": 1.9151706772623432e-05, "grad_norm": 2.15625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 1710, "epoch": 1, "loss": 0.2795, "lr": 1.9150499009989174e-05, "grad_norm": 6.21875, "samples_per_sec": 10.22, "gpu_mem_gb": 6.84} {"step": 1711, "epoch": 1, "loss": 0.0037, "lr": 1.9149290426324658e-05, "grad_norm": 2.578125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1712, "epoch": 1, "loss": 0.039, "lr": 1.9148081021738327e-05, "grad_norm": 1.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1713, "epoch": 1, "loss": 0.1985, "lr": 1.914687079633869e-05, "grad_norm": 3.375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.84} {"step": 1714, "epoch": 1, "loss": 0.1485, "lr": 1.9145659750234334e-05, "grad_norm": 5.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 1715, "epoch": 1, "loss": 0.1766, "lr": 1.9144447883533922e-05, "grad_norm": 6.375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.84} {"step": 1716, "epoch": 1, "loss": 0.0453, "lr": 1.9143235196346194e-05, "grad_norm": 3.78125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1717, "epoch": 1, "loss": 0.0366, "lr": 1.9142021688779954e-05, "grad_norm": 1.7734375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1718, "epoch": 1, "loss": 0.0052, "lr": 1.914080736094408e-05, "grad_norm": 8.6875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 1719, "epoch": 1, "loss": 0.0019, "lr": 1.9139592212947536e-05, "grad_norm": 3.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1720, "epoch": 1, "loss": 0.0014, "lr": 1.9138376244899347e-05, "grad_norm": 5.28125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 1721, "epoch": 1, "loss": 0.0286, "lr": 1.9137159456908615e-05, "grad_norm": 2.578125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1722, "epoch": 1, "loss": 0.001, "lr": 1.9135941849084522e-05, "grad_norm": 3.765625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1723, "epoch": 1, "loss": 0.0072, "lr": 1.9134723421536312e-05, "grad_norm": 3.984375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 1724, "epoch": 1, "loss": 0.0004, "lr": 1.913350417437331e-05, "grad_norm": 0.65625, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 1725, "epoch": 1, "loss": 0.1148, "lr": 1.9132284107704918e-05, "grad_norm": 6.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 1726, "epoch": 1, "loss": 0.1365, "lr": 1.9131063221640597e-05, "grad_norm": 2.84375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 1727, "epoch": 1, "loss": 0.1748, "lr": 1.9129841516289902e-05, "grad_norm": 3.65625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1728, "epoch": 1, "loss": 0.0002, "lr": 1.9128618991762442e-05, "grad_norm": 2.578125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1729, "epoch": 1, "loss": 0.0017, "lr": 1.9127395648167916e-05, "grad_norm": 1.9921875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1730, "epoch": 1, "loss": 0.0004, "lr": 1.912617148561608e-05, "grad_norm": 2.109375, "samples_per_sec": 10.27, "gpu_mem_gb": 6.83} {"step": 1731, "epoch": 1, "loss": 0.032, "lr": 1.9124946504216782e-05, "grad_norm": 6.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1732, "epoch": 1, "loss": 0.0046, "lr": 1.9123720704079928e-05, "grad_norm": 3.484375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1733, "epoch": 1, "loss": 0.0008, "lr": 1.9122494085315502e-05, "grad_norm": 2.875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1734, "epoch": 1, "loss": 0.0006, "lr": 1.9121266648033564e-05, "grad_norm": 3.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1735, "epoch": 1, "loss": 0.1835, "lr": 1.912003839234425e-05, "grad_norm": 6.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1736, "epoch": 1, "loss": 0.0007, "lr": 1.911880931835776e-05, "grad_norm": 1.84375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1737, "epoch": 1, "loss": 0.0047, "lr": 1.9117579426184374e-05, "grad_norm": 3.203125, "samples_per_sec": 10.42, "gpu_mem_gb": 6.85} {"step": 1738, "epoch": 1, "loss": 0.054, "lr": 1.9116348715934446e-05, "grad_norm": 4.09375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.85} {"step": 1739, "epoch": 1, "loss": 0.0012, "lr": 1.91151171877184e-05, "grad_norm": 2.640625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1740, "epoch": 1, "loss": 0.1697, "lr": 1.9113884841646736e-05, "grad_norm": 7.28125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 1741, "epoch": 1, "loss": 0.1012, "lr": 1.9112651677830025e-05, "grad_norm": 4.9375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1742, "epoch": 1, "loss": 0.0058, "lr": 1.9111417696378916e-05, "grad_norm": 1.2734375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1743, "epoch": 1, "loss": 0.3197, "lr": 1.911018289740413e-05, "grad_norm": 7.8125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1744, "epoch": 1, "loss": 0.0003, "lr": 1.9108947281016453e-05, "grad_norm": 6.75, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1745, "epoch": 1, "loss": 0.0533, "lr": 1.9107710847326753e-05, "grad_norm": 8.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1746, "epoch": 1, "loss": 0.0091, "lr": 1.910647359644597e-05, "grad_norm": 5.03125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 1747, "epoch": 1, "loss": 0.0321, "lr": 1.910523552848512e-05, "grad_norm": 5.375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 1748, "epoch": 1, "loss": 0.097, "lr": 1.9103996643555285e-05, "grad_norm": 4.5625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1749, "epoch": 1, "loss": 0.2835, "lr": 1.9102756941767625e-05, "grad_norm": 4.15625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1750, "epoch": 1, "loss": 0.0003, "lr": 1.910151642323337e-05, "grad_norm": 5.75, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 1751, "epoch": 1, "loss": 0.0665, "lr": 1.910027508806383e-05, "grad_norm": 3.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1752, "epoch": 1, "loss": 0.0009, "lr": 1.9099032936370382e-05, "grad_norm": 2.640625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1753, "epoch": 1, "loss": 0.0353, "lr": 1.909778996826448e-05, "grad_norm": 3.75, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 1754, "epoch": 1, "loss": 0.1236, "lr": 1.9096546183857644e-05, "grad_norm": 4.84375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 1755, "epoch": 1, "loss": 0.1035, "lr": 1.909530158326148e-05, "grad_norm": 6.1875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 1756, "epoch": 1, "loss": 0.0963, "lr": 1.9094056166587654e-05, "grad_norm": 5.0625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 1757, "epoch": 1, "loss": 0.0009, "lr": 1.909280993394792e-05, "grad_norm": 4.46875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1758, "epoch": 1, "loss": 0.0008, "lr": 1.9091562885454085e-05, "grad_norm": 4.03125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1759, "epoch": 1, "loss": 0.0061, "lr": 1.9090315021218047e-05, "grad_norm": 0.3984375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1760, "epoch": 1, "loss": 0.0001, "lr": 1.908906634135177e-05, "grad_norm": 6.375, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 1761, "epoch": 1, "loss": 0.1057, "lr": 1.9087816845967296e-05, "grad_norm": 4.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1762, "epoch": 1, "loss": 0.0985, "lr": 1.9086566535176726e-05, "grad_norm": 5.59375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1763, "epoch": 1, "loss": 0.0002, "lr": 1.9085315409092254e-05, "grad_norm": 0.84375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 1764, "epoch": 1, "loss": 0.1111, "lr": 1.9084063467826137e-05, "grad_norm": 7.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 1765, "epoch": 1, "loss": 0.1263, "lr": 1.90828107114907e-05, "grad_norm": 12.1875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1766, "epoch": 1, "loss": 0.1588, "lr": 1.9081557140198352e-05, "grad_norm": 4.03125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1767, "epoch": 1, "loss": 0.0027, "lr": 1.9080302754061564e-05, "grad_norm": 1.2734375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 1768, "epoch": 1, "loss": 0.0007, "lr": 1.907904755319289e-05, "grad_norm": 4.28125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1769, "epoch": 1, "loss": 0.0772, "lr": 1.9077791537704956e-05, "grad_norm": 5.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1770, "epoch": 1, "loss": 0.0732, "lr": 1.9076534707710453e-05, "grad_norm": 4.25, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 1771, "epoch": 1, "loss": 0.0038, "lr": 1.907527706332215e-05, "grad_norm": 4.34375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 1772, "epoch": 1, "loss": 0.0975, "lr": 1.9074018604652894e-05, "grad_norm": 16.875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1773, "epoch": 1, "loss": 0.1439, "lr": 1.9072759331815602e-05, "grad_norm": 5.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 1774, "epoch": 1, "loss": 0.0085, "lr": 1.9071499244923255e-05, "grad_norm": 9.5625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1775, "epoch": 1, "loss": 0.0006, "lr": 1.9070238344088917e-05, "grad_norm": 5.28125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 1776, "epoch": 1, "loss": 0.0019, "lr": 1.9068976629425725e-05, "grad_norm": 1.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1777, "epoch": 1, "loss": 0.0753, "lr": 1.9067714101046886e-05, "grad_norm": 4.71875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1778, "epoch": 1, "loss": 0.0056, "lr": 1.906645075906568e-05, "grad_norm": 0.8359375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 1779, "epoch": 1, "loss": 0.0156, "lr": 1.906518660359546e-05, "grad_norm": 5.15625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1780, "epoch": 1, "loss": 0.0897, "lr": 1.906392163474965e-05, "grad_norm": 5.6875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 1781, "epoch": 1, "loss": 0.0003, "lr": 1.906265585264176e-05, "grad_norm": 8.875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1782, "epoch": 1, "loss": 0.0005, "lr": 1.9061389257385347e-05, "grad_norm": 1.1484375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1783, "epoch": 1, "loss": 0.0012, "lr": 1.906012184909407e-05, "grad_norm": 3.234375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1784, "epoch": 1, "loss": 0.0613, "lr": 1.905885362788164e-05, "grad_norm": 4.40625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1785, "epoch": 1, "loss": 0.0051, "lr": 1.905758459386185e-05, "grad_norm": 6.21875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1786, "epoch": 1, "loss": 0.1225, "lr": 1.905631474714857e-05, "grad_norm": 5.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1787, "epoch": 1, "loss": 0.0051, "lr": 1.9055044087855728e-05, "grad_norm": 1.7265625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 1788, "epoch": 1, "loss": 0.0811, "lr": 1.905377261609734e-05, "grad_norm": 3.890625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1789, "epoch": 1, "loss": 0.1349, "lr": 1.9052500331987483e-05, "grad_norm": 6.5625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 1790, "epoch": 1, "loss": 0.0041, "lr": 1.9051227235640323e-05, "grad_norm": 5.0, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 1791, "epoch": 1, "loss": 0.0624, "lr": 1.9049953327170078e-05, "grad_norm": 2.984375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 1792, "epoch": 1, "loss": 0.0016, "lr": 1.9048678606691058e-05, "grad_norm": 2.75, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 1793, "epoch": 1, "loss": 0.0679, "lr": 1.9047403074317636e-05, "grad_norm": 6.90625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1794, "epoch": 1, "loss": 0.0435, "lr": 1.9046126730164252e-05, "grad_norm": 5.1875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 1795, "epoch": 1, "loss": 0.0462, "lr": 1.9044849574345437e-05, "grad_norm": 5.8125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1796, "epoch": 1, "loss": 0.0746, "lr": 1.9043571606975776e-05, "grad_norm": 5.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1797, "epoch": 1, "loss": 0.0616, "lr": 1.904229282816994e-05, "grad_norm": 3.875, "samples_per_sec": 10.37, "gpu_mem_gb": 6.85} {"step": 1798, "epoch": 1, "loss": 0.0734, "lr": 1.9041013238042664e-05, "grad_norm": 7.46875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1799, "epoch": 1, "loss": 0.0031, "lr": 1.903973283670876e-05, "grad_norm": 2.75, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1800, "epoch": 1, "loss": 0.1582, "lr": 1.9038451624283113e-05, "grad_norm": 4.21875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1801, "epoch": 1, "loss": 0.002, "lr": 1.9037169600880683e-05, "grad_norm": 4.75, "samples_per_sec": 10.6, "gpu_mem_gb": 6.85} {"step": 1802, "epoch": 1, "loss": 0.21, "lr": 1.9035886766616497e-05, "grad_norm": 7.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.84} {"step": 1803, "epoch": 1, "loss": 0.0051, "lr": 1.9034603121605654e-05, "grad_norm": 4.5, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 1804, "epoch": 1, "loss": 0.0685, "lr": 1.9033318665963336e-05, "grad_norm": 5.84375, "samples_per_sec": 10.13, "gpu_mem_gb": 6.85} {"step": 1805, "epoch": 1, "loss": 0.013, "lr": 1.9032033399804786e-05, "grad_norm": 4.28125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1806, "epoch": 1, "loss": 0.0001, "lr": 1.903074732324533e-05, "grad_norm": 2.84375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1807, "epoch": 1, "loss": 0.2014, "lr": 1.9029460436400355e-05, "grad_norm": 9.9375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1808, "epoch": 1, "loss": 0.001, "lr": 1.902817273938533e-05, "grad_norm": 5.1875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1809, "epoch": 1, "loss": 0.0862, "lr": 1.9026884232315796e-05, "grad_norm": 4.46875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 1810, "epoch": 1, "loss": 0.0674, "lr": 1.902559491530736e-05, "grad_norm": 4.8125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 1811, "epoch": 1, "loss": 0.0156, "lr": 1.902430478847571e-05, "grad_norm": 4.75, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 1812, "epoch": 1, "loss": 0.1024, "lr": 1.9023013851936603e-05, "grad_norm": 5.8125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1813, "epoch": 1, "loss": 0.0143, "lr": 1.9021722105805865e-05, "grad_norm": 3.703125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1814, "epoch": 1, "loss": 0.0655, "lr": 1.9020429550199403e-05, "grad_norm": 4.25, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 1815, "epoch": 1, "loss": 0.4053, "lr": 1.901913618523319e-05, "grad_norm": 9.75, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 1816, "epoch": 1, "loss": 0.0012, "lr": 1.901784201102327e-05, "grad_norm": 0.3984375, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 1817, "epoch": 1, "loss": 0.0002, "lr": 1.9016547027685765e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1818, "epoch": 1, "loss": 0.0182, "lr": 1.9015251235336873e-05, "grad_norm": 3.171875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1819, "epoch": 1, "loss": 0.0002, "lr": 1.9013954634092847e-05, "grad_norm": 4.375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 1820, "epoch": 1, "loss": 0.0087, "lr": 1.9012657224070038e-05, "grad_norm": 6.8125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1821, "epoch": 1, "loss": 0.0073, "lr": 1.901135900538485e-05, "grad_norm": 2.84375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 1822, "epoch": 1, "loss": 0.0264, "lr": 1.9010059978153767e-05, "grad_norm": 3.609375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 1823, "epoch": 1, "loss": 0.0009, "lr": 1.900876014249334e-05, "grad_norm": 2.984375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1824, "epoch": 1, "loss": 0.035, "lr": 1.9007459498520203e-05, "grad_norm": 4.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1825, "epoch": 1, "loss": 0.037, "lr": 1.900615804635106e-05, "grad_norm": 6.84375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 1826, "epoch": 1, "loss": 0.0644, "lr": 1.900485578610267e-05, "grad_norm": 5.03125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1827, "epoch": 1, "loss": 0.0465, "lr": 1.9003552717891893e-05, "grad_norm": 3.3125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1828, "epoch": 1, "loss": 0.0092, "lr": 1.9002248841835638e-05, "grad_norm": 8.125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1829, "epoch": 1, "loss": 0.0013, "lr": 1.9000944158050903e-05, "grad_norm": 3.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1830, "epoch": 1, "loss": 0.0624, "lr": 1.8999638666654744e-05, "grad_norm": 5.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1831, "epoch": 1, "loss": 0.0332, "lr": 1.89983323677643e-05, "grad_norm": 5.375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 1832, "epoch": 1, "loss": 0.0318, "lr": 1.8997025261496777e-05, "grad_norm": 4.6875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1833, "epoch": 1, "loss": 0.1381, "lr": 1.8995717347969454e-05, "grad_norm": 3.40625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.84} {"step": 1834, "epoch": 1, "loss": 0.0245, "lr": 1.8994408627299693e-05, "grad_norm": 4.1875, "samples_per_sec": 10.53, "gpu_mem_gb": 6.82} {"step": 1835, "epoch": 1, "loss": 0.0066, "lr": 1.8993099099604908e-05, "grad_norm": 6.0625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1836, "epoch": 1, "loss": 0.2513, "lr": 1.89917887650026e-05, "grad_norm": 5.875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 1837, "epoch": 1, "loss": 0.036, "lr": 1.8990477623610343e-05, "grad_norm": 6.53125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 1838, "epoch": 1, "loss": 0.0072, "lr": 1.8989165675545776e-05, "grad_norm": 2.796875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1839, "epoch": 1, "loss": 0.0553, "lr": 1.898785292092661e-05, "grad_norm": 6.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 1840, "epoch": 1, "loss": 0.1002, "lr": 1.8986539359870644e-05, "grad_norm": 4.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1841, "epoch": 1, "loss": 0.0555, "lr": 1.8985224992495723e-05, "grad_norm": 3.890625, "samples_per_sec": 10.25, "gpu_mem_gb": 6.83} {"step": 1842, "epoch": 1, "loss": 0.0038, "lr": 1.898390981891979e-05, "grad_norm": 3.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1843, "epoch": 1, "loss": 0.0018, "lr": 1.898259383926084e-05, "grad_norm": 2.890625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1844, "epoch": 1, "loss": 0.0242, "lr": 1.8981277053636963e-05, "grad_norm": 2.953125, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 1845, "epoch": 1, "loss": 0.1823, "lr": 1.8979959462166295e-05, "grad_norm": 7.4375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.84} {"step": 1846, "epoch": 1, "loss": 0.1006, "lr": 1.897864106496706e-05, "grad_norm": 3.96875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1847, "epoch": 1, "loss": 0.0176, "lr": 1.8977321862157557e-05, "grad_norm": 5.15625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1848, "epoch": 1, "loss": 0.0422, "lr": 1.8976001853856142e-05, "grad_norm": 6.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1849, "epoch": 1, "loss": 0.0279, "lr": 1.8974681040181266e-05, "grad_norm": 3.1875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 1850, "epoch": 1, "loss": 0.0925, "lr": 1.8973359421251424e-05, "grad_norm": 6.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1851, "epoch": 1, "loss": 0.0846, "lr": 1.8972036997185213e-05, "grad_norm": 5.0625, "samples_per_sec": 10.31, "gpu_mem_gb": 6.82} {"step": 1852, "epoch": 1, "loss": 0.0641, "lr": 1.8970713768101277e-05, "grad_norm": 4.25, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 1853, "epoch": 1, "loss": 0.1114, "lr": 1.8969389734118348e-05, "grad_norm": 2.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 1854, "epoch": 1, "loss": 0.0098, "lr": 1.896806489535523e-05, "grad_norm": 5.84375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1855, "epoch": 1, "loss": 0.0006, "lr": 1.8966739251930783e-05, "grad_norm": 6.9375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 1856, "epoch": 1, "loss": 0.0509, "lr": 1.8965412803963957e-05, "grad_norm": 4.96875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1857, "epoch": 1, "loss": 0.1582, "lr": 1.8964085551573767e-05, "grad_norm": 4.28125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1858, "epoch": 1, "loss": 0.024, "lr": 1.89627574948793e-05, "grad_norm": 3.703125, "samples_per_sec": 10.19, "gpu_mem_gb": 6.83} {"step": 1859, "epoch": 1, "loss": 0.0436, "lr": 1.896142863399972e-05, "grad_norm": 5.375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1860, "epoch": 1, "loss": 0.0066, "lr": 1.8960098969054253e-05, "grad_norm": 5.40625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 1861, "epoch": 1, "loss": 0.0043, "lr": 1.8958768500162213e-05, "grad_norm": 3.75, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1862, "epoch": 1, "loss": 0.0484, "lr": 1.8957437227442966e-05, "grad_norm": 3.21875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.85} {"step": 1863, "epoch": 1, "loss": 0.0019, "lr": 1.8956105151015966e-05, "grad_norm": 4.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 1864, "epoch": 1, "loss": 0.105, "lr": 1.8954772271000738e-05, "grad_norm": 4.59375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 1865, "epoch": 1, "loss": 0.0063, "lr": 1.8953438587516863e-05, "grad_norm": 3.21875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 1866, "epoch": 1, "loss": 0.0157, "lr": 1.8952104100684017e-05, "grad_norm": 5.9375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 1867, "epoch": 1, "loss": 0.0388, "lr": 1.8950768810621932e-05, "grad_norm": 7.34375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 1868, "epoch": 1, "loss": 0.0003, "lr": 1.8949432717450418e-05, "grad_norm": 3.3125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1869, "epoch": 1, "loss": 0.0017, "lr": 1.8948095821289358e-05, "grad_norm": 6.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 1870, "epoch": 1, "loss": 0.0737, "lr": 1.89467581222587e-05, "grad_norm": 4.625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 1871, "epoch": 1, "loss": 0.2416, "lr": 1.894541962047848e-05, "grad_norm": 6.21875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 1872, "epoch": 1, "loss": 0.104, "lr": 1.8944080316068785e-05, "grad_norm": 2.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1873, "epoch": 1, "loss": 0.0714, "lr": 1.8942740209149785e-05, "grad_norm": 3.890625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 1874, "epoch": 1, "loss": 0.2384, "lr": 1.8941399299841723e-05, "grad_norm": 7.65625, "samples_per_sec": 8.82, "gpu_mem_gb": 6.83} {"step": 1875, "epoch": 1, "loss": 0.0014, "lr": 1.8940057588264918e-05, "grad_norm": 0.47265625, "samples_per_sec": 11.64, "gpu_mem_gb": 6.82} {"step": 1876, "epoch": 1, "loss": 0.0295, "lr": 1.8938715074539753e-05, "grad_norm": 2.484375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1877, "epoch": 1, "loss": 0.0084, "lr": 1.8937371758786676e-05, "grad_norm": 4.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1878, "epoch": 1, "loss": 0.0142, "lr": 1.8936027641126228e-05, "grad_norm": 2.6875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 1879, "epoch": 1, "loss": 0.0011, "lr": 1.8934682721679004e-05, "grad_norm": 8.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1880, "epoch": 1, "loss": 0.0628, "lr": 1.893333700056568e-05, "grad_norm": 2.109375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 1881, "epoch": 1, "loss": 0.0577, "lr": 1.8931990477906997e-05, "grad_norm": 2.265625, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1882, "epoch": 1, "loss": 0.0077, "lr": 1.8930643153823777e-05, "grad_norm": 6.34375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 1883, "epoch": 1, "loss": 0.0063, "lr": 1.8929295028436907e-05, "grad_norm": 4.75, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 1884, "epoch": 1, "loss": 0.0006, "lr": 1.8927946101867348e-05, "grad_norm": 6.90625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 1885, "epoch": 1, "loss": 0.2168, "lr": 1.892659637423613e-05, "grad_norm": 4.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1886, "epoch": 1, "loss": 0.0009, "lr": 1.892524584566436e-05, "grad_norm": 5.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1887, "epoch": 1, "loss": 0.1016, "lr": 1.8923894516273216e-05, "grad_norm": 4.6875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.85} {"step": 1888, "epoch": 1, "loss": 0.0013, "lr": 1.8922542386183942e-05, "grad_norm": 3.203125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 1889, "epoch": 1, "loss": 0.0015, "lr": 1.892118945551786e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 1890, "epoch": 1, "loss": 0.2407, "lr": 1.8919835724396363e-05, "grad_norm": 4.6875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 1891, "epoch": 1, "loss": 0.1903, "lr": 1.891848119294092e-05, "grad_norm": 4.9375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 1892, "epoch": 1, "loss": 0.0004, "lr": 1.8917125861273056e-05, "grad_norm": 4.4375, "samples_per_sec": 11.03, "gpu_mem_gb": 6.82} {"step": 1893, "epoch": 1, "loss": 0.0417, "lr": 1.8915769729514384e-05, "grad_norm": 4.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1894, "epoch": 1, "loss": 0.0034, "lr": 1.891441279778658e-05, "grad_norm": 5.1875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 1895, "epoch": 1, "loss": 0.0537, "lr": 1.89130550662114e-05, "grad_norm": 7.0625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 1896, "epoch": 1, "loss": 0.1363, "lr": 1.8911696534910664e-05, "grad_norm": 5.15625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 1897, "epoch": 1, "loss": 0.0601, "lr": 1.8910337204006265e-05, "grad_norm": 5.0625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1898, "epoch": 1, "loss": 0.0005, "lr": 1.8908977073620175e-05, "grad_norm": 4.21875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 1899, "epoch": 1, "loss": 0.0013, "lr": 1.8907616143874427e-05, "grad_norm": 5.28125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1900, "epoch": 1, "loss": 0.0089, "lr": 1.8906254414891127e-05, "grad_norm": 5.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1901, "epoch": 1, "loss": 0.1576, "lr": 1.8904891886792465e-05, "grad_norm": 4.96875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 1902, "epoch": 1, "loss": 0.0024, "lr": 1.890352855970069e-05, "grad_norm": 2.296875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1903, "epoch": 1, "loss": 0.0491, "lr": 1.8902164433738123e-05, "grad_norm": 7.3125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 1904, "epoch": 1, "loss": 0.0386, "lr": 1.8900799509027168e-05, "grad_norm": 5.59375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1905, "epoch": 1, "loss": 0.0007, "lr": 1.889943378569029e-05, "grad_norm": 0.92578125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1906, "epoch": 1, "loss": 0.1363, "lr": 1.8898067263850025e-05, "grad_norm": 10.1875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 1907, "epoch": 1, "loss": 0.0079, "lr": 1.889669994362899e-05, "grad_norm": 2.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1908, "epoch": 1, "loss": 0.0371, "lr": 1.889533182514986e-05, "grad_norm": 5.09375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1909, "epoch": 1, "loss": 0.0116, "lr": 1.88939629085354e-05, "grad_norm": 4.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 1910, "epoch": 1, "loss": 0.0764, "lr": 1.8892593193908434e-05, "grad_norm": 3.34375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 1911, "epoch": 1, "loss": 0.0124, "lr": 1.8891222681391853e-05, "grad_norm": 2.125, "samples_per_sec": 10.39, "gpu_mem_gb": 6.85} {"step": 1912, "epoch": 1, "loss": 0.0204, "lr": 1.888985137110863e-05, "grad_norm": 3.953125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 1913, "epoch": 1, "loss": 0.0791, "lr": 1.8888479263181813e-05, "grad_norm": 3.78125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1914, "epoch": 1, "loss": 0.012, "lr": 1.8887106357734506e-05, "grad_norm": 4.34375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 1915, "epoch": 1, "loss": 0.1662, "lr": 1.8885732654889898e-05, "grad_norm": 4.65625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 1916, "epoch": 1, "loss": 0.1711, "lr": 1.888435815477124e-05, "grad_norm": 7.34375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 1917, "epoch": 1, "loss": 0.0018, "lr": 1.888298285750186e-05, "grad_norm": 2.515625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1918, "epoch": 1, "loss": 0.0013, "lr": 1.888160676320516e-05, "grad_norm": 6.625, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 1919, "epoch": 1, "loss": 0.0002, "lr": 1.8880229872004616e-05, "grad_norm": 3.140625, "samples_per_sec": 10.35, "gpu_mem_gb": 6.82} {"step": 1920, "epoch": 1, "loss": 0.0004, "lr": 1.8878852184023754e-05, "grad_norm": 4.4375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 1921, "epoch": 1, "loss": 0.0063, "lr": 1.88774736993862e-05, "grad_norm": 3.46875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1922, "epoch": 1, "loss": 0.1507, "lr": 1.8876094418215634e-05, "grad_norm": 6.5, "samples_per_sec": 10.97, "gpu_mem_gb": 6.84} {"step": 1923, "epoch": 1, "loss": 0.2229, "lr": 1.8874714340635813e-05, "grad_norm": 9.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 1924, "epoch": 1, "loss": 0.1991, "lr": 1.8873333466770568e-05, "grad_norm": 5.9375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1925, "epoch": 1, "loss": 0.0536, "lr": 1.8871951796743792e-05, "grad_norm": 5.84375, "samples_per_sec": 10.39, "gpu_mem_gb": 6.82} {"step": 1926, "epoch": 1, "loss": 0.001, "lr": 1.887056933067946e-05, "grad_norm": 5.78125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 1927, "epoch": 1, "loss": 0.0732, "lr": 1.8869186068701613e-05, "grad_norm": 5.1875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 1928, "epoch": 1, "loss": 0.0173, "lr": 1.8867802010934363e-05, "grad_norm": 4.78125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 1929, "epoch": 1, "loss": 0.2731, "lr": 1.88664171575019e-05, "grad_norm": 7.9375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1930, "epoch": 1, "loss": 0.0012, "lr": 1.886503150852847e-05, "grad_norm": 3.4375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 1931, "epoch": 1, "loss": 0.2081, "lr": 1.8863645064138412e-05, "grad_norm": 4.59375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 1932, "epoch": 1, "loss": 0.0541, "lr": 1.886225782445612e-05, "grad_norm": 3.78125, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 1933, "epoch": 1, "loss": 0.0102, "lr": 1.886086978960606e-05, "grad_norm": 2.34375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1934, "epoch": 1, "loss": 0.037, "lr": 1.885948095971278e-05, "grad_norm": 1.90625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 1935, "epoch": 1, "loss": 0.1206, "lr": 1.8858091334900893e-05, "grad_norm": 4.40625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1936, "epoch": 1, "loss": 0.1951, "lr": 1.8856700915295084e-05, "grad_norm": 4.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1937, "epoch": 1, "loss": 0.085, "lr": 1.88553097010201e-05, "grad_norm": 9.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 1938, "epoch": 1, "loss": 0.0469, "lr": 1.885391769220078e-05, "grad_norm": 4.3125, "samples_per_sec": 10.25, "gpu_mem_gb": 6.83} {"step": 1939, "epoch": 1, "loss": 0.1725, "lr": 1.885252488896201e-05, "grad_norm": 5.03125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 1940, "epoch": 1, "loss": 0.01, "lr": 1.8851131291428768e-05, "grad_norm": 3.8125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1941, "epoch": 1, "loss": 0.0391, "lr": 1.8849736899726093e-05, "grad_norm": 6.0625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1942, "epoch": 1, "loss": 0.0007, "lr": 1.8848341713979097e-05, "grad_norm": 3.40625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1943, "epoch": 1, "loss": 0.0019, "lr": 1.884694573431296e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 1944, "epoch": 1, "loss": 0.0449, "lr": 1.8845548960852948e-05, "grad_norm": 5.0625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 1945, "epoch": 1, "loss": 0.0033, "lr": 1.884415139372437e-05, "grad_norm": 4.90625, "samples_per_sec": 10.35, "gpu_mem_gb": 6.85} {"step": 1946, "epoch": 1, "loss": 0.1478, "lr": 1.8842753033052635e-05, "grad_norm": 5.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 1947, "epoch": 1, "loss": 0.0542, "lr": 1.8841353878963203e-05, "grad_norm": 4.5625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 1948, "epoch": 1, "loss": 0.0461, "lr": 1.8839953931581622e-05, "grad_norm": 5.5625, "samples_per_sec": 10.34, "gpu_mem_gb": 6.85} {"step": 1949, "epoch": 1, "loss": 0.2534, "lr": 1.8838553191033496e-05, "grad_norm": 5.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 1950, "epoch": 1, "loss": 0.0004, "lr": 1.8837151657444513e-05, "grad_norm": 6.15625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 1951, "epoch": 1, "loss": 0.0975, "lr": 1.8835749330940418e-05, "grad_norm": 6.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 1952, "epoch": 1, "loss": 0.0501, "lr": 1.883434621164704e-05, "grad_norm": 3.65625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1953, "epoch": 1, "loss": 0.0008, "lr": 1.8832942299690274e-05, "grad_norm": 2.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 1954, "epoch": 1, "loss": 0.0089, "lr": 1.8831537595196085e-05, "grad_norm": 4.6875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 1955, "epoch": 1, "loss": 0.0597, "lr": 1.8830132098290508e-05, "grad_norm": 5.5625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.84} {"step": 1956, "epoch": 1, "loss": 0.0003, "lr": 1.8828725809099657e-05, "grad_norm": 4.375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 1957, "epoch": 1, "loss": 0.0381, "lr": 1.882731872774971e-05, "grad_norm": 8.0625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 1958, "epoch": 1, "loss": 0.2403, "lr": 1.8825910854366914e-05, "grad_norm": 7.46875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 1959, "epoch": 1, "loss": 0.0006, "lr": 1.8824502189077593e-05, "grad_norm": 1.46875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 1960, "epoch": 1, "loss": 0.2369, "lr": 1.882309273200814e-05, "grad_norm": 7.0, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1961, "epoch": 1, "loss": 0.0063, "lr": 1.882168248328502e-05, "grad_norm": 0.44921875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 1962, "epoch": 1, "loss": 0.0013, "lr": 1.8820271443034766e-05, "grad_norm": 3.203125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 1963, "epoch": 1, "loss": 0.0109, "lr": 1.8818859611383985e-05, "grad_norm": 3.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 1964, "epoch": 1, "loss": 0.1225, "lr": 1.8817446988459348e-05, "grad_norm": 4.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1965, "epoch": 1, "loss": 0.0051, "lr": 1.8816033574387615e-05, "grad_norm": 0.68359375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 1966, "epoch": 1, "loss": 0.5092, "lr": 1.8814619369295597e-05, "grad_norm": 8.75, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1967, "epoch": 1, "loss": 0.1317, "lr": 1.881320437331018e-05, "grad_norm": 8.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1968, "epoch": 1, "loss": 0.0008, "lr": 1.8811788586558334e-05, "grad_norm": 3.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 1969, "epoch": 1, "loss": 0.0008, "lr": 1.8810372009167084e-05, "grad_norm": 5.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1970, "epoch": 1, "loss": 0.0461, "lr": 1.8808954641263538e-05, "grad_norm": 5.375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1971, "epoch": 1, "loss": 0.1646, "lr": 1.880753648297486e-05, "grad_norm": 5.25, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 1972, "epoch": 1, "loss": 0.0293, "lr": 1.880611753442831e-05, "grad_norm": 3.046875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.85} {"step": 1973, "epoch": 1, "loss": 0.0014, "lr": 1.880469779575119e-05, "grad_norm": 3.609375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1974, "epoch": 1, "loss": 0.001, "lr": 1.8803277267070888e-05, "grad_norm": 2.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1975, "epoch": 1, "loss": 0.0047, "lr": 1.8801855948514868e-05, "grad_norm": 2.546875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 1976, "epoch": 1, "loss": 0.1272, "lr": 1.880043384021065e-05, "grad_norm": 3.390625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 1977, "epoch": 1, "loss": 0.0009, "lr": 1.879901094228584e-05, "grad_norm": 6.8125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 1978, "epoch": 1, "loss": 0.0387, "lr": 1.8797587254868106e-05, "grad_norm": 6.1875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 1979, "epoch": 1, "loss": 0.0008, "lr": 1.8796162778085184e-05, "grad_norm": 13.125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 1980, "epoch": 1, "loss": 0.1603, "lr": 1.879473751206489e-05, "grad_norm": 6.59375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 1981, "epoch": 1, "loss": 0.0027, "lr": 1.8793311456935107e-05, "grad_norm": 6.46875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 1982, "epoch": 1, "loss": 0.1221, "lr": 1.879188461282379e-05, "grad_norm": 4.34375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 1983, "epoch": 1, "loss": 0.0009, "lr": 1.879045697985895e-05, "grad_norm": 3.015625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 1984, "epoch": 1, "loss": 0.0081, "lr": 1.8789028558168697e-05, "grad_norm": 4.21875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 1985, "epoch": 1, "loss": 0.2168, "lr": 1.878759934788119e-05, "grad_norm": 6.4375, "samples_per_sec": 10.32, "gpu_mem_gb": 6.83} {"step": 1986, "epoch": 1, "loss": 0.0446, "lr": 1.8786169349124658e-05, "grad_norm": 4.71875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 1987, "epoch": 1, "loss": 0.0107, "lr": 1.878473856202742e-05, "grad_norm": 6.09375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 1988, "epoch": 1, "loss": 0.1943, "lr": 1.8783306986717852e-05, "grad_norm": 3.125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 1989, "epoch": 1, "loss": 0.0041, "lr": 1.8781874623324396e-05, "grad_norm": 7.8125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 1990, "epoch": 1, "loss": 0.3579, "lr": 1.8780441471975574e-05, "grad_norm": 6.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1991, "epoch": 1, "loss": 0.3541, "lr": 1.877900753279998e-05, "grad_norm": 6.28125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 1992, "epoch": 1, "loss": 0.0008, "lr": 1.8777572805926267e-05, "grad_norm": 2.296875, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 1993, "epoch": 1, "loss": 0.0016, "lr": 1.877613729148317e-05, "grad_norm": 1.9765625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 1994, "epoch": 1, "loss": 0.0624, "lr": 1.877470098959949e-05, "grad_norm": 5.03125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 1995, "epoch": 1, "loss": 0.0046, "lr": 1.87732639004041e-05, "grad_norm": 2.578125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 1996, "epoch": 1, "loss": 0.0801, "lr": 1.8771826024025944e-05, "grad_norm": 3.15625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 1997, "epoch": 1, "loss": 0.0045, "lr": 1.8770387360594035e-05, "grad_norm": 5.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 1998, "epoch": 1, "loss": 0.2671, "lr": 1.8768947910237452e-05, "grad_norm": 6.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 1999, "epoch": 1, "loss": 0.039, "lr": 1.8767507673085362e-05, "grad_norm": 2.234375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 2000, "epoch": 1, "loss": 0.2718, "lr": 1.876606664926698e-05, "grad_norm": 7.09375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2001, "epoch": 1, "loss": 0.0027, "lr": 1.8764624838911608e-05, "grad_norm": 1.9921875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2002, "epoch": 1, "loss": 0.0408, "lr": 1.876318224214861e-05, "grad_norm": 8.1875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 2003, "epoch": 1, "loss": 0.3435, "lr": 1.8761738859107423e-05, "grad_norm": 5.9375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2004, "epoch": 1, "loss": 0.0097, "lr": 1.8760294689917556e-05, "grad_norm": 4.65625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 2005, "epoch": 1, "loss": 0.0128, "lr": 1.8758849734708587e-05, "grad_norm": 3.453125, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 2006, "epoch": 1, "loss": 0.0867, "lr": 1.8757403993610165e-05, "grad_norm": 4.9375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2007, "epoch": 1, "loss": 0.0372, "lr": 1.8755957466752007e-05, "grad_norm": 4.4375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2008, "epoch": 1, "loss": 0.0798, "lr": 1.8754510154263906e-05, "grad_norm": 4.15625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2009, "epoch": 1, "loss": 0.1891, "lr": 1.875306205627572e-05, "grad_norm": 7.03125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 2010, "epoch": 1, "loss": 0.0023, "lr": 1.8751613172917386e-05, "grad_norm": 3.953125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2011, "epoch": 1, "loss": 0.0016, "lr": 1.8750163504318898e-05, "grad_norm": 11.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2012, "epoch": 1, "loss": 0.0404, "lr": 1.8748713050610328e-05, "grad_norm": 3.328125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 2013, "epoch": 1, "loss": 0.0022, "lr": 1.8747261811921825e-05, "grad_norm": 2.8125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2014, "epoch": 1, "loss": 0.0011, "lr": 1.8745809788383594e-05, "grad_norm": 5.03125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2015, "epoch": 1, "loss": 0.074, "lr": 1.8744356980125922e-05, "grad_norm": 3.828125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2016, "epoch": 1, "loss": 0.0012, "lr": 1.8742903387279165e-05, "grad_norm": 4.65625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2017, "epoch": 1, "loss": 0.0014, "lr": 1.874144900997374e-05, "grad_norm": 8.5625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2018, "epoch": 1, "loss": 0.112, "lr": 1.8739993848340147e-05, "grad_norm": 7.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2019, "epoch": 1, "loss": 0.0226, "lr": 1.873853790250895e-05, "grad_norm": 5.9375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2020, "epoch": 1, "loss": 0.0599, "lr": 1.873708117261078e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2021, "epoch": 1, "loss": 0.1423, "lr": 1.8735623658776348e-05, "grad_norm": 17.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2022, "epoch": 1, "loss": 0.0542, "lr": 1.8734165361136422e-05, "grad_norm": 5.28125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 2023, "epoch": 1, "loss": 0.0026, "lr": 1.8732706279821858e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2024, "epoch": 1, "loss": 0.009, "lr": 1.8731246414963566e-05, "grad_norm": 7.53125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2025, "epoch": 1, "loss": 0.0037, "lr": 1.8729785766692533e-05, "grad_norm": 6.28125, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 2026, "epoch": 1, "loss": 0.0007, "lr": 1.8728324335139814e-05, "grad_norm": 3.390625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2027, "epoch": 1, "loss": 0.2333, "lr": 1.8726862120436546e-05, "grad_norm": 5.15625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2028, "epoch": 1, "loss": 0.1696, "lr": 1.8725399122713914e-05, "grad_norm": 4.0, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2029, "epoch": 1, "loss": 0.069, "lr": 1.8723935342103195e-05, "grad_norm": 5.0625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 2030, "epoch": 1, "loss": 0.0389, "lr": 1.872247077873572e-05, "grad_norm": 4.4375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 2031, "epoch": 1, "loss": 0.1331, "lr": 1.87210054327429e-05, "grad_norm": 4.9375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2032, "epoch": 1, "loss": 0.0277, "lr": 1.8719539304256216e-05, "grad_norm": 2.75, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 2033, "epoch": 1, "loss": 0.0808, "lr": 1.8718072393407217e-05, "grad_norm": 7.53125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2034, "epoch": 1, "loss": 0.0009, "lr": 1.8716604700327516e-05, "grad_norm": 3.703125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2035, "epoch": 1, "loss": 0.0258, "lr": 1.8715136225148806e-05, "grad_norm": 10.125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2036, "epoch": 1, "loss": 0.1632, "lr": 1.8713666968002842e-05, "grad_norm": 8.3125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 2037, "epoch": 1, "loss": 0.0002, "lr": 1.8712196929021463e-05, "grad_norm": 3.984375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2038, "epoch": 1, "loss": 0.0017, "lr": 1.8710726108336562e-05, "grad_norm": 1.171875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2039, "epoch": 1, "loss": 0.1186, "lr": 1.8709254506080107e-05, "grad_norm": 5.78125, "samples_per_sec": 10.27, "gpu_mem_gb": 6.83} {"step": 2040, "epoch": 1, "loss": 0.0162, "lr": 1.870778212238414e-05, "grad_norm": 6.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2041, "epoch": 1, "loss": 0.0389, "lr": 1.8706308957380774e-05, "grad_norm": 6.84375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2042, "epoch": 1, "loss": 0.0058, "lr": 1.870483501120218e-05, "grad_norm": 4.6875, "samples_per_sec": 10.35, "gpu_mem_gb": 6.82} {"step": 2043, "epoch": 1, "loss": 0.0746, "lr": 1.8703360283980618e-05, "grad_norm": 3.078125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2044, "epoch": 1, "loss": 0.3439, "lr": 1.8701884775848404e-05, "grad_norm": 6.28125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2045, "epoch": 1, "loss": 0.0021, "lr": 1.870040848693793e-05, "grad_norm": 5.59375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 2046, "epoch": 1, "loss": 0.0255, "lr": 1.8698931417381654e-05, "grad_norm": 5.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2047, "epoch": 1, "loss": 0.0015, "lr": 1.8697453567312104e-05, "grad_norm": 2.15625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 2048, "epoch": 1, "loss": 0.0056, "lr": 1.8695974936861886e-05, "grad_norm": 1.125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 2049, "epoch": 1, "loss": 0.1159, "lr": 1.869449552616367e-05, "grad_norm": 6.15625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2050, "epoch": 1, "loss": 0.004, "lr": 1.8693015335350193e-05, "grad_norm": 5.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2051, "epoch": 1, "loss": 0.1595, "lr": 1.8691534364554265e-05, "grad_norm": 6.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2052, "epoch": 1, "loss": 0.0411, "lr": 1.869005261390877e-05, "grad_norm": 4.03125, "samples_per_sec": 10.39, "gpu_mem_gb": 6.82} {"step": 2053, "epoch": 1, "loss": 0.0051, "lr": 1.8688570083546658e-05, "grad_norm": 4.375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 2054, "epoch": 1, "loss": 0.0869, "lr": 1.8687086773600945e-05, "grad_norm": 2.953125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2055, "epoch": 1, "loss": 0.1739, "lr": 1.8685602684204722e-05, "grad_norm": 10.0625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2056, "epoch": 1, "loss": 0.001, "lr": 1.8684117815491154e-05, "grad_norm": 2.59375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2057, "epoch": 1, "loss": 0.0399, "lr": 1.868263216759347e-05, "grad_norm": 3.71875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.84} {"step": 2058, "epoch": 1, "loss": 0.0561, "lr": 1.8681145740644965e-05, "grad_norm": 5.84375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 2059, "epoch": 1, "loss": 0.0038, "lr": 1.8679658534779017e-05, "grad_norm": 3.984375, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 2060, "epoch": 1, "loss": 0.2296, "lr": 1.8678170550129056e-05, "grad_norm": 4.46875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2061, "epoch": 1, "loss": 0.1633, "lr": 1.8676681786828598e-05, "grad_norm": 4.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.84} {"step": 2062, "epoch": 1, "loss": 0.0471, "lr": 1.8675192245011224e-05, "grad_norm": 2.234375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2063, "epoch": 1, "loss": 0.1341, "lr": 1.8673701924810582e-05, "grad_norm": 9.125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2064, "epoch": 1, "loss": 0.0066, "lr": 1.8672210826360388e-05, "grad_norm": 3.734375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2065, "epoch": 1, "loss": 0.0006, "lr": 1.8670718949794432e-05, "grad_norm": 1.484375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 2066, "epoch": 1, "loss": 0.0035, "lr": 1.8669226295246576e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.45, "gpu_mem_gb": 6.85} {"step": 2067, "epoch": 1, "loss": 0.0445, "lr": 1.8667732862850744e-05, "grad_norm": 5.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 2068, "epoch": 1, "loss": 0.0919, "lr": 1.866623865274094e-05, "grad_norm": 3.84375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2069, "epoch": 1, "loss": 0.1756, "lr": 1.866474366505123e-05, "grad_norm": 5.9375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2070, "epoch": 1, "loss": 0.0011, "lr": 1.866324789991575e-05, "grad_norm": 4.8125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 2071, "epoch": 1, "loss": 0.1197, "lr": 1.8661751357468708e-05, "grad_norm": 7.78125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2072, "epoch": 1, "loss": 0.1257, "lr": 1.866025403784439e-05, "grad_norm": 3.171875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 2073, "epoch": 1, "loss": 0.2598, "lr": 1.8658755941177127e-05, "grad_norm": 4.65625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2074, "epoch": 1, "loss": 0.0203, "lr": 1.8657257067601355e-05, "grad_norm": 4.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2075, "epoch": 1, "loss": 0.1177, "lr": 1.8655757417251544e-05, "grad_norm": 4.15625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2076, "epoch": 1, "loss": 0.0231, "lr": 1.865425699026226e-05, "grad_norm": 4.5, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 2077, "epoch": 1, "loss": 0.0947, "lr": 1.8652755786768127e-05, "grad_norm": 13.0, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2078, "epoch": 1, "loss": 0.0112, "lr": 1.865125380690384e-05, "grad_norm": 2.375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2079, "epoch": 1, "loss": 0.017, "lr": 1.8649751050804164e-05, "grad_norm": 7.0625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 2080, "epoch": 1, "loss": 0.0025, "lr": 1.8648247518603936e-05, "grad_norm": 5.65625, "samples_per_sec": 10.29, "gpu_mem_gb": 6.83} {"step": 2081, "epoch": 1, "loss": 0.1329, "lr": 1.8646743210438058e-05, "grad_norm": 6.78125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2082, "epoch": 1, "loss": 0.0012, "lr": 1.864523812644151e-05, "grad_norm": 1.0390625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2083, "epoch": 1, "loss": 0.0017, "lr": 1.8643732266749327e-05, "grad_norm": 5.46875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2084, "epoch": 1, "loss": 0.1169, "lr": 1.864222563149663e-05, "grad_norm": 5.6875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 2085, "epoch": 1, "loss": 0.0005, "lr": 1.8640718220818598e-05, "grad_norm": 3.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2086, "epoch": 1, "loss": 0.0079, "lr": 1.8639210034850486e-05, "grad_norm": 3.359375, "samples_per_sec": 10.37, "gpu_mem_gb": 6.83} {"step": 2087, "epoch": 1, "loss": 0.0015, "lr": 1.8637701073727617e-05, "grad_norm": 5.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2088, "epoch": 1, "loss": 0.0004, "lr": 1.863619133758538e-05, "grad_norm": 6.6875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2089, "epoch": 1, "loss": 0.0011, "lr": 1.8634680826559238e-05, "grad_norm": 1.2109375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2090, "epoch": 1, "loss": 0.0049, "lr": 1.863316954078472e-05, "grad_norm": 6.9375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 2091, "epoch": 1, "loss": 0.1716, "lr": 1.863165748039743e-05, "grad_norm": 4.75, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2092, "epoch": 1, "loss": 0.0047, "lr": 1.863014464553303e-05, "grad_norm": 3.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2093, "epoch": 1, "loss": 0.0012, "lr": 1.8628631036327273e-05, "grad_norm": 1.5859375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 2094, "epoch": 1, "loss": 0.0843, "lr": 1.8627116652915957e-05, "grad_norm": 2.5625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2095, "epoch": 1, "loss": 0.0267, "lr": 1.8625601495434968e-05, "grad_norm": 6.53125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 2096, "epoch": 1, "loss": 0.0778, "lr": 1.8624085564020245e-05, "grad_norm": 5.59375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 2097, "epoch": 1, "loss": 0.0232, "lr": 1.862256885880781e-05, "grad_norm": 2.5625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 2098, "epoch": 1, "loss": 0.0119, "lr": 1.8621051379933748e-05, "grad_norm": 7.84375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 2099, "epoch": 1, "loss": 0.0013, "lr": 1.8619533127534218e-05, "grad_norm": 2.671875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 2100, "epoch": 1, "loss": 0.0023, "lr": 1.8618014101745444e-05, "grad_norm": 4.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 2101, "epoch": 1, "loss": 0.2837, "lr": 1.8616494302703718e-05, "grad_norm": 9.625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 2102, "epoch": 1, "loss": 0.0016, "lr": 1.8614973730545412e-05, "grad_norm": 3.359375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2103, "epoch": 1, "loss": 0.2748, "lr": 1.861345238540695e-05, "grad_norm": 7.53125, "samples_per_sec": 8.84, "gpu_mem_gb": 6.83} {"step": 2104, "epoch": 1, "loss": 0.0057, "lr": 1.861193026742484e-05, "grad_norm": 3.796875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2105, "epoch": 1, "loss": 0.0499, "lr": 1.8610407376735657e-05, "grad_norm": 3.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2106, "epoch": 1, "loss": 0.0013, "lr": 1.8608883713476035e-05, "grad_norm": 11.6875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 2107, "epoch": 1, "loss": 0.002, "lr": 1.8607359277782692e-05, "grad_norm": 1.921875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2108, "epoch": 1, "loss": 0.2492, "lr": 1.8605834069792405e-05, "grad_norm": 4.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2109, "epoch": 1, "loss": 0.0005, "lr": 1.8604308089642025e-05, "grad_norm": 2.078125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2110, "epoch": 1, "loss": 0.1267, "lr": 1.8602781337468472e-05, "grad_norm": 8.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2111, "epoch": 1, "loss": 0.0217, "lr": 1.860125381340873e-05, "grad_norm": 4.34375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2112, "epoch": 1, "loss": 0.0006, "lr": 1.8599725517599858e-05, "grad_norm": 1.6484375, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 2113, "epoch": 1, "loss": 0.1192, "lr": 1.8598196450178984e-05, "grad_norm": 8.0, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 2114, "epoch": 1, "loss": 0.0013, "lr": 1.8596666611283303e-05, "grad_norm": 2.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2115, "epoch": 1, "loss": 0.0179, "lr": 1.8595136001050083e-05, "grad_norm": 2.59375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2116, "epoch": 1, "loss": 0.0012, "lr": 1.859360461961665e-05, "grad_norm": 1.0625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 2117, "epoch": 1, "loss": 0.0264, "lr": 1.8592072467120416e-05, "grad_norm": 7.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2118, "epoch": 1, "loss": 0.1019, "lr": 1.8590539543698852e-05, "grad_norm": 5.9375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2119, "epoch": 1, "loss": 0.0034, "lr": 1.85890058494895e-05, "grad_norm": 3.125, "samples_per_sec": 10.34, "gpu_mem_gb": 6.82} {"step": 2120, "epoch": 1, "loss": 0.003, "lr": 1.8587471384629965e-05, "grad_norm": 1.828125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 2121, "epoch": 1, "loss": 0.0002, "lr": 1.8585936149257935e-05, "grad_norm": 2.9375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2122, "epoch": 1, "loss": 0.0012, "lr": 1.8584400143511154e-05, "grad_norm": 3.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2123, "epoch": 1, "loss": 0.0008, "lr": 1.8582863367527447e-05, "grad_norm": 6.46875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2124, "epoch": 1, "loss": 0.2932, "lr": 1.858132582144469e-05, "grad_norm": 4.53125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2125, "epoch": 1, "loss": 0.0567, "lr": 1.8579787505400852e-05, "grad_norm": 6.90625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2126, "epoch": 1, "loss": 0.0693, "lr": 1.857824841953395e-05, "grad_norm": 10.875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 2127, "epoch": 1, "loss": 0.0037, "lr": 1.8576708563982085e-05, "grad_norm": 3.140625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 2128, "epoch": 1, "loss": 0.0021, "lr": 1.8575167938883412e-05, "grad_norm": 5.1875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2129, "epoch": 1, "loss": 0.0028, "lr": 1.857362654437618e-05, "grad_norm": 2.140625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 2130, "epoch": 1, "loss": 0.0016, "lr": 1.857208438059867e-05, "grad_norm": 0.3515625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 2131, "epoch": 1, "loss": 0.0902, "lr": 1.857054144768927e-05, "grad_norm": 5.59375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2132, "epoch": 1, "loss": 0.0362, "lr": 1.856899774578641e-05, "grad_norm": 4.5, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 2133, "epoch": 1, "loss": 0.0005, "lr": 1.8567453275028606e-05, "grad_norm": 1.4765625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 2134, "epoch": 1, "loss": 0.0751, "lr": 1.8565908035554428e-05, "grad_norm": 5.53125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2135, "epoch": 1, "loss": 0.0008, "lr": 1.856436202750253e-05, "grad_norm": 0.76953125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2136, "epoch": 1, "loss": 0.0123, "lr": 1.8562815251011626e-05, "grad_norm": 1.0703125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2137, "epoch": 1, "loss": 0.1243, "lr": 1.85612677062205e-05, "grad_norm": 3.734375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 2138, "epoch": 1, "loss": 0.0326, "lr": 1.8559719393268002e-05, "grad_norm": 2.203125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 2139, "epoch": 1, "loss": 0.1129, "lr": 1.8558170312293064e-05, "grad_norm": 11.9375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2140, "epoch": 1, "loss": 0.0004, "lr": 1.855662046343467e-05, "grad_norm": 3.796875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 2141, "epoch": 1, "loss": 0.0003, "lr": 1.8555069846831882e-05, "grad_norm": 0.13671875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2142, "epoch": 1, "loss": 0.1316, "lr": 1.8553518462623826e-05, "grad_norm": 5.59375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2143, "epoch": 1, "loss": 0.0016, "lr": 1.855196631094971e-05, "grad_norm": 5.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2144, "epoch": 1, "loss": 0.0018, "lr": 1.8550413391948792e-05, "grad_norm": 4.46875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2145, "epoch": 1, "loss": 0.2253, "lr": 1.8548859705760414e-05, "grad_norm": 8.125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 2146, "epoch": 1, "loss": 0.0035, "lr": 1.8547305252523976e-05, "grad_norm": 4.46875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 2147, "epoch": 1, "loss": 0.0016, "lr": 1.854575003237895e-05, "grad_norm": 7.9375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2148, "epoch": 1, "loss": 0.0439, "lr": 1.8544194045464888e-05, "grad_norm": 7.59375, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 2149, "epoch": 1, "loss": 0.028, "lr": 1.8542637291921393e-05, "grad_norm": 2.734375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2150, "epoch": 1, "loss": 0.197, "lr": 1.8541079771888145e-05, "grad_norm": 6.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2151, "epoch": 1, "loss": 0.0267, "lr": 1.8539521485504898e-05, "grad_norm": 7.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2152, "epoch": 1, "loss": 0.0329, "lr": 1.8537962432911465e-05, "grad_norm": 2.390625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 2153, "epoch": 1, "loss": 0.0012, "lr": 1.8536402614247734e-05, "grad_norm": 2.34375, "samples_per_sec": 10.39, "gpu_mem_gb": 6.83} {"step": 2154, "epoch": 1, "loss": 0.0664, "lr": 1.853484202965366e-05, "grad_norm": 4.5625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2155, "epoch": 1, "loss": 0.0541, "lr": 1.853328067926927e-05, "grad_norm": 4.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2156, "epoch": 1, "loss": 0.0017, "lr": 1.8531718563234646e-05, "grad_norm": 3.984375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 2157, "epoch": 1, "loss": 0.0089, "lr": 1.8530155681689962e-05, "grad_norm": 2.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2158, "epoch": 1, "loss": 0.2432, "lr": 1.8528592034775438e-05, "grad_norm": 5.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2159, "epoch": 1, "loss": 0.0206, "lr": 1.852702762263138e-05, "grad_norm": 2.734375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2160, "epoch": 1, "loss": 0.001, "lr": 1.8525462445398147e-05, "grad_norm": 3.78125, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 2161, "epoch": 1, "loss": 0.2557, "lr": 1.8523896503216185e-05, "grad_norm": 6.40625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2162, "epoch": 1, "loss": 0.0895, "lr": 1.852232979622599e-05, "grad_norm": 9.4375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2163, "epoch": 1, "loss": 0.0302, "lr": 1.8520762324568138e-05, "grad_norm": 2.90625, "samples_per_sec": 10.38, "gpu_mem_gb": 6.83} {"step": 2164, "epoch": 1, "loss": 0.324, "lr": 1.851919408838327e-05, "grad_norm": 13.1875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2165, "epoch": 1, "loss": 0.0237, "lr": 1.85176250878121e-05, "grad_norm": 5.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2166, "epoch": 1, "loss": 0.072, "lr": 1.8516055322995402e-05, "grad_norm": 10.875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2167, "epoch": 1, "loss": 0.001, "lr": 1.8514484794074028e-05, "grad_norm": 4.1875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.85} {"step": 2168, "epoch": 1, "loss": 0.0443, "lr": 1.851291350118889e-05, "grad_norm": 7.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2169, "epoch": 1, "loss": 0.0399, "lr": 1.851134144448097e-05, "grad_norm": 9.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2170, "epoch": 1, "loss": 0.0679, "lr": 1.850976862409133e-05, "grad_norm": 4.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2171, "epoch": 1, "loss": 0.3099, "lr": 1.8508195040161084e-05, "grad_norm": 5.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2172, "epoch": 1, "loss": 0.04, "lr": 1.8506620692831427e-05, "grad_norm": 9.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 2173, "epoch": 1, "loss": 0.0012, "lr": 1.8505045582243614e-05, "grad_norm": 3.75, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83} {"step": 2174, "epoch": 1, "loss": 0.0204, "lr": 1.8503469708538976e-05, "grad_norm": 6.46875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2175, "epoch": 1, "loss": 0.1175, "lr": 1.85018930718589e-05, "grad_norm": 4.46875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2176, "epoch": 1, "loss": 0.3617, "lr": 1.8500315672344864e-05, "grad_norm": 5.125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2177, "epoch": 1, "loss": 0.0298, "lr": 1.849873751013839e-05, "grad_norm": 2.96875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2178, "epoch": 1, "loss": 0.0018, "lr": 1.8497158585381076e-05, "grad_norm": 2.25, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2179, "epoch": 1, "loss": 0.0008, "lr": 1.84955788982146e-05, "grad_norm": 2.640625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2180, "epoch": 1, "loss": 0.003, "lr": 1.84939984487807e-05, "grad_norm": 1.171875, "samples_per_sec": 10.39, "gpu_mem_gb": 6.83} {"step": 2181, "epoch": 1, "loss": 0.2535, "lr": 1.8492417237221178e-05, "grad_norm": 5.03125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 2182, "epoch": 1, "loss": 0.0161, "lr": 1.8490835263677906e-05, "grad_norm": 5.46875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2183, "epoch": 1, "loss": 0.0058, "lr": 1.8489252528292828e-05, "grad_norm": 3.765625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2184, "epoch": 1, "loss": 0.1048, "lr": 1.848766903120796e-05, "grad_norm": 5.4375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2185, "epoch": 1, "loss": 0.0003, "lr": 1.848608477256538e-05, "grad_norm": 5.375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 2186, "epoch": 1, "loss": 0.1582, "lr": 1.8484499752507234e-05, "grad_norm": 9.8125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2187, "epoch": 1, "loss": 0.0218, "lr": 1.8482913971175737e-05, "grad_norm": 3.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2188, "epoch": 1, "loss": 0.0131, "lr": 1.848132742871318e-05, "grad_norm": 0.76953125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 2189, "epoch": 1, "loss": 0.1105, "lr": 1.8479740125261907e-05, "grad_norm": 3.671875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 2190, "epoch": 1, "loss": 0.0389, "lr": 1.8478152060964343e-05, "grad_norm": 3.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2191, "epoch": 1, "loss": 0.0037, "lr": 1.847656323596298e-05, "grad_norm": 6.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2192, "epoch": 1, "loss": 0.1292, "lr": 1.8474973650400376e-05, "grad_norm": 7.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 2193, "epoch": 1, "loss": 0.072, "lr": 1.847338330441915e-05, "grad_norm": 3.15625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2194, "epoch": 1, "loss": 0.0345, "lr": 1.8471792198162002e-05, "grad_norm": 6.0625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 2195, "epoch": 1, "loss": 0.1395, "lr": 1.8470200331771694e-05, "grad_norm": 4.53125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2196, "epoch": 1, "loss": 0.0415, "lr": 1.846860770539105e-05, "grad_norm": 2.53125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2197, "epoch": 1, "loss": 0.0461, "lr": 1.846701431916298e-05, "grad_norm": 3.8125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2198, "epoch": 1, "loss": 0.0022, "lr": 1.846542017323044e-05, "grad_norm": 5.5625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 2199, "epoch": 1, "loss": 0.0727, "lr": 1.8463825267736472e-05, "grad_norm": 6.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2200, "epoch": 1, "loss": 0.0136, "lr": 1.8462229602824177e-05, "grad_norm": 0.51171875, "samples_per_sec": 10.32, "gpu_mem_gb": 6.83} {"step": 2201, "epoch": 1, "loss": 0.0392, "lr": 1.846063317863673e-05, "grad_norm": 3.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 2202, "epoch": 1, "loss": 0.0004, "lr": 1.8459035995317364e-05, "grad_norm": 3.03125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2203, "epoch": 1, "loss": 0.007, "lr": 1.8457438053009386e-05, "grad_norm": 1.734375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2204, "epoch": 1, "loss": 0.0366, "lr": 1.845583935185618e-05, "grad_norm": 3.3125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2205, "epoch": 1, "loss": 0.1917, "lr": 1.845423989200118e-05, "grad_norm": 5.03125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2206, "epoch": 1, "loss": 0.2527, "lr": 1.8452639673587908e-05, "grad_norm": 4.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2207, "epoch": 1, "loss": 0.0392, "lr": 1.845103869675993e-05, "grad_norm": 2.671875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.84} {"step": 2208, "epoch": 1, "loss": 0.0598, "lr": 1.8449436961660908e-05, "grad_norm": 3.046875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2209, "epoch": 1, "loss": 0.0009, "lr": 1.8447834468434556e-05, "grad_norm": 4.40625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 2210, "epoch": 1, "loss": 0.0035, "lr": 1.844623121722465e-05, "grad_norm": 2.515625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2211, "epoch": 1, "loss": 0.0035, "lr": 1.844462720817505e-05, "grad_norm": 4.375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 2212, "epoch": 1, "loss": 0.0002, "lr": 1.844302244142967e-05, "grad_norm": 4.28125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 2213, "epoch": 1, "loss": 0.0985, "lr": 1.84414169171325e-05, "grad_norm": 5.28125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2214, "epoch": 1, "loss": 0.1139, "lr": 1.8439810635427596e-05, "grad_norm": 5.125, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 2215, "epoch": 1, "loss": 0.0009, "lr": 1.8438203596459084e-05, "grad_norm": 3.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2216, "epoch": 1, "loss": 0.0009, "lr": 1.8436595800371154e-05, "grad_norm": 3.609375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2217, "epoch": 1, "loss": 0.1708, "lr": 1.843498724730807e-05, "grad_norm": 6.8125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2218, "epoch": 1, "loss": 0.0016, "lr": 1.843337793741415e-05, "grad_norm": 4.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2219, "epoch": 1, "loss": 0.1231, "lr": 1.8431767870833796e-05, "grad_norm": 4.375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2220, "epoch": 1, "loss": 0.0031, "lr": 1.8430157047711473e-05, "grad_norm": 0.8515625, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 2221, "epoch": 1, "loss": 0.0041, "lr": 1.8428545468191713e-05, "grad_norm": 4.75, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2222, "epoch": 1, "loss": 0.016, "lr": 1.842693313241911e-05, "grad_norm": 3.96875, "samples_per_sec": 11.01, "gpu_mem_gb": 6.82} {"step": 2223, "epoch": 1, "loss": 0.0003, "lr": 1.8425320040538332e-05, "grad_norm": 4.40625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2224, "epoch": 1, "loss": 0.2661, "lr": 1.8423706192694118e-05, "grad_norm": 5.9375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2225, "epoch": 1, "loss": 0.0021, "lr": 1.8422091589031268e-05, "grad_norm": 3.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2226, "epoch": 1, "loss": 0.0559, "lr": 1.8420476229694656e-05, "grad_norm": 2.203125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2227, "epoch": 1, "loss": 0.0687, "lr": 1.8418860114829212e-05, "grad_norm": 4.875, "samples_per_sec": 10.53, "gpu_mem_gb": 6.82} {"step": 2228, "epoch": 1, "loss": 0.0433, "lr": 1.8417243244579952e-05, "grad_norm": 2.71875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2229, "epoch": 1, "loss": 0.0341, "lr": 1.8415625619091945e-05, "grad_norm": 4.0, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2230, "epoch": 1, "loss": 0.001, "lr": 1.8414007238510332e-05, "grad_norm": 5.65625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2231, "epoch": 1, "loss": 0.1667, "lr": 1.8412388102980325e-05, "grad_norm": 3.171875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2232, "epoch": 1, "loss": 0.052, "lr": 1.8410768212647204e-05, "grad_norm": 3.953125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2233, "epoch": 1, "loss": 0.015, "lr": 1.8409147567656306e-05, "grad_norm": 2.859375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2234, "epoch": 1, "loss": 0.0814, "lr": 1.840752616815305e-05, "grad_norm": 3.234375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 2235, "epoch": 1, "loss": 0.0089, "lr": 1.8405904014282914e-05, "grad_norm": 2.953125, "samples_per_sec": 11.02, "gpu_mem_gb": 6.85} {"step": 2236, "epoch": 1, "loss": 0.0476, "lr": 1.8404281106191447e-05, "grad_norm": 3.53125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2237, "epoch": 1, "loss": 0.0013, "lr": 1.840265744402426e-05, "grad_norm": 6.8125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.85} {"step": 2238, "epoch": 1, "loss": 0.0016, "lr": 1.8401033027927048e-05, "grad_norm": 3.984375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2239, "epoch": 1, "loss": 0.0446, "lr": 1.839940785804555e-05, "grad_norm": 5.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2240, "epoch": 1, "loss": 0.0031, "lr": 1.839778193452559e-05, "grad_norm": 3.9375, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 2241, "epoch": 1, "loss": 0.0101, "lr": 1.839615525751305e-05, "grad_norm": 3.296875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 2242, "epoch": 1, "loss": 0.008, "lr": 1.8394527827153892e-05, "grad_norm": 2.609375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2243, "epoch": 1, "loss": 0.0513, "lr": 1.8392899643594135e-05, "grad_norm": 3.609375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2244, "epoch": 1, "loss": 0.0186, "lr": 1.8391270706979864e-05, "grad_norm": 4.25, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 2245, "epoch": 1, "loss": 0.1653, "lr": 1.8389641017457234e-05, "grad_norm": 5.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2246, "epoch": 1, "loss": 0.1588, "lr": 1.8388010575172476e-05, "grad_norm": 3.25, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2247, "epoch": 1, "loss": 0.0732, "lr": 1.838637938027188e-05, "grad_norm": 8.6875, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 2248, "epoch": 1, "loss": 0.0034, "lr": 1.8384747432901804e-05, "grad_norm": 4.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 2249, "epoch": 1, "loss": 0.0188, "lr": 1.8383114733208672e-05, "grad_norm": 2.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2250, "epoch": 1, "loss": 0.0085, "lr": 1.8381481281338983e-05, "grad_norm": 3.78125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2251, "epoch": 1, "loss": 0.0354, "lr": 1.8379847077439294e-05, "grad_norm": 4.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2252, "epoch": 1, "loss": 0.0212, "lr": 1.837821212165624e-05, "grad_norm": 1.546875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2253, "epoch": 1, "loss": 0.0002, "lr": 1.8376576414136513e-05, "grad_norm": 3.21875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2254, "epoch": 1, "loss": 0.016, "lr": 1.8374939955026883e-05, "grad_norm": 4.21875, "samples_per_sec": 10.26, "gpu_mem_gb": 6.83} {"step": 2255, "epoch": 1, "loss": 0.0015, "lr": 1.8373302744474172e-05, "grad_norm": 3.078125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2256, "epoch": 1, "loss": 0.1443, "lr": 1.8371664782625287e-05, "grad_norm": 5.65625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2257, "epoch": 1, "loss": 0.0577, "lr": 1.8370026069627192e-05, "grad_norm": 4.53125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2258, "epoch": 1, "loss": 0.0012, "lr": 1.836838660562692e-05, "grad_norm": 7.5625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 2259, "epoch": 1, "loss": 0.1689, "lr": 1.8366746390771576e-05, "grad_norm": 5.375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.84} {"step": 2260, "epoch": 1, "loss": 0.0905, "lr": 1.8365105425208325e-05, "grad_norm": 3.671875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2261, "epoch": 1, "loss": 0.0042, "lr": 1.83634637090844e-05, "grad_norm": 8.6875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 2262, "epoch": 1, "loss": 0.4393, "lr": 1.836182124254711e-05, "grad_norm": 6.9375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2263, "epoch": 1, "loss": 0.0466, "lr": 1.8360178025743822e-05, "grad_norm": 5.0625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2264, "epoch": 1, "loss": 0.1339, "lr": 1.835853405882198e-05, "grad_norm": 6.21875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 2265, "epoch": 1, "loss": 0.0171, "lr": 1.835688934192908e-05, "grad_norm": 7.84375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2266, "epoch": 1, "loss": 0.0463, "lr": 1.83552438752127e-05, "grad_norm": 5.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2267, "epoch": 1, "loss": 0.0759, "lr": 1.835359765882048e-05, "grad_norm": 2.9375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 2268, "epoch": 1, "loss": 0.0012, "lr": 1.8351950692900127e-05, "grad_norm": 0.66796875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2269, "epoch": 1, "loss": 0.0075, "lr": 1.8350302977599415e-05, "grad_norm": 4.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2270, "epoch": 1, "loss": 0.1067, "lr": 1.8348654513066183e-05, "grad_norm": 7.65625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2271, "epoch": 1, "loss": 0.0053, "lr": 1.8347005299448342e-05, "grad_norm": 3.796875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2272, "epoch": 1, "loss": 0.2508, "lr": 1.834535533689387e-05, "grad_norm": 5.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 2273, "epoch": 1, "loss": 0.4012, "lr": 1.8343704625550806e-05, "grad_norm": 17.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2274, "epoch": 1, "loss": 0.0145, "lr": 1.8342053165567263e-05, "grad_norm": 2.875, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 2275, "epoch": 1, "loss": 0.0044, "lr": 1.834040095709142e-05, "grad_norm": 4.875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2276, "epoch": 1, "loss": 0.0016, "lr": 1.8338748000271517e-05, "grad_norm": 0.69921875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 2277, "epoch": 1, "loss": 0.0441, "lr": 1.8337094295255866e-05, "grad_norm": 3.390625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 2278, "epoch": 1, "loss": 0.0242, "lr": 1.833543984219285e-05, "grad_norm": 4.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2279, "epoch": 1, "loss": 0.0025, "lr": 1.8333784641230917e-05, "grad_norm": 8.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2280, "epoch": 1, "loss": 0.0209, "lr": 1.8332128692518576e-05, "grad_norm": 6.90625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2281, "epoch": 1, "loss": 0.0372, "lr": 1.8330471996204408e-05, "grad_norm": 4.96875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 2282, "epoch": 1, "loss": 0.021, "lr": 1.832881455243706e-05, "grad_norm": 2.25, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2283, "epoch": 1, "loss": 0.4367, "lr": 1.8327156361365247e-05, "grad_norm": 6.375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.84} {"step": 2284, "epoch": 1, "loss": 0.1296, "lr": 1.832549742313775e-05, "grad_norm": 4.6875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2285, "epoch": 1, "loss": 0.0463, "lr": 1.8323837737903418e-05, "grad_norm": 3.296875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2286, "epoch": 1, "loss": 0.0017, "lr": 1.8322177305811165e-05, "grad_norm": 6.78125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 2287, "epoch": 1, "loss": 0.0164, "lr": 1.832051612700998e-05, "grad_norm": 6.9375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 2288, "epoch": 1, "loss": 0.1139, "lr": 1.8318854201648905e-05, "grad_norm": 2.34375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 2289, "epoch": 1, "loss": 0.1497, "lr": 1.831719152987706e-05, "grad_norm": 3.796875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2290, "epoch": 1, "loss": 0.0031, "lr": 1.8315528111843627e-05, "grad_norm": 5.03125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2291, "epoch": 1, "loss": 0.0715, "lr": 1.831386394769786e-05, "grad_norm": 4.28125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2292, "epoch": 1, "loss": 0.0087, "lr": 1.831219903758907e-05, "grad_norm": 4.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2293, "epoch": 1, "loss": 0.0109, "lr": 1.831053338166665e-05, "grad_norm": 4.0, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2294, "epoch": 1, "loss": 0.0018, "lr": 1.830886698008004e-05, "grad_norm": 2.4375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 2295, "epoch": 1, "loss": 0.2158, "lr": 1.830719983297877e-05, "grad_norm": 5.28125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2296, "epoch": 1, "loss": 0.1378, "lr": 1.830553194051242e-05, "grad_norm": 6.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2297, "epoch": 1, "loss": 0.009, "lr": 1.830386330283064e-05, "grad_norm": 3.78125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2298, "epoch": 1, "loss": 0.1169, "lr": 1.830219392008315e-05, "grad_norm": 5.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2299, "epoch": 1, "loss": 0.0019, "lr": 1.8300523792419737e-05, "grad_norm": 4.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2300, "epoch": 1, "loss": 0.0401, "lr": 1.8298852919990254e-05, "grad_norm": 5.125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2301, "epoch": 1, "loss": 0.0387, "lr": 1.8297181302944616e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2302, "epoch": 1, "loss": 0.0018, "lr": 1.8295508941432814e-05, "grad_norm": 2.625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 2303, "epoch": 1, "loss": 0.0353, "lr": 1.8293835835604904e-05, "grad_norm": 5.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2304, "epoch": 1, "loss": 0.2404, "lr": 1.8292161985610994e-05, "grad_norm": 5.90625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2305, "epoch": 1, "loss": 0.0034, "lr": 1.829048739160128e-05, "grad_norm": 2.875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 2306, "epoch": 1, "loss": 0.0106, "lr": 1.8288812053726015e-05, "grad_norm": 5.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2307, "epoch": 1, "loss": 0.0354, "lr": 1.8287135972135515e-05, "grad_norm": 1.34375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2308, "epoch": 1, "loss": 0.1635, "lr": 1.828545914698017e-05, "grad_norm": 8.1875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.84} {"step": 2309, "epoch": 1, "loss": 0.0977, "lr": 1.8283781578410428e-05, "grad_norm": 3.109375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2310, "epoch": 1, "loss": 0.1555, "lr": 1.8282103266576817e-05, "grad_norm": 5.9375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2311, "epoch": 1, "loss": 0.0252, "lr": 1.8280424211629922e-05, "grad_norm": 4.71875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 2312, "epoch": 1, "loss": 0.0304, "lr": 1.8278744413720387e-05, "grad_norm": 5.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2313, "epoch": 1, "loss": 0.0192, "lr": 1.8277063872998946e-05, "grad_norm": 4.84375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2314, "epoch": 1, "loss": 0.1833, "lr": 1.8275382589616377e-05, "grad_norm": 3.578125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2315, "epoch": 1, "loss": 0.1149, "lr": 1.8273700563723538e-05, "grad_norm": 7.6875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2316, "epoch": 1, "loss": 0.3204, "lr": 1.8272017795471345e-05, "grad_norm": 6.28125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.84} {"step": 2317, "epoch": 1, "loss": 0.1283, "lr": 1.8270334285010788e-05, "grad_norm": 5.0, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2318, "epoch": 1, "loss": 0.0003, "lr": 1.826865003249292e-05, "grad_norm": 0.7890625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2319, "epoch": 1, "loss": 0.048, "lr": 1.8266965038068856e-05, "grad_norm": 4.40625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 2320, "epoch": 1, "loss": 0.0925, "lr": 1.8265279301889795e-05, "grad_norm": 1.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 2321, "epoch": 1, "loss": 0.2324, "lr": 1.826359282410697e-05, "grad_norm": 5.34375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 2322, "epoch": 1, "loss": 0.006, "lr": 1.8261905604871722e-05, "grad_norm": 9.0625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 2323, "epoch": 1, "loss": 0.1608, "lr": 1.8260217644335424e-05, "grad_norm": 3.078125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2324, "epoch": 1, "loss": 0.1336, "lr": 1.8258528942649528e-05, "grad_norm": 4.15625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 2325, "epoch": 1, "loss": 0.0015, "lr": 1.825683949996556e-05, "grad_norm": 4.46875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 2326, "epoch": 1, "loss": 0.0048, "lr": 1.8255149316435102e-05, "grad_norm": 3.015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 2327, "epoch": 1, "loss": 0.0004, "lr": 1.8253458392209803e-05, "grad_norm": 2.875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2328, "epoch": 1, "loss": 0.1884, "lr": 1.8251766727441393e-05, "grad_norm": 5.4375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 2329, "epoch": 1, "loss": 0.0487, "lr": 1.825007432228164e-05, "grad_norm": 5.21875, "samples_per_sec": 8.89, "gpu_mem_gb": 6.83} {"step": 2330, "epoch": 1, "loss": 0.0918, "lr": 1.824838117688241e-05, "grad_norm": 7.21875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 2331, "epoch": 1, "loss": 0.013, "lr": 1.824668729139561e-05, "grad_norm": 2.53125, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 2332, "epoch": 1, "loss": 0.0576, "lr": 1.8244992665973235e-05, "grad_norm": 4.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 2333, "epoch": 1, "loss": 0.0511, "lr": 1.8243297300767322e-05, "grad_norm": 8.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.84} {"step": 2334, "epoch": 1, "loss": 0.0884, "lr": 1.824160119593e-05, "grad_norm": 4.21875, "samples_per_sec": 10.53, "gpu_mem_gb": 6.85} {"step": 2335, "epoch": 1, "loss": 0.2376, "lr": 1.8239904351613448e-05, "grad_norm": 4.28125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.84} {"step": 2336, "epoch": 1, "loss": 0.1897, "lr": 1.8238206767969917e-05, "grad_norm": 5.875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2337, "epoch": 1, "loss": 0.0902, "lr": 1.823650844515172e-05, "grad_norm": 5.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2338, "epoch": 1, "loss": 0.0737, "lr": 1.823480938331124e-05, "grad_norm": 4.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2339, "epoch": 1, "loss": 0.0027, "lr": 1.823310958260093e-05, "grad_norm": 3.3125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2340, "epoch": 1, "loss": 0.0512, "lr": 1.82314090431733e-05, "grad_norm": 3.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2341, "epoch": 1, "loss": 0.0225, "lr": 1.8229707765180933e-05, "grad_norm": 3.625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2342, "epoch": 1, "loss": 0.1962, "lr": 1.8228005748776474e-05, "grad_norm": 9.75, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2343, "epoch": 1, "loss": 0.1179, "lr": 1.822630299411264e-05, "grad_norm": 5.03125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.84} {"step": 2344, "epoch": 1, "loss": 0.0114, "lr": 1.8224599501342212e-05, "grad_norm": 3.28125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2345, "epoch": 1, "loss": 0.0648, "lr": 1.8222895270618032e-05, "grad_norm": 4.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2346, "epoch": 1, "loss": 0.0021, "lr": 1.8221190302093017e-05, "grad_norm": 3.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2347, "epoch": 1, "loss": 0.0225, "lr": 1.8219484595920144e-05, "grad_norm": 4.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 2348, "epoch": 1, "loss": 0.0035, "lr": 1.821777815225245e-05, "grad_norm": 1.5078125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 2349, "epoch": 1, "loss": 0.014, "lr": 1.8216070971243063e-05, "grad_norm": 4.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2350, "epoch": 1, "loss": 0.0014, "lr": 1.8214363053045145e-05, "grad_norm": 4.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2351, "epoch": 1, "loss": 0.0036, "lr": 1.8212654397811944e-05, "grad_norm": 4.09375, "samples_per_sec": 10.33, "gpu_mem_gb": 6.83} {"step": 2352, "epoch": 1, "loss": 0.0017, "lr": 1.8210945005696773e-05, "grad_norm": 4.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2353, "epoch": 1, "loss": 0.2992, "lr": 1.8209234876853e-05, "grad_norm": 5.21875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2354, "epoch": 1, "loss": 0.1766, "lr": 1.8207524011434074e-05, "grad_norm": 5.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2355, "epoch": 1, "loss": 0.0234, "lr": 1.82058124095935e-05, "grad_norm": 5.53125, "samples_per_sec": 10.32, "gpu_mem_gb": 6.85} {"step": 2356, "epoch": 1, "loss": 0.1482, "lr": 1.820410007148485e-05, "grad_norm": 7.15625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 2357, "epoch": 1, "loss": 0.0724, "lr": 1.820238699726177e-05, "grad_norm": 4.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2358, "epoch": 1, "loss": 0.0698, "lr": 1.8200673187077956e-05, "grad_norm": 3.703125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2359, "epoch": 1, "loss": 0.0036, "lr": 1.8198958641087188e-05, "grad_norm": 2.953125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2360, "epoch": 1, "loss": 0.1265, "lr": 1.81972433594433e-05, "grad_norm": 5.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 2361, "epoch": 1, "loss": 0.1277, "lr": 1.81955273423002e-05, "grad_norm": 5.9375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2362, "epoch": 1, "loss": 0.0005, "lr": 1.8193810589811853e-05, "grad_norm": 3.140625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2363, "epoch": 1, "loss": 0.0517, "lr": 1.8192093102132294e-05, "grad_norm": 5.65625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2364, "epoch": 1, "loss": 0.0984, "lr": 1.8190374879415634e-05, "grad_norm": 3.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2365, "epoch": 1, "loss": 0.0011, "lr": 1.818865592181603e-05, "grad_norm": 5.875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2366, "epoch": 1, "loss": 0.1321, "lr": 1.8186936229487723e-05, "grad_norm": 2.5, "samples_per_sec": 11.0, "gpu_mem_gb": 6.84} {"step": 2367, "epoch": 1, "loss": 0.0016, "lr": 1.8185215802585013e-05, "grad_norm": 5.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2368, "epoch": 1, "loss": 0.0457, "lr": 1.818349464126226e-05, "grad_norm": 6.0, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2369, "epoch": 1, "loss": 0.0013, "lr": 1.81817727456739e-05, "grad_norm": 3.890625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2370, "epoch": 1, "loss": 0.0258, "lr": 1.818005011597443e-05, "grad_norm": 4.03125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2371, "epoch": 1, "loss": 0.0414, "lr": 1.817832675231841e-05, "grad_norm": 5.5, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 2372, "epoch": 1, "loss": 0.1922, "lr": 1.817660265486047e-05, "grad_norm": 3.875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.84} {"step": 2373, "epoch": 1, "loss": 0.004, "lr": 1.817487782375531e-05, "grad_norm": 2.828125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2374, "epoch": 1, "loss": 0.0048, "lr": 1.817315225915769e-05, "grad_norm": 4.28125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2375, "epoch": 1, "loss": 0.1091, "lr": 1.8171425961222426e-05, "grad_norm": 4.375, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 2376, "epoch": 1, "loss": 0.0001, "lr": 1.816969893010442e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2377, "epoch": 1, "loss": 0.0008, "lr": 1.816797116595863e-05, "grad_norm": 2.59375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2378, "epoch": 1, "loss": 0.0767, "lr": 1.816624266894008e-05, "grad_norm": 3.953125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2379, "epoch": 1, "loss": 0.0872, "lr": 1.816451343920386e-05, "grad_norm": 4.90625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2380, "epoch": 1, "loss": 0.0092, "lr": 1.8162783476905118e-05, "grad_norm": 3.046875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 2381, "epoch": 1, "loss": 0.0019, "lr": 1.8161052782199084e-05, "grad_norm": 5.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 2382, "epoch": 1, "loss": 0.0015, "lr": 1.8159321355241042e-05, "grad_norm": 2.015625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2383, "epoch": 1, "loss": 0.0575, "lr": 1.815758919618634e-05, "grad_norm": 8.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2384, "epoch": 1, "loss": 0.0304, "lr": 1.8155856305190406e-05, "grad_norm": 5.625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2385, "epoch": 1, "loss": 0.0512, "lr": 1.8154122682408716e-05, "grad_norm": 4.90625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2386, "epoch": 1, "loss": 0.0979, "lr": 1.8152388327996822e-05, "grad_norm": 9.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2387, "epoch": 1, "loss": 0.0015, "lr": 1.815065324211034e-05, "grad_norm": 2.21875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2388, "epoch": 1, "loss": 0.043, "lr": 1.8148917424904952e-05, "grad_norm": 2.625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 2389, "epoch": 1, "loss": 0.2554, "lr": 1.8147180876536402e-05, "grad_norm": 7.6875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2390, "epoch": 1, "loss": 0.0016, "lr": 1.8145443597160503e-05, "grad_norm": 4.34375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 2391, "epoch": 1, "loss": 0.0045, "lr": 1.8143705586933128e-05, "grad_norm": 2.375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 2392, "epoch": 1, "loss": 0.0009, "lr": 1.814196684601023e-05, "grad_norm": 3.546875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2393, "epoch": 1, "loss": 0.0006, "lr": 1.8140227374547815e-05, "grad_norm": 4.6875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2394, "epoch": 1, "loss": 0.0024, "lr": 1.813848717270195e-05, "grad_norm": 4.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2395, "epoch": 1, "loss": 0.0639, "lr": 1.8136746240628783e-05, "grad_norm": 2.859375, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 2396, "epoch": 1, "loss": 0.0003, "lr": 1.8135004578484515e-05, "grad_norm": 7.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2397, "epoch": 1, "loss": 0.0036, "lr": 1.8133262186425424e-05, "grad_norm": 0.55859375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2398, "epoch": 1, "loss": 0.0758, "lr": 1.8131519064607837e-05, "grad_norm": 3.25, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2399, "epoch": 1, "loss": 0.1099, "lr": 1.8129775213188163e-05, "grad_norm": 7.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2400, "epoch": 1, "loss": 0.0917, "lr": 1.812803063232286e-05, "grad_norm": 4.28125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 2401, "epoch": 1, "loss": 0.0018, "lr": 1.8126285322168472e-05, "grad_norm": 7.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2402, "epoch": 1, "loss": 0.0021, "lr": 1.8124539282881593e-05, "grad_norm": 5.8125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 2403, "epoch": 1, "loss": 0.0579, "lr": 1.8122792514618886e-05, "grad_norm": 5.28125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 2404, "epoch": 1, "loss": 0.113, "lr": 1.8121045017537076e-05, "grad_norm": 2.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2405, "epoch": 1, "loss": 0.0003, "lr": 1.811929679179297e-05, "grad_norm": 1.6015625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2406, "epoch": 1, "loss": 0.0003, "lr": 1.8117547837543416e-05, "grad_norm": 2.5, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2407, "epoch": 1, "loss": 0.135, "lr": 1.811579815494534e-05, "grad_norm": 7.84375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2408, "epoch": 1, "loss": 0.1089, "lr": 1.811404774415574e-05, "grad_norm": 6.75, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2409, "epoch": 1, "loss": 0.0065, "lr": 1.8112296605331664e-05, "grad_norm": 6.84375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 2410, "epoch": 1, "loss": 0.0712, "lr": 1.811054473863024e-05, "grad_norm": 4.625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2411, "epoch": 1, "loss": 0.0047, "lr": 1.810879214420865e-05, "grad_norm": 3.234375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2412, "epoch": 1, "loss": 0.0296, "lr": 1.810703882222415e-05, "grad_norm": 5.3125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2413, "epoch": 1, "loss": 0.0994, "lr": 1.810528477283405e-05, "grad_norm": 6.15625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 2414, "epoch": 1, "loss": 0.0329, "lr": 1.810352999619574e-05, "grad_norm": 6.28125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2415, "epoch": 1, "loss": 0.1004, "lr": 1.8101774492466665e-05, "grad_norm": 3.625, "samples_per_sec": 10.38, "gpu_mem_gb": 6.83} {"step": 2416, "epoch": 1, "loss": 0.191, "lr": 1.8100018261804337e-05, "grad_norm": 4.8125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2417, "epoch": 1, "loss": 0.0528, "lr": 1.8098261304366334e-05, "grad_norm": 2.921875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2418, "epoch": 1, "loss": 0.0044, "lr": 1.8096503620310298e-05, "grad_norm": 3.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2419, "epoch": 1, "loss": 0.0011, "lr": 1.809474520979394e-05, "grad_norm": 6.0, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 2420, "epoch": 1, "loss": 0.3201, "lr": 1.8092986072975035e-05, "grad_norm": 7.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2421, "epoch": 1, "loss": 0.032, "lr": 1.8091226210011423e-05, "grad_norm": 5.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2422, "epoch": 1, "loss": 0.0004, "lr": 1.8089465621061e-05, "grad_norm": 3.546875, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 2423, "epoch": 1, "loss": 0.0008, "lr": 1.808770430628174e-05, "grad_norm": 3.03125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2424, "epoch": 1, "loss": 0.1503, "lr": 1.8085942265831682e-05, "grad_norm": 5.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2425, "epoch": 1, "loss": 0.145, "lr": 1.8084179499868916e-05, "grad_norm": 10.0625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2426, "epoch": 1, "loss": 0.0347, "lr": 1.8082416008551616e-05, "grad_norm": 4.84375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2427, "epoch": 1, "loss": 0.1131, "lr": 1.8080651792038006e-05, "grad_norm": 5.6875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 2428, "epoch": 1, "loss": 0.0479, "lr": 1.807888685048638e-05, "grad_norm": 5.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 2429, "epoch": 1, "loss": 0.0115, "lr": 1.80771211840551e-05, "grad_norm": 5.71875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 2430, "epoch": 1, "loss": 0.0058, "lr": 1.807535479290259e-05, "grad_norm": 2.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2431, "epoch": 1, "loss": 0.0001, "lr": 1.8073587677187344e-05, "grad_norm": 6.875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2432, "epoch": 1, "loss": 0.0739, "lr": 1.8071819837067907e-05, "grad_norm": 4.71875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2433, "epoch": 1, "loss": 0.0029, "lr": 1.8070051272702905e-05, "grad_norm": 5.375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2434, "epoch": 1, "loss": 0.0539, "lr": 1.8068281984251027e-05, "grad_norm": 3.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2435, "epoch": 1, "loss": 0.0139, "lr": 1.8066511971871015e-05, "grad_norm": 7.34375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 2436, "epoch": 1, "loss": 0.0602, "lr": 1.806474123572169e-05, "grad_norm": 6.34375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 2437, "epoch": 1, "loss": 0.022, "lr": 1.8062969775961924e-05, "grad_norm": 6.65625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2438, "epoch": 1, "loss": 0.0079, "lr": 1.806119759275067e-05, "grad_norm": 4.65625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2439, "epoch": 1, "loss": 0.3586, "lr": 1.8059424686246928e-05, "grad_norm": 6.75, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2440, "epoch": 1, "loss": 0.0396, "lr": 1.8057651056609784e-05, "grad_norm": 2.53125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.85} {"step": 2441, "epoch": 1, "loss": 0.3311, "lr": 1.8055876703998368e-05, "grad_norm": 7.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2442, "epoch": 1, "loss": 0.1523, "lr": 1.8054101628571888e-05, "grad_norm": 5.625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 2443, "epoch": 1, "loss": 0.0668, "lr": 1.805232583048961e-05, "grad_norm": 5.65625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2444, "epoch": 1, "loss": 0.164, "lr": 1.8050549309910873e-05, "grad_norm": 4.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2445, "epoch": 1, "loss": 0.032, "lr": 1.8048772066995074e-05, "grad_norm": 2.03125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 2446, "epoch": 1, "loss": 0.3246, "lr": 1.804699410190167e-05, "grad_norm": 4.75, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 2447, "epoch": 1, "loss": 0.2629, "lr": 1.8045215414790197e-05, "grad_norm": 5.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2448, "epoch": 1, "loss": 0.018, "lr": 1.8043436005820247e-05, "grad_norm": 3.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2449, "epoch": 1, "loss": 0.2695, "lr": 1.8041655875151477e-05, "grad_norm": 8.75, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2450, "epoch": 1, "loss": 0.0074, "lr": 1.8039875022943603e-05, "grad_norm": 2.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2451, "epoch": 1, "loss": 0.0381, "lr": 1.8038093449356423e-05, "grad_norm": 2.359375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2452, "epoch": 1, "loss": 0.123, "lr": 1.8036311154549783e-05, "grad_norm": 5.90625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2453, "epoch": 1, "loss": 0.0038, "lr": 1.80345281386836e-05, "grad_norm": 4.875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 2454, "epoch": 1, "loss": 0.0388, "lr": 1.803274440191786e-05, "grad_norm": 4.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2455, "epoch": 1, "loss": 0.0015, "lr": 1.80309599444126e-05, "grad_norm": 0.875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2456, "epoch": 1, "loss": 0.0614, "lr": 1.802917476632794e-05, "grad_norm": 5.78125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 2457, "epoch": 1, "loss": 0.1014, "lr": 1.8027388867824052e-05, "grad_norm": 6.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2458, "epoch": 1, "loss": 0.0006, "lr": 1.8025602249061173e-05, "grad_norm": 2.6875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2459, "epoch": 1, "loss": 0.0002, "lr": 1.8023814910199613e-05, "grad_norm": 3.765625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 2460, "epoch": 1, "loss": 0.0063, "lr": 1.8022026851399737e-05, "grad_norm": 7.4375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 2461, "epoch": 1, "loss": 0.0009, "lr": 1.8020238072821982e-05, "grad_norm": 6.25, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 2462, "epoch": 1, "loss": 0.0172, "lr": 1.801844857462684e-05, "grad_norm": 3.671875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 2463, "epoch": 1, "loss": 0.0078, "lr": 1.8016658356974885e-05, "grad_norm": 3.6875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.85} {"step": 2464, "epoch": 1, "loss": 0.1016, "lr": 1.8014867420026738e-05, "grad_norm": 3.296875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2465, "epoch": 1, "loss": 0.0529, "lr": 1.8013075763943095e-05, "grad_norm": 6.6875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2466, "epoch": 1, "loss": 0.098, "lr": 1.8011283388884703e-05, "grad_norm": 4.90625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 2467, "epoch": 1, "loss": 0.0012, "lr": 1.8009490295012394e-05, "grad_norm": 2.046875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2468, "epoch": 1, "loss": 0.0064, "lr": 1.800769648248705e-05, "grad_norm": 5.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2469, "epoch": 1, "loss": 0.0062, "lr": 1.800590195146962e-05, "grad_norm": 1.9296875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2470, "epoch": 1, "loss": 0.0012, "lr": 1.8004106702121116e-05, "grad_norm": 2.3125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2471, "epoch": 1, "loss": 0.1152, "lr": 1.8002310734602625e-05, "grad_norm": 5.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2472, "epoch": 1, "loss": 0.1157, "lr": 1.8000514049075283e-05, "grad_norm": 6.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2473, "epoch": 1, "loss": 0.0265, "lr": 1.7998716645700302e-05, "grad_norm": 2.109375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2474, "epoch": 1, "loss": 0.1481, "lr": 1.7996918524638956e-05, "grad_norm": 10.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2475, "epoch": 1, "loss": 0.165, "lr": 1.799511968605258e-05, "grad_norm": 6.71875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2476, "epoch": 1, "loss": 0.0048, "lr": 1.7993320130102567e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2477, "epoch": 1, "loss": 0.3086, "lr": 1.7991519856950396e-05, "grad_norm": 10.75, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 2478, "epoch": 1, "loss": 0.0041, "lr": 1.7989718866757587e-05, "grad_norm": 6.1875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 2479, "epoch": 1, "loss": 0.1216, "lr": 1.798791715968574e-05, "grad_norm": 4.625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2480, "epoch": 1, "loss": 0.0031, "lr": 1.798611473589651e-05, "grad_norm": 4.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2481, "epoch": 1, "loss": 0.0335, "lr": 1.798431159555162e-05, "grad_norm": 6.0625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 2482, "epoch": 1, "loss": 0.002, "lr": 1.798250773881286e-05, "grad_norm": 2.546875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2483, "epoch": 1, "loss": 0.0164, "lr": 1.798070316584208e-05, "grad_norm": 3.484375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2484, "epoch": 1, "loss": 0.0799, "lr": 1.7978897876801192e-05, "grad_norm": 3.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2485, "epoch": 1, "loss": 0.0177, "lr": 1.797709187185218e-05, "grad_norm": 5.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2486, "epoch": 1, "loss": 0.0047, "lr": 1.797528515115709e-05, "grad_norm": 4.28125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 2487, "epoch": 1, "loss": 0.0279, "lr": 1.7973477714878027e-05, "grad_norm": 2.6875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2488, "epoch": 1, "loss": 0.0232, "lr": 1.7971669563177165e-05, "grad_norm": 3.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2489, "epoch": 1, "loss": 0.0219, "lr": 1.7969860696216738e-05, "grad_norm": 4.4375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 2490, "epoch": 1, "loss": 0.0854, "lr": 1.7968051114159046e-05, "grad_norm": 5.8125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 2491, "epoch": 1, "loss": 0.1389, "lr": 1.7966240817166464e-05, "grad_norm": 3.40625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2492, "epoch": 1, "loss": 0.0163, "lr": 1.7964429805401406e-05, "grad_norm": 2.28125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2493, "epoch": 1, "loss": 0.0572, "lr": 1.796261807902638e-05, "grad_norm": 3.40625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2494, "epoch": 1, "loss": 0.0077, "lr": 1.7960805638203936e-05, "grad_norm": 2.265625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2495, "epoch": 1, "loss": 0.0151, "lr": 1.7958992483096693e-05, "grad_norm": 1.140625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2496, "epoch": 1, "loss": 0.0006, "lr": 1.7957178613867343e-05, "grad_norm": 4.4375, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 2497, "epoch": 1, "loss": 0.0008, "lr": 1.7955364030678635e-05, "grad_norm": 7.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2498, "epoch": 1, "loss": 0.1647, "lr": 1.7953548733693378e-05, "grad_norm": 3.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.84} {"step": 2499, "epoch": 1, "loss": 0.1776, "lr": 1.7951732723074453e-05, "grad_norm": 6.0, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2500, "epoch": 1, "loss": 0.1564, "lr": 1.7949915998984803e-05, "grad_norm": 7.96875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2501, "epoch": 1, "loss": 0.0313, "lr": 1.794809856158743e-05, "grad_norm": 5.8125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2502, "epoch": 1, "loss": 0.0232, "lr": 1.794628041104541e-05, "grad_norm": 2.84375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2503, "epoch": 1, "loss": 0.0032, "lr": 1.7944461547521874e-05, "grad_norm": 6.40625, "samples_per_sec": 10.3, "gpu_mem_gb": 6.82} {"step": 2504, "epoch": 1, "loss": 0.0052, "lr": 1.7942641971180017e-05, "grad_norm": 3.703125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2505, "epoch": 1, "loss": 0.0892, "lr": 1.79408216821831e-05, "grad_norm": 4.09375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2506, "epoch": 1, "loss": 0.0084, "lr": 1.7939000680694456e-05, "grad_norm": 3.734375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 2507, "epoch": 1, "loss": 0.0469, "lr": 1.7937178966877472e-05, "grad_norm": 3.90625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2508, "epoch": 1, "loss": 0.0214, "lr": 1.79353565408956e-05, "grad_norm": 1.1640625, "samples_per_sec": 11.0, "gpu_mem_gb": 6.85} {"step": 2509, "epoch": 1, "loss": 0.0038, "lr": 1.7933533402912354e-05, "grad_norm": 6.5625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2510, "epoch": 1, "loss": 0.1122, "lr": 1.793170955309132e-05, "grad_norm": 7.0, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 2511, "epoch": 1, "loss": 0.0957, "lr": 1.7929884991596144e-05, "grad_norm": 5.84375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2512, "epoch": 1, "loss": 0.0005, "lr": 1.7928059718590535e-05, "grad_norm": 2.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2513, "epoch": 1, "loss": 0.0067, "lr": 1.792623373423826e-05, "grad_norm": 1.9140625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2514, "epoch": 1, "loss": 0.0003, "lr": 1.792440703870316e-05, "grad_norm": 3.609375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2515, "epoch": 1, "loss": 0.0008, "lr": 1.792257963214914e-05, "grad_norm": 3.5625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 2516, "epoch": 1, "loss": 0.0008, "lr": 1.792075151474016e-05, "grad_norm": 3.890625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2517, "epoch": 1, "loss": 0.0945, "lr": 1.7918922686640248e-05, "grad_norm": 4.59375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 2518, "epoch": 1, "loss": 0.0019, "lr": 1.7917093148013494e-05, "grad_norm": 1.2109375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2519, "epoch": 1, "loss": 0.005, "lr": 1.791526289902406e-05, "grad_norm": 6.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2520, "epoch": 1, "loss": 0.0028, "lr": 1.791343193983616e-05, "grad_norm": 4.15625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 2521, "epoch": 1, "loss": 0.0484, "lr": 1.7911600270614077e-05, "grad_norm": 9.25, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2522, "epoch": 1, "loss": 0.0004, "lr": 1.790976789152216e-05, "grad_norm": 4.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2523, "epoch": 1, "loss": 0.0008, "lr": 1.790793480272482e-05, "grad_norm": 2.59375, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 2524, "epoch": 1, "loss": 0.0003, "lr": 1.790610100438653e-05, "grad_norm": 3.796875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2525, "epoch": 1, "loss": 0.3295, "lr": 1.790426649667183e-05, "grad_norm": 6.9375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2526, "epoch": 1, "loss": 0.0039, "lr": 1.7902431279745317e-05, "grad_norm": 2.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 2527, "epoch": 1, "loss": 0.0636, "lr": 1.790059535377166e-05, "grad_norm": 3.234375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2528, "epoch": 1, "loss": 0.0003, "lr": 1.789875871891559e-05, "grad_norm": 16.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2529, "epoch": 1, "loss": 0.0574, "lr": 1.789692137534189e-05, "grad_norm": 18.875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2530, "epoch": 1, "loss": 0.0221, "lr": 1.7895083323215427e-05, "grad_norm": 6.65625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2531, "epoch": 1, "loss": 0.0207, "lr": 1.7893244562701114e-05, "grad_norm": 3.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2532, "epoch": 1, "loss": 0.3492, "lr": 1.789140509396394e-05, "grad_norm": 7.0625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2533, "epoch": 1, "loss": 0.121, "lr": 1.7889564917168943e-05, "grad_norm": 3.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.84} {"step": 2534, "epoch": 1, "loss": 0.0042, "lr": 1.7887724032481237e-05, "grad_norm": 2.125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2535, "epoch": 1, "loss": 0.0001, "lr": 1.7885882440066e-05, "grad_norm": 1.84375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2536, "epoch": 1, "loss": 0.0689, "lr": 1.7884040140088465e-05, "grad_norm": 7.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2537, "epoch": 1, "loss": 0.0012, "lr": 1.7882197132713933e-05, "grad_norm": 0.2216796875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 2538, "epoch": 1, "loss": 0.0062, "lr": 1.788035341810777e-05, "grad_norm": 8.375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2539, "epoch": 1, "loss": 0.0806, "lr": 1.78785089964354e-05, "grad_norm": 6.84375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2540, "epoch": 1, "loss": 0.0023, "lr": 1.7876663867862317e-05, "grad_norm": 4.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2541, "epoch": 1, "loss": 0.1383, "lr": 1.7874818032554074e-05, "grad_norm": 4.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2542, "epoch": 1, "loss": 0.0056, "lr": 1.7872971490676293e-05, "grad_norm": 2.765625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2543, "epoch": 1, "loss": 0.1809, "lr": 1.787112424239465e-05, "grad_norm": 3.390625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.84} {"step": 2544, "epoch": 1, "loss": 0.0192, "lr": 1.7869276287874894e-05, "grad_norm": 5.78125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 2545, "epoch": 1, "loss": 0.126, "lr": 1.7867427627282827e-05, "grad_norm": 6.78125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2546, "epoch": 1, "loss": 0.0696, "lr": 1.7865578260784328e-05, "grad_norm": 7.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2547, "epoch": 1, "loss": 0.2129, "lr": 1.7863728188545326e-05, "grad_norm": 5.1875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 2548, "epoch": 1, "loss": 0.0058, "lr": 1.7861877410731822e-05, "grad_norm": 1.1484375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2549, "epoch": 1, "loss": 0.1319, "lr": 1.786002592750988e-05, "grad_norm": 5.90625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2550, "epoch": 1, "loss": 0.0005, "lr": 1.7858173739045615e-05, "grad_norm": 7.71875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2551, "epoch": 1, "loss": 0.006, "lr": 1.7856320845505224e-05, "grad_norm": 0.63671875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2552, "epoch": 1, "loss": 0.0342, "lr": 1.7854467247054957e-05, "grad_norm": 4.6875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2553, "epoch": 1, "loss": 0.0832, "lr": 1.7852612943861128e-05, "grad_norm": 4.96875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2554, "epoch": 1, "loss": 0.0269, "lr": 1.7850757936090108e-05, "grad_norm": 4.78125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2555, "epoch": 1, "loss": 0.003, "lr": 1.7848902223908345e-05, "grad_norm": 4.65625, "samples_per_sec": 8.92, "gpu_mem_gb": 6.85} {"step": 2556, "epoch": 1, "loss": 0.1371, "lr": 1.7847045807482347e-05, "grad_norm": 3.53125, "samples_per_sec": 11.8, "gpu_mem_gb": 6.84} {"step": 2557, "epoch": 1, "loss": 0.2414, "lr": 1.7845188686978673e-05, "grad_norm": 5.03125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2558, "epoch": 1, "loss": 0.0198, "lr": 1.7843330862563958e-05, "grad_norm": 2.46875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 2559, "epoch": 1, "loss": 0.0017, "lr": 1.7841472334404893e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2560, "epoch": 1, "loss": 0.0048, "lr": 1.7839613102668233e-05, "grad_norm": 1.6953125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2561, "epoch": 1, "loss": 0.4563, "lr": 1.7837753167520805e-05, "grad_norm": 7.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2562, "epoch": 1, "loss": 0.0015, "lr": 1.7835892529129486e-05, "grad_norm": 1.515625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 2563, "epoch": 1, "loss": 0.0494, "lr": 1.7834031187661224e-05, "grad_norm": 2.359375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2564, "epoch": 1, "loss": 0.1076, "lr": 1.783216914328303e-05, "grad_norm": 2.828125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 2565, "epoch": 1, "loss": 0.1429, "lr": 1.7830306396161972e-05, "grad_norm": 6.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2566, "epoch": 1, "loss": 0.1792, "lr": 1.7828442946465188e-05, "grad_norm": 5.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.84} {"step": 2567, "epoch": 1, "loss": 0.1008, "lr": 1.7826578794359878e-05, "grad_norm": 3.09375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 2568, "epoch": 1, "loss": 0.3356, "lr": 1.78247139400133e-05, "grad_norm": 5.6875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2569, "epoch": 1, "loss": 0.065, "lr": 1.782284838359278e-05, "grad_norm": 6.78125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2570, "epoch": 1, "loss": 0.0447, "lr": 1.7820982125265705e-05, "grad_norm": 2.03125, "samples_per_sec": 10.45, "gpu_mem_gb": 6.85} {"step": 2571, "epoch": 1, "loss": 0.0018, "lr": 1.781911516519953e-05, "grad_norm": 2.203125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 2572, "epoch": 1, "loss": 0.0017, "lr": 1.781724750356176e-05, "grad_norm": 2.40625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2573, "epoch": 1, "loss": 0.0067, "lr": 1.781537914051998e-05, "grad_norm": 4.0625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2574, "epoch": 1, "loss": 0.027, "lr": 1.781351007624182e-05, "grad_norm": 2.46875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 2575, "epoch": 1, "loss": 0.0057, "lr": 1.781164031089499e-05, "grad_norm": 3.015625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2576, "epoch": 1, "loss": 0.0035, "lr": 1.780976984464725e-05, "grad_norm": 6.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2577, "epoch": 1, "loss": 0.0022, "lr": 1.780789867766643e-05, "grad_norm": 5.46875, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 2578, "epoch": 1, "loss": 0.0034, "lr": 1.7806026810120423e-05, "grad_norm": 1.6953125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2579, "epoch": 1, "loss": 0.0086, "lr": 1.7804154242177178e-05, "grad_norm": 3.703125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2580, "epoch": 1, "loss": 0.001, "lr": 1.7802280974004717e-05, "grad_norm": 3.703125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2581, "epoch": 1, "loss": 0.0308, "lr": 1.7800407005771112e-05, "grad_norm": 4.46875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2582, "epoch": 1, "loss": 0.0048, "lr": 1.779853233764451e-05, "grad_norm": 2.203125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2583, "epoch": 1, "loss": 0.0001, "lr": 1.779665696979312e-05, "grad_norm": 0.412109375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2584, "epoch": 1, "loss": 0.017, "lr": 1.7794780902385198e-05, "grad_norm": 1.46875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2585, "epoch": 1, "loss": 0.0004, "lr": 1.7792904135589085e-05, "grad_norm": 7.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2586, "epoch": 1, "loss": 0.1888, "lr": 1.779102666957317e-05, "grad_norm": 4.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2587, "epoch": 1, "loss": 0.2127, "lr": 1.7789148504505912e-05, "grad_norm": 5.75, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 2588, "epoch": 1, "loss": 0.0695, "lr": 1.7787269640555825e-05, "grad_norm": 6.125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2589, "epoch": 1, "loss": 0.0908, "lr": 1.7785390077891493e-05, "grad_norm": 11.9375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 2590, "epoch": 1, "loss": 0.0266, "lr": 1.7783509816681563e-05, "grad_norm": 4.40625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 2591, "epoch": 1, "loss": 0.2373, "lr": 1.7781628857094736e-05, "grad_norm": 4.25, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 2592, "epoch": 1, "loss": 0.0032, "lr": 1.7779747199299785e-05, "grad_norm": 6.90625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2593, "epoch": 1, "loss": 0.1542, "lr": 1.7777864843465542e-05, "grad_norm": 3.96875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2594, "epoch": 1, "loss": 0.1509, "lr": 1.77759817897609e-05, "grad_norm": 7.625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 2595, "epoch": 1, "loss": 0.0021, "lr": 1.777409803835482e-05, "grad_norm": 5.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2596, "epoch": 1, "loss": 0.0037, "lr": 1.7772213589416317e-05, "grad_norm": 6.625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2597, "epoch": 1, "loss": 0.0272, "lr": 1.7770328443114476e-05, "grad_norm": 6.125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 2598, "epoch": 1, "loss": 0.0758, "lr": 1.7768442599618444e-05, "grad_norm": 7.75, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2599, "epoch": 1, "loss": 0.0376, "lr": 1.7766556059097424e-05, "grad_norm": 4.46875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2600, "epoch": 1, "loss": 0.0812, "lr": 1.776466882172069e-05, "grad_norm": 4.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 2601, "epoch": 1, "loss": 0.0296, "lr": 1.7762780887657576e-05, "grad_norm": 3.8125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2602, "epoch": 1, "loss": 0.0025, "lr": 1.7760892257077474e-05, "grad_norm": 2.546875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2603, "epoch": 1, "loss": 0.0016, "lr": 1.775900293014984e-05, "grad_norm": 4.21875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2604, "epoch": 1, "loss": 0.0549, "lr": 1.77571129070442e-05, "grad_norm": 2.6875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2605, "epoch": 1, "loss": 0.1107, "lr": 1.7755222187930132e-05, "grad_norm": 4.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2606, "epoch": 1, "loss": 0.0826, "lr": 1.775333077297728e-05, "grad_norm": 5.90625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 2607, "epoch": 1, "loss": 0.002, "lr": 1.775143866235536e-05, "grad_norm": 5.09375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2608, "epoch": 1, "loss": 0.0342, "lr": 1.7749545856234134e-05, "grad_norm": 4.28125, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 2609, "epoch": 1, "loss": 0.0206, "lr": 1.7747652354783434e-05, "grad_norm": 6.78125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 2610, "epoch": 1, "loss": 0.0348, "lr": 1.7745758158173164e-05, "grad_norm": 5.84375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2611, "epoch": 1, "loss": 0.004, "lr": 1.774386326657327e-05, "grad_norm": 2.78125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.85} {"step": 2612, "epoch": 1, "loss": 0.0485, "lr": 1.7741967680153776e-05, "grad_norm": 4.125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2613, "epoch": 1, "loss": 0.0288, "lr": 1.774007139908476e-05, "grad_norm": 4.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2614, "epoch": 1, "loss": 0.007, "lr": 1.7738174423536377e-05, "grad_norm": 2.671875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 2615, "epoch": 1, "loss": 0.0009, "lr": 1.7736276753678826e-05, "grad_norm": 0.330078125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2616, "epoch": 1, "loss": 0.0022, "lr": 1.7734378389682373e-05, "grad_norm": 3.90625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2617, "epoch": 1, "loss": 0.0038, "lr": 1.7732479331717353e-05, "grad_norm": 1.78125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2618, "epoch": 1, "loss": 0.2077, "lr": 1.773057957995416e-05, "grad_norm": 6.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2619, "epoch": 1, "loss": 0.1695, "lr": 1.7728679134563247e-05, "grad_norm": 4.3125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2620, "epoch": 1, "loss": 0.0033, "lr": 1.7726777995715136e-05, "grad_norm": 1.2890625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2621, "epoch": 1, "loss": 0.0019, "lr": 1.7724876163580403e-05, "grad_norm": 8.625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2622, "epoch": 1, "loss": 0.0107, "lr": 1.772297363832969e-05, "grad_norm": 4.53125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2623, "epoch": 1, "loss": 0.4231, "lr": 1.7721070420133702e-05, "grad_norm": 9.125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2624, "epoch": 1, "loss": 0.0211, "lr": 1.771916650916321e-05, "grad_norm": 4.03125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2625, "epoch": 1, "loss": 0.2377, "lr": 1.7717261905589037e-05, "grad_norm": 5.15625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2626, "epoch": 1, "loss": 0.0053, "lr": 1.771535660958208e-05, "grad_norm": 1.0390625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2627, "epoch": 1, "loss": 0.5333, "lr": 1.7713450621313286e-05, "grad_norm": 6.375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2628, "epoch": 1, "loss": 0.2013, "lr": 1.7711543940953667e-05, "grad_norm": 5.84375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 2629, "epoch": 1, "loss": 0.1056, "lr": 1.7709636568674315e-05, "grad_norm": 4.4375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2630, "epoch": 1, "loss": 0.0751, "lr": 1.7707728504646353e-05, "grad_norm": 1.796875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2631, "epoch": 1, "loss": 0.1946, "lr": 1.7705819749040994e-05, "grad_norm": 6.75, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 2632, "epoch": 1, "loss": 0.0019, "lr": 1.770391030202949e-05, "grad_norm": 5.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2633, "epoch": 1, "loss": 0.0004, "lr": 1.7702000163783183e-05, "grad_norm": 4.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2634, "epoch": 1, "loss": 0.0019, "lr": 1.7700089334473446e-05, "grad_norm": 1.703125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 2635, "epoch": 1, "loss": 0.3187, "lr": 1.7698177814271732e-05, "grad_norm": 5.5, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 2636, "epoch": 1, "loss": 0.1255, "lr": 1.7696265603349557e-05, "grad_norm": 4.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2637, "epoch": 1, "loss": 0.2157, "lr": 1.7694352701878486e-05, "grad_norm": 4.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2638, "epoch": 1, "loss": 0.0837, "lr": 1.7692439110030166e-05, "grad_norm": 14.75, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 2639, "epoch": 1, "loss": 0.0014, "lr": 1.7690524827976287e-05, "grad_norm": 2.0625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 2640, "epoch": 1, "loss": 0.0154, "lr": 1.768860985588861e-05, "grad_norm": 3.953125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2641, "epoch": 1, "loss": 0.04, "lr": 1.7686694193938952e-05, "grad_norm": 5.53125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2642, "epoch": 1, "loss": 0.3387, "lr": 1.7684777842299206e-05, "grad_norm": 6.78125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 2643, "epoch": 1, "loss": 0.0278, "lr": 1.7682860801141306e-05, "grad_norm": 3.328125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 2644, "epoch": 1, "loss": 0.0018, "lr": 1.7680943070637265e-05, "grad_norm": 3.234375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 2645, "epoch": 1, "loss": 0.0039, "lr": 1.7679024650959152e-05, "grad_norm": 5.15625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 2646, "epoch": 1, "loss": 0.1602, "lr": 1.7677105542279093e-05, "grad_norm": 3.640625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2647, "epoch": 1, "loss": 0.0419, "lr": 1.7675185744769286e-05, "grad_norm": 3.8125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 2648, "epoch": 1, "loss": 0.1068, "lr": 1.7673265258601984e-05, "grad_norm": 9.0, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2649, "epoch": 1, "loss": 0.0003, "lr": 1.76713440839495e-05, "grad_norm": 1.859375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 2650, "epoch": 1, "loss": 0.0697, "lr": 1.7669422220984215e-05, "grad_norm": 5.15625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2651, "epoch": 1, "loss": 0.0033, "lr": 1.7667499669878565e-05, "grad_norm": 3.5, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 2652, "epoch": 1, "loss": 0.0246, "lr": 1.7665576430805053e-05, "grad_norm": 3.828125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2653, "epoch": 1, "loss": 0.2891, "lr": 1.7663652503936245e-05, "grad_norm": 5.8125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2654, "epoch": 1, "loss": 0.0004, "lr": 1.766172788944476e-05, "grad_norm": 3.734375, "samples_per_sec": 10.48, "gpu_mem_gb": 6.85} {"step": 2655, "epoch": 1, "loss": 0.0132, "lr": 1.7659802587503288e-05, "grad_norm": 4.0625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2656, "epoch": 1, "loss": 0.0203, "lr": 1.765787659828458e-05, "grad_norm": 2.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2657, "epoch": 1, "loss": 0.0029, "lr": 1.7655949921961437e-05, "grad_norm": 0.376953125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2658, "epoch": 1, "loss": 0.0005, "lr": 1.7654022558706738e-05, "grad_norm": 7.71875, "samples_per_sec": 10.3, "gpu_mem_gb": 6.82} {"step": 2659, "epoch": 1, "loss": 0.3098, "lr": 1.7652094508693412e-05, "grad_norm": 5.90625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2660, "epoch": 1, "loss": 0.0117, "lr": 1.765016577209446e-05, "grad_norm": 2.234375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2661, "epoch": 1, "loss": 0.1421, "lr": 1.7648236349082928e-05, "grad_norm": 2.203125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 2662, "epoch": 1, "loss": 0.0422, "lr": 1.764630623983194e-05, "grad_norm": 4.96875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2663, "epoch": 1, "loss": 0.001, "lr": 1.764437544451468e-05, "grad_norm": 2.796875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2664, "epoch": 1, "loss": 0.0004, "lr": 1.7642443963304377e-05, "grad_norm": 4.90625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2665, "epoch": 1, "loss": 0.0013, "lr": 1.7640511796374348e-05, "grad_norm": 4.46875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2666, "epoch": 1, "loss": 0.0027, "lr": 1.763857894389794e-05, "grad_norm": 2.609375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2667, "epoch": 1, "loss": 0.0571, "lr": 1.7636645406048597e-05, "grad_norm": 2.890625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 2668, "epoch": 1, "loss": 0.0443, "lr": 1.763471118299979e-05, "grad_norm": 5.25, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 2669, "epoch": 1, "loss": 0.1974, "lr": 1.7632776274925078e-05, "grad_norm": 4.53125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2670, "epoch": 1, "loss": 0.0155, "lr": 1.7630840681998068e-05, "grad_norm": 3.671875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2671, "epoch": 1, "loss": 0.2371, "lr": 1.762890440439243e-05, "grad_norm": 7.75, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 2672, "epoch": 1, "loss": 0.0876, "lr": 1.7626967442281894e-05, "grad_norm": 7.5, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2673, "epoch": 1, "loss": 0.127, "lr": 1.7625029795840262e-05, "grad_norm": 2.734375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 2674, "epoch": 1, "loss": 0.021, "lr": 1.7623091465241387e-05, "grad_norm": 0.9921875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2675, "epoch": 1, "loss": 0.0379, "lr": 1.7621152450659182e-05, "grad_norm": 4.21875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2676, "epoch": 1, "loss": 0.0008, "lr": 1.7619212752267628e-05, "grad_norm": 1.6796875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2677, "epoch": 1, "loss": 0.0369, "lr": 1.761727237024077e-05, "grad_norm": 2.359375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2678, "epoch": 1, "loss": 0.0004, "lr": 1.76153313047527e-05, "grad_norm": 4.1875, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 2679, "epoch": 1, "loss": 0.0188, "lr": 1.7613389555977586e-05, "grad_norm": 3.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2680, "epoch": 1, "loss": 0.0018, "lr": 1.761144712408965e-05, "grad_norm": 3.9375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2681, "epoch": 1, "loss": 0.0013, "lr": 1.7609504009263175e-05, "grad_norm": 5.71875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2682, "epoch": 1, "loss": 0.0957, "lr": 1.7607560211672515e-05, "grad_norm": 5.21875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2683, "epoch": 1, "loss": 0.0006, "lr": 1.760561573149207e-05, "grad_norm": 5.59375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2684, "epoch": 1, "loss": 0.0016, "lr": 1.760367056889631e-05, "grad_norm": 6.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2685, "epoch": 1, "loss": 0.0025, "lr": 1.760172472405977e-05, "grad_norm": 2.0625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 2686, "epoch": 1, "loss": 0.2872, "lr": 1.7599778197157036e-05, "grad_norm": 6.09375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2687, "epoch": 1, "loss": 0.0056, "lr": 1.7597830988362758e-05, "grad_norm": 4.28125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 2688, "epoch": 1, "loss": 0.0506, "lr": 1.759588309785166e-05, "grad_norm": 4.21875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2689, "epoch": 1, "loss": 0.0005, "lr": 1.7593934525798508e-05, "grad_norm": 1.9453125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2690, "epoch": 1, "loss": 0.0023, "lr": 1.759198527237814e-05, "grad_norm": 6.34375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2691, "epoch": 1, "loss": 0.1897, "lr": 1.7590035337765455e-05, "grad_norm": 5.3125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2692, "epoch": 1, "loss": 0.0478, "lr": 1.7588084722135413e-05, "grad_norm": 6.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 2693, "epoch": 1, "loss": 0.1252, "lr": 1.758613342566303e-05, "grad_norm": 6.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2694, "epoch": 1, "loss": 0.0871, "lr": 1.758418144852338e-05, "grad_norm": 4.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2695, "epoch": 1, "loss": 0.0324, "lr": 1.758222879089162e-05, "grad_norm": 5.25, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 2696, "epoch": 1, "loss": 0.0219, "lr": 1.758027545294294e-05, "grad_norm": 2.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2697, "epoch": 1, "loss": 0.1057, "lr": 1.757832143485261e-05, "grad_norm": 5.46875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 2698, "epoch": 1, "loss": 0.1147, "lr": 1.757636673679595e-05, "grad_norm": 5.0625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 2699, "epoch": 1, "loss": 0.0011, "lr": 1.7574411358948347e-05, "grad_norm": 4.59375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2700, "epoch": 1, "loss": 0.1123, "lr": 1.757245530148525e-05, "grad_norm": 3.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2701, "epoch": 1, "loss": 0.0011, "lr": 1.7570498564582166e-05, "grad_norm": 0.27734375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2702, "epoch": 1, "loss": 0.0011, "lr": 1.7568541148414665e-05, "grad_norm": 1.0703125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 2703, "epoch": 1, "loss": 0.2906, "lr": 1.7566583053158367e-05, "grad_norm": 4.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2704, "epoch": 1, "loss": 0.0234, "lr": 1.7564624278988977e-05, "grad_norm": 4.8125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2705, "epoch": 1, "loss": 0.082, "lr": 1.7562664826082233e-05, "grad_norm": 3.375, "samples_per_sec": 10.31, "gpu_mem_gb": 6.85} {"step": 2706, "epoch": 1, "loss": 0.0875, "lr": 1.756070469461396e-05, "grad_norm": 6.3125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2707, "epoch": 1, "loss": 0.1282, "lr": 1.755874388476002e-05, "grad_norm": 6.0625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2708, "epoch": 1, "loss": 0.0027, "lr": 1.7556782396696356e-05, "grad_norm": 4.78125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2709, "epoch": 1, "loss": 0.0031, "lr": 1.7554820230598958e-05, "grad_norm": 1.3671875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2710, "epoch": 1, "loss": 0.0012, "lr": 1.7552857386643878e-05, "grad_norm": 3.96875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2711, "epoch": 1, "loss": 0.019, "lr": 1.755089386500724e-05, "grad_norm": 2.6875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2712, "epoch": 1, "loss": 0.0524, "lr": 1.7548929665865217e-05, "grad_norm": 4.21875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 2713, "epoch": 1, "loss": 0.1296, "lr": 1.7546964789394047e-05, "grad_norm": 6.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 2714, "epoch": 1, "loss": 0.0794, "lr": 1.7544999235770035e-05, "grad_norm": 2.390625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2715, "epoch": 1, "loss": 0.0002, "lr": 1.7543033005169532e-05, "grad_norm": 5.53125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 2716, "epoch": 1, "loss": 0.0823, "lr": 1.7541066097768965e-05, "grad_norm": 1.734375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 2717, "epoch": 1, "loss": 0.0013, "lr": 1.753909851374481e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2718, "epoch": 1, "loss": 0.3286, "lr": 1.7537130253273613e-05, "grad_norm": 7.84375, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 2719, "epoch": 1, "loss": 0.0584, "lr": 1.7535161316531975e-05, "grad_norm": 3.8125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2720, "epoch": 1, "loss": 0.0016, "lr": 1.7533191703696557e-05, "grad_norm": 3.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 2721, "epoch": 1, "loss": 0.0114, "lr": 1.7531221414944087e-05, "grad_norm": 3.0625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2722, "epoch": 1, "loss": 0.0087, "lr": 1.7529250450451347e-05, "grad_norm": 4.71875, "samples_per_sec": 10.43, "gpu_mem_gb": 6.82} {"step": 2723, "epoch": 1, "loss": 0.085, "lr": 1.752727881039518e-05, "grad_norm": 8.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2724, "epoch": 1, "loss": 0.0755, "lr": 1.7525306494952498e-05, "grad_norm": 4.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2725, "epoch": 1, "loss": 0.0137, "lr": 1.7523333504300265e-05, "grad_norm": 5.25, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 2726, "epoch": 1, "loss": 0.0971, "lr": 1.7521359838615506e-05, "grad_norm": 7.34375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2727, "epoch": 1, "loss": 0.0573, "lr": 1.7519385498075308e-05, "grad_norm": 4.40625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2728, "epoch": 1, "loss": 0.0035, "lr": 1.751741048285682e-05, "grad_norm": 3.59375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2729, "epoch": 1, "loss": 0.1199, "lr": 1.7515434793137252e-05, "grad_norm": 3.734375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 2730, "epoch": 1, "loss": 0.006, "lr": 1.751345842909387e-05, "grad_norm": 2.234375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2731, "epoch": 1, "loss": 0.0042, "lr": 1.751148139090401e-05, "grad_norm": 5.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2732, "epoch": 1, "loss": 0.0474, "lr": 1.7509503678745055e-05, "grad_norm": 3.921875, "samples_per_sec": 10.29, "gpu_mem_gb": 6.82} {"step": 2733, "epoch": 1, "loss": 0.0006, "lr": 1.7507525292794458e-05, "grad_norm": 5.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2734, "epoch": 1, "loss": 0.0099, "lr": 1.7505546233229732e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2735, "epoch": 1, "loss": 0.0165, "lr": 1.7503566500228446e-05, "grad_norm": 4.3125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2736, "epoch": 1, "loss": 0.0804, "lr": 1.7501586093968236e-05, "grad_norm": 6.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2737, "epoch": 1, "loss": 0.0003, "lr": 1.7499605014626786e-05, "grad_norm": 7.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2738, "epoch": 1, "loss": 0.0096, "lr": 1.749762326238186e-05, "grad_norm": 3.359375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2739, "epoch": 1, "loss": 0.1424, "lr": 1.7495640837411265e-05, "grad_norm": 8.1875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 2740, "epoch": 1, "loss": 0.0136, "lr": 1.749365773989287e-05, "grad_norm": 2.328125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 2741, "epoch": 1, "loss": 0.0029, "lr": 1.7491673970004615e-05, "grad_norm": 5.03125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2742, "epoch": 1, "loss": 0.042, "lr": 1.7489689527924492e-05, "grad_norm": 3.1875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2743, "epoch": 1, "loss": 0.1006, "lr": 1.7487704413830557e-05, "grad_norm": 8.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2744, "epoch": 1, "loss": 0.0276, "lr": 1.7485718627900926e-05, "grad_norm": 3.890625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2745, "epoch": 1, "loss": 0.0031, "lr": 1.7483732170313766e-05, "grad_norm": 2.90625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 2746, "epoch": 1, "loss": 0.037, "lr": 1.7481745041247322e-05, "grad_norm": 3.453125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 2747, "epoch": 1, "loss": 0.0009, "lr": 1.7479757240879887e-05, "grad_norm": 0.58984375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2748, "epoch": 1, "loss": 0.06, "lr": 1.747776876938981e-05, "grad_norm": 7.03125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 2749, "epoch": 1, "loss": 0.1363, "lr": 1.7475779626955514e-05, "grad_norm": 3.546875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2750, "epoch": 1, "loss": 0.0055, "lr": 1.747378981375548e-05, "grad_norm": 1.453125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2751, "epoch": 1, "loss": 0.1189, "lr": 1.7471799329968232e-05, "grad_norm": 3.1875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 2752, "epoch": 1, "loss": 0.0013, "lr": 1.7469808175772373e-05, "grad_norm": 2.734375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2753, "epoch": 1, "loss": 0.0049, "lr": 1.746781635134656e-05, "grad_norm": 3.578125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2754, "epoch": 1, "loss": 0.1613, "lr": 1.7465823856869505e-05, "grad_norm": 4.46875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2755, "epoch": 1, "loss": 0.0008, "lr": 1.7463830692519995e-05, "grad_norm": 3.84375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2756, "epoch": 1, "loss": 0.0004, "lr": 1.7461836858476858e-05, "grad_norm": 5.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2757, "epoch": 1, "loss": 0.0084, "lr": 1.7459842354918995e-05, "grad_norm": 0.8984375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 2758, "epoch": 1, "loss": 0.1082, "lr": 1.7457847182025357e-05, "grad_norm": 5.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2759, "epoch": 1, "loss": 0.1248, "lr": 1.7455851339974975e-05, "grad_norm": 3.015625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 2760, "epoch": 1, "loss": 0.0354, "lr": 1.7453854828946913e-05, "grad_norm": 3.96875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2761, "epoch": 1, "loss": 0.0601, "lr": 1.745185764912031e-05, "grad_norm": 4.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2762, "epoch": 1, "loss": 0.027, "lr": 1.744985980067437e-05, "grad_norm": 1.6171875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 2763, "epoch": 1, "loss": 0.0188, "lr": 1.7447861283788346e-05, "grad_norm": 5.53125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 2764, "epoch": 1, "loss": 0.0202, "lr": 1.744586209864156e-05, "grad_norm": 4.15625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2765, "epoch": 1, "loss": 0.2838, "lr": 1.744386224541338e-05, "grad_norm": 5.0, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 2766, "epoch": 1, "loss": 0.046, "lr": 1.7441861724283253e-05, "grad_norm": 3.203125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.85} {"step": 2767, "epoch": 1, "loss": 0.0015, "lr": 1.743986053543067e-05, "grad_norm": 1.6015625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2768, "epoch": 1, "loss": 0.0005, "lr": 1.743785867903519e-05, "grad_norm": 2.328125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 2769, "epoch": 1, "loss": 0.0023, "lr": 1.7435856155276426e-05, "grad_norm": 3.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 2770, "epoch": 1, "loss": 0.0004, "lr": 1.7433852964334062e-05, "grad_norm": 5.4375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 2771, "epoch": 1, "loss": 0.0241, "lr": 1.7431849106387835e-05, "grad_norm": 2.765625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2772, "epoch": 1, "loss": 0.0002, "lr": 1.7429844581617532e-05, "grad_norm": 1.40625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2773, "epoch": 1, "loss": 0.0307, "lr": 1.7427839390203016e-05, "grad_norm": 7.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2774, "epoch": 1, "loss": 0.0002, "lr": 1.7425833532324207e-05, "grad_norm": 1.875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2775, "epoch": 1, "loss": 0.0016, "lr": 1.742382700816107e-05, "grad_norm": 1.4921875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2776, "epoch": 1, "loss": 0.0006, "lr": 1.742181981789365e-05, "grad_norm": 0.671875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2777, "epoch": 1, "loss": 0.1225, "lr": 1.7419811961702037e-05, "grad_norm": 2.296875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2778, "epoch": 1, "loss": 0.021, "lr": 1.7417803439766394e-05, "grad_norm": 4.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 2779, "epoch": 1, "loss": 0.107, "lr": 1.7415794252266924e-05, "grad_norm": 11.1875, "samples_per_sec": 10.25, "gpu_mem_gb": 6.83} {"step": 2780, "epoch": 1, "loss": 0.0006, "lr": 1.7413784399383914e-05, "grad_norm": 1.828125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2781, "epoch": 1, "loss": 0.1949, "lr": 1.741177388129769e-05, "grad_norm": 5.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2782, "epoch": 1, "loss": 0.0013, "lr": 1.7409762698188652e-05, "grad_norm": 3.03125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 2783, "epoch": 1, "loss": 0.1082, "lr": 1.740775085023725e-05, "grad_norm": 5.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2784, "epoch": 1, "loss": 0.1099, "lr": 1.7405738337623998e-05, "grad_norm": 3.421875, "samples_per_sec": 8.89, "gpu_mem_gb": 6.84} {"step": 2785, "epoch": 1, "loss": 0.0365, "lr": 1.740372516052947e-05, "grad_norm": 3.1875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2786, "epoch": 1, "loss": 0.0165, "lr": 1.74017113191343e-05, "grad_norm": 1.5625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 2787, "epoch": 1, "loss": 0.0246, "lr": 1.7399696813619172e-05, "grad_norm": 2.515625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2788, "epoch": 1, "loss": 0.0007, "lr": 1.739768164416485e-05, "grad_norm": 4.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2789, "epoch": 1, "loss": 0.0007, "lr": 1.7395665810952137e-05, "grad_norm": 4.3125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2790, "epoch": 1, "loss": 0.1501, "lr": 1.7393649314161908e-05, "grad_norm": 3.28125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 2791, "epoch": 1, "loss": 0.0926, "lr": 1.7391632153975095e-05, "grad_norm": 4.03125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2792, "epoch": 1, "loss": 0.0202, "lr": 1.738961433057268e-05, "grad_norm": 2.625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.85} {"step": 2793, "epoch": 1, "loss": 0.0035, "lr": 1.738759584413572e-05, "grad_norm": 2.609375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2794, "epoch": 1, "loss": 0.0023, "lr": 1.7385576694845324e-05, "grad_norm": 1.3671875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2795, "epoch": 1, "loss": 0.006, "lr": 1.7383556882882658e-05, "grad_norm": 6.9375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2796, "epoch": 1, "loss": 0.0414, "lr": 1.7381536408428948e-05, "grad_norm": 6.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2797, "epoch": 1, "loss": 0.2102, "lr": 1.7379515271665485e-05, "grad_norm": 4.4375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 2798, "epoch": 1, "loss": 0.1169, "lr": 1.737749347277362e-05, "grad_norm": 9.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2799, "epoch": 1, "loss": 0.037, "lr": 1.737547101193475e-05, "grad_norm": 5.78125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 2800, "epoch": 1, "loss": 0.0032, "lr": 1.7373447889330344e-05, "grad_norm": 4.78125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 2801, "epoch": 1, "loss": 0.047, "lr": 1.737142410514193e-05, "grad_norm": 1.046875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2802, "epoch": 1, "loss": 0.1094, "lr": 1.736939965955109e-05, "grad_norm": 4.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2803, "epoch": 1, "loss": 0.0886, "lr": 1.736737455273947e-05, "grad_norm": 7.5625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2804, "epoch": 1, "loss": 0.0825, "lr": 1.7365348784888768e-05, "grad_norm": 5.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2805, "epoch": 1, "loss": 0.1408, "lr": 1.7363322356180753e-05, "grad_norm": 6.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2806, "epoch": 1, "loss": 0.0244, "lr": 1.736129526679724e-05, "grad_norm": 6.21875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2807, "epoch": 1, "loss": 0.0025, "lr": 1.7359267516920116e-05, "grad_norm": 3.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2808, "epoch": 1, "loss": 0.1825, "lr": 1.735723910673132e-05, "grad_norm": 3.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2809, "epoch": 1, "loss": 0.003, "lr": 1.7355210036412845e-05, "grad_norm": 2.234375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.85} {"step": 2810, "epoch": 1, "loss": 0.0008, "lr": 1.735318030614676e-05, "grad_norm": 7.09375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 2811, "epoch": 1, "loss": 0.0923, "lr": 1.7351149916115174e-05, "grad_norm": 5.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2812, "epoch": 1, "loss": 0.0028, "lr": 1.734911886650027e-05, "grad_norm": 0.5703125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2813, "epoch": 1, "loss": 0.0455, "lr": 1.7347087157484282e-05, "grad_norm": 6.0625, "samples_per_sec": 10.34, "gpu_mem_gb": 6.84} {"step": 2814, "epoch": 1, "loss": 0.0006, "lr": 1.7345054789249503e-05, "grad_norm": 3.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 2815, "epoch": 1, "loss": 0.028, "lr": 1.7343021761978294e-05, "grad_norm": 1.4140625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2816, "epoch": 1, "loss": 0.0148, "lr": 1.7340988075853064e-05, "grad_norm": 3.15625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.85} {"step": 2817, "epoch": 1, "loss": 0.0013, "lr": 1.7338953731056287e-05, "grad_norm": 5.4375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 2818, "epoch": 1, "loss": 0.2183, "lr": 1.7336918727770494e-05, "grad_norm": 4.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2819, "epoch": 1, "loss": 0.0012, "lr": 1.733488306617828e-05, "grad_norm": 0.259765625, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 2820, "epoch": 1, "loss": 0.0019, "lr": 1.733284674646229e-05, "grad_norm": 2.484375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2821, "epoch": 1, "loss": 0.0608, "lr": 1.7330809768805234e-05, "grad_norm": 3.484375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2822, "epoch": 1, "loss": 0.0133, "lr": 1.732877213338988e-05, "grad_norm": 2.703125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2823, "epoch": 1, "loss": 0.0209, "lr": 1.732673384039906e-05, "grad_norm": 5.21875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 2824, "epoch": 1, "loss": 0.0054, "lr": 1.7324694890015655e-05, "grad_norm": 3.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 2825, "epoch": 1, "loss": 0.0124, "lr": 1.7322655282422617e-05, "grad_norm": 3.234375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 2826, "epoch": 1, "loss": 0.0593, "lr": 1.732061501780294e-05, "grad_norm": 6.40625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 2827, "epoch": 1, "loss": 0.0932, "lr": 1.7318574096339694e-05, "grad_norm": 4.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2828, "epoch": 1, "loss": 0.0216, "lr": 1.7316532518216004e-05, "grad_norm": 2.984375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2829, "epoch": 1, "loss": 0.0075, "lr": 1.731449028361504e-05, "grad_norm": 4.03125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 2830, "epoch": 1, "loss": 0.0428, "lr": 1.7312447392720052e-05, "grad_norm": 5.59375, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 2831, "epoch": 1, "loss": 0.0002, "lr": 1.731040384571433e-05, "grad_norm": 2.140625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2832, "epoch": 1, "loss": 0.1081, "lr": 1.730835964278124e-05, "grad_norm": 7.71875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 2833, "epoch": 1, "loss": 0.1207, "lr": 1.73063147841042e-05, "grad_norm": 3.109375, "samples_per_sec": 10.24, "gpu_mem_gb": 6.85} {"step": 2834, "epoch": 1, "loss": 0.2005, "lr": 1.7304269269866677e-05, "grad_norm": 4.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2835, "epoch": 1, "loss": 0.1646, "lr": 1.7302223100252205e-05, "grad_norm": 3.15625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2836, "epoch": 1, "loss": 0.0243, "lr": 1.7300176275444386e-05, "grad_norm": 0.96875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2837, "epoch": 1, "loss": 0.0012, "lr": 1.7298128795626863e-05, "grad_norm": 0.1474609375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 2838, "epoch": 1, "loss": 0.0202, "lr": 1.7296080660983354e-05, "grad_norm": 5.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2839, "epoch": 1, "loss": 0.073, "lr": 1.729403187169762e-05, "grad_norm": 2.703125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 2840, "epoch": 1, "loss": 0.1531, "lr": 1.7291982427953497e-05, "grad_norm": 3.765625, "samples_per_sec": 10.33, "gpu_mem_gb": 6.84} {"step": 2841, "epoch": 1, "loss": 0.0006, "lr": 1.7289932329934864e-05, "grad_norm": 4.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 2842, "epoch": 1, "loss": 0.0688, "lr": 1.7287881577825672e-05, "grad_norm": 4.78125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2843, "epoch": 1, "loss": 0.1362, "lr": 1.7285830171809925e-05, "grad_norm": 4.3125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 2844, "epoch": 1, "loss": 0.0042, "lr": 1.7283778112071683e-05, "grad_norm": 4.09375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 2845, "epoch": 1, "loss": 0.0665, "lr": 1.728172539879507e-05, "grad_norm": 7.03125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2846, "epoch": 1, "loss": 0.0927, "lr": 1.7279672032164267e-05, "grad_norm": 7.4375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 2847, "epoch": 1, "loss": 0.0022, "lr": 1.7277618012363508e-05, "grad_norm": 1.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2848, "epoch": 1, "loss": 0.0, "lr": 1.727556333957709e-05, "grad_norm": 3.3125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2849, "epoch": 1, "loss": 0.0022, "lr": 1.7273508013989374e-05, "grad_norm": 6.875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 2850, "epoch": 1, "loss": 0.0005, "lr": 1.7271452035784775e-05, "grad_norm": 5.84375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2851, "epoch": 1, "loss": 0.2562, "lr": 1.726939540514776e-05, "grad_norm": 6.96875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 2852, "epoch": 1, "loss": 0.0083, "lr": 1.7267338122262867e-05, "grad_norm": 2.828125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 2853, "epoch": 1, "loss": 0.051, "lr": 1.726528018731468e-05, "grad_norm": 6.5625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 2854, "epoch": 1, "loss": 0.2205, "lr": 1.7263221600487852e-05, "grad_norm": 5.625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.84} {"step": 2855, "epoch": 1, "loss": 0.0371, "lr": 1.7261162361967088e-05, "grad_norm": 8.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2856, "epoch": 1, "loss": 0.3405, "lr": 1.7259102471937154e-05, "grad_norm": 7.1875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 2857, "epoch": 1, "loss": 0.005, "lr": 1.7257041930582878e-05, "grad_norm": 4.59375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 2858, "epoch": 1, "loss": 0.0003, "lr": 1.7254980738089133e-05, "grad_norm": 3.671875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2859, "epoch": 1, "loss": 0.041, "lr": 1.725291889464087e-05, "grad_norm": 5.5, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 2860, "epoch": 1, "loss": 0.0396, "lr": 1.7250856400423083e-05, "grad_norm": 6.0625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 2861, "epoch": 1, "loss": 0.0143, "lr": 1.724879325562083e-05, "grad_norm": 2.546875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2862, "epoch": 1, "loss": 0.1509, "lr": 1.7246729460419224e-05, "grad_norm": 5.28125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2863, "epoch": 1, "loss": 0.0173, "lr": 1.724466501500345e-05, "grad_norm": 5.71875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 2864, "epoch": 1, "loss": 0.0311, "lr": 1.7242599919558728e-05, "grad_norm": 3.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2865, "epoch": 1, "loss": 0.0009, "lr": 1.724053417427036e-05, "grad_norm": 4.625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 2866, "epoch": 1, "loss": 0.3129, "lr": 1.7238467779323685e-05, "grad_norm": 6.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 2867, "epoch": 1, "loss": 0.004, "lr": 1.7236400734904118e-05, "grad_norm": 1.8515625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2868, "epoch": 1, "loss": 0.2154, "lr": 1.7234333041197127e-05, "grad_norm": 3.578125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2869, "epoch": 1, "loss": 0.0642, "lr": 1.723226469838823e-05, "grad_norm": 1.859375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2870, "epoch": 1, "loss": 0.0253, "lr": 1.723019570666301e-05, "grad_norm": 4.40625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 2871, "epoch": 1, "loss": 0.0046, "lr": 1.722812606620711e-05, "grad_norm": 3.453125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2872, "epoch": 1, "loss": 0.0277, "lr": 1.722605577720623e-05, "grad_norm": 3.234375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 2873, "epoch": 1, "loss": 0.0304, "lr": 1.7223984839846123e-05, "grad_norm": 5.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 2874, "epoch": 1, "loss": 0.0065, "lr": 1.7221913254312604e-05, "grad_norm": 2.5, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 2875, "epoch": 1, "loss": 0.0007, "lr": 1.721984102079155e-05, "grad_norm": 3.125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 2876, "epoch": 1, "loss": 0.0711, "lr": 1.7217768139468897e-05, "grad_norm": 3.765625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 2877, "epoch": 1, "loss": 0.0905, "lr": 1.7215694610530624e-05, "grad_norm": 5.09375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2878, "epoch": 1, "loss": 0.1315, "lr": 1.7213620434162782e-05, "grad_norm": 9.8125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2879, "epoch": 1, "loss": 0.0863, "lr": 1.7211545610551485e-05, "grad_norm": 3.703125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 2880, "epoch": 1, "loss": 0.1477, "lr": 1.7209470139882887e-05, "grad_norm": 5.875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 2881, "epoch": 1, "loss": 0.1926, "lr": 1.7207394022343216e-05, "grad_norm": 3.96875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 2882, "epoch": 1, "loss": 0.2142, "lr": 1.7205317258118745e-05, "grad_norm": 3.921875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 2883, "epoch": 1, "loss": 0.0053, "lr": 1.720323984739582e-05, "grad_norm": 6.46875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 2884, "epoch": 1, "loss": 0.0006, "lr": 1.7201161790360833e-05, "grad_norm": 4.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2885, "epoch": 1, "loss": 0.0392, "lr": 1.7199083087200242e-05, "grad_norm": 3.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2886, "epoch": 1, "loss": 0.0004, "lr": 1.7197003738100556e-05, "grad_norm": 3.515625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2887, "epoch": 1, "loss": 0.0017, "lr": 1.719492374324834e-05, "grad_norm": 5.96875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 2888, "epoch": 1, "loss": 0.2453, "lr": 1.719284310283023e-05, "grad_norm": 7.28125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2889, "epoch": 1, "loss": 0.1073, "lr": 1.719076181703291e-05, "grad_norm": 5.3125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2890, "epoch": 1, "loss": 0.2016, "lr": 1.718867988604312e-05, "grad_norm": 6.1875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2891, "epoch": 1, "loss": 0.2343, "lr": 1.7186597310047663e-05, "grad_norm": 4.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2892, "epoch": 1, "loss": 0.0301, "lr": 1.7184514089233403e-05, "grad_norm": 5.1875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2893, "epoch": 1, "loss": 0.0022, "lr": 1.7182430223787255e-05, "grad_norm": 6.09375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 2894, "epoch": 1, "loss": 0.0906, "lr": 1.718034571389619e-05, "grad_norm": 4.5, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2895, "epoch": 1, "loss": 0.0016, "lr": 1.7178260559747247e-05, "grad_norm": 4.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 2896, "epoch": 1, "loss": 0.0001, "lr": 1.717617476152751e-05, "grad_norm": 4.71875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 2897, "epoch": 1, "loss": 0.2249, "lr": 1.7174088319424137e-05, "grad_norm": 9.5, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2898, "epoch": 1, "loss": 0.0568, "lr": 1.7172001233624325e-05, "grad_norm": 3.078125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 2899, "epoch": 1, "loss": 0.0395, "lr": 1.716991350431534e-05, "grad_norm": 5.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2900, "epoch": 1, "loss": 0.0038, "lr": 1.7167825131684516e-05, "grad_norm": 1.8515625, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 2901, "epoch": 1, "loss": 0.0309, "lr": 1.7165736115919216e-05, "grad_norm": 2.703125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 2902, "epoch": 1, "loss": 0.0195, "lr": 1.7163646457206886e-05, "grad_norm": 2.265625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2903, "epoch": 1, "loss": 0.0647, "lr": 1.716155615573502e-05, "grad_norm": 5.3125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.84} {"step": 2904, "epoch": 1, "loss": 0.0008, "lr": 1.7159465211691167e-05, "grad_norm": 1.7890625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2905, "epoch": 1, "loss": 0.0001, "lr": 1.7157373625262943e-05, "grad_norm": 0.6328125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2906, "epoch": 1, "loss": 0.0025, "lr": 1.7155281396638014e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 2907, "epoch": 1, "loss": 0.0002, "lr": 1.7153188526004103e-05, "grad_norm": 13.0625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 2908, "epoch": 1, "loss": 0.0059, "lr": 1.7151095013548996e-05, "grad_norm": 4.40625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2909, "epoch": 1, "loss": 0.0017, "lr": 1.714900085946053e-05, "grad_norm": 1.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2910, "epoch": 1, "loss": 0.0019, "lr": 1.7146906063926613e-05, "grad_norm": 2.4375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 2911, "epoch": 1, "loss": 0.1246, "lr": 1.7144810627135188e-05, "grad_norm": 6.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2912, "epoch": 1, "loss": 0.0283, "lr": 1.714271454927428e-05, "grad_norm": 6.09375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2913, "epoch": 1, "loss": 0.0023, "lr": 1.714061783053195e-05, "grad_norm": 4.125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2914, "epoch": 1, "loss": 0.0079, "lr": 1.7138520471096334e-05, "grad_norm": 4.6875, "samples_per_sec": 10.31, "gpu_mem_gb": 6.82} {"step": 2915, "epoch": 1, "loss": 0.0076, "lr": 1.7136422471155613e-05, "grad_norm": 4.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2916, "epoch": 1, "loss": 0.002, "lr": 1.7134323830898036e-05, "grad_norm": 4.09375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2917, "epoch": 1, "loss": 0.2044, "lr": 1.7132224550511897e-05, "grad_norm": 7.09375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.84} {"step": 2918, "epoch": 1, "loss": 0.0001, "lr": 1.713012463018556e-05, "grad_norm": 1.7109375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2919, "epoch": 1, "loss": 0.1111, "lr": 1.7128024070107438e-05, "grad_norm": 3.65625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2920, "epoch": 1, "loss": 0.0218, "lr": 1.7125922870466004e-05, "grad_norm": 6.71875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 2921, "epoch": 1, "loss": 0.002, "lr": 1.7123821031449792e-05, "grad_norm": 1.9609375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 2922, "epoch": 1, "loss": 0.0013, "lr": 1.7121718553247384e-05, "grad_norm": 0.96484375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2923, "epoch": 1, "loss": 0.0949, "lr": 1.711961543604743e-05, "grad_norm": 6.40625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 2924, "epoch": 1, "loss": 0.0044, "lr": 1.711751168003863e-05, "grad_norm": 6.53125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 2925, "epoch": 1, "loss": 0.1121, "lr": 1.7115407285409745e-05, "grad_norm": 4.8125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 2926, "epoch": 1, "loss": 0.1382, "lr": 1.7113302252349594e-05, "grad_norm": 3.03125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2927, "epoch": 1, "loss": 0.0674, "lr": 1.711119658104705e-05, "grad_norm": 5.4375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 2928, "epoch": 1, "loss": 0.1933, "lr": 1.7109090271691044e-05, "grad_norm": 5.6875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 2929, "epoch": 1, "loss": 0.0024, "lr": 1.7106983324470564e-05, "grad_norm": 1.171875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 2930, "epoch": 1, "loss": 0.0832, "lr": 1.7104875739574656e-05, "grad_norm": 5.0625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 2931, "epoch": 1, "loss": 0.0707, "lr": 1.710276751719243e-05, "grad_norm": 5.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2932, "epoch": 1, "loss": 0.037, "lr": 1.7100658657513038e-05, "grad_norm": 3.28125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 2933, "epoch": 1, "loss": 0.0035, "lr": 1.70985491607257e-05, "grad_norm": 5.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 2934, "epoch": 1, "loss": 0.0025, "lr": 1.7096439027019695e-05, "grad_norm": 0.390625, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 2935, "epoch": 1, "loss": 0.1636, "lr": 1.7094328256584355e-05, "grad_norm": 3.4375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2936, "epoch": 1, "loss": 0.0017, "lr": 1.709221684960906e-05, "grad_norm": 3.328125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2937, "epoch": 1, "loss": 0.0009, "lr": 1.709010480628327e-05, "grad_norm": 9.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 2938, "epoch": 1, "loss": 0.2917, "lr": 1.7087992126796483e-05, "grad_norm": 5.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2939, "epoch": 1, "loss": 0.3596, "lr": 1.7085878811338255e-05, "grad_norm": 6.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2940, "epoch": 1, "loss": 0.0002, "lr": 1.7083764860098206e-05, "grad_norm": 4.1875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2941, "epoch": 1, "loss": 0.006, "lr": 1.708165027326601e-05, "grad_norm": 2.859375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 2942, "epoch": 1, "loss": 0.0008, "lr": 1.7079535051031408e-05, "grad_norm": 4.96875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 2943, "epoch": 1, "loss": 0.0008, "lr": 1.7077419193584178e-05, "grad_norm": 4.28125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 2944, "epoch": 1, "loss": 0.2545, "lr": 1.7075302701114166e-05, "grad_norm": 7.125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 2945, "epoch": 1, "loss": 0.0022, "lr": 1.707318557381128e-05, "grad_norm": 6.1875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 2946, "epoch": 1, "loss": 0.0449, "lr": 1.7071067811865477e-05, "grad_norm": 5.90625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2947, "epoch": 1, "loss": 0.0788, "lr": 1.7068949415466773e-05, "grad_norm": 2.421875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2948, "epoch": 1, "loss": 0.0018, "lr": 1.706683038480524e-05, "grad_norm": 3.09375, "samples_per_sec": 10.29, "gpu_mem_gb": 6.83} {"step": 2949, "epoch": 1, "loss": 0.1384, "lr": 1.706471072007102e-05, "grad_norm": 6.0625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 2950, "epoch": 1, "loss": 0.186, "lr": 1.7062590421454285e-05, "grad_norm": 6.125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 2951, "epoch": 1, "loss": 0.059, "lr": 1.7060469489145283e-05, "grad_norm": 3.484375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 2952, "epoch": 1, "loss": 0.0381, "lr": 1.7058347923334316e-05, "grad_norm": 5.0625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 2953, "epoch": 1, "loss": 0.0022, "lr": 1.7056225724211746e-05, "grad_norm": 3.28125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 2954, "epoch": 1, "loss": 0.0028, "lr": 1.7054102891967986e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 2955, "epoch": 1, "loss": 0.0037, "lr": 1.7051979426793506e-05, "grad_norm": 4.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 2956, "epoch": 1, "loss": 0.2329, "lr": 1.704985532887884e-05, "grad_norm": 5.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 2957, "epoch": 1, "loss": 0.002, "lr": 1.704773059841456e-05, "grad_norm": 4.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 2958, "epoch": 1, "loss": 0.0005, "lr": 1.7045605235591317e-05, "grad_norm": 4.375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2959, "epoch": 1, "loss": 0.0685, "lr": 1.7043479240599813e-05, "grad_norm": 4.5, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2960, "epoch": 1, "loss": 0.1146, "lr": 1.7041352613630794e-05, "grad_norm": 4.40625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 2961, "epoch": 1, "loss": 0.0006, "lr": 1.7039225354875078e-05, "grad_norm": 0.22265625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 2962, "epoch": 1, "loss": 0.1527, "lr": 1.7037097464523532e-05, "grad_norm": 8.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2963, "epoch": 1, "loss": 0.0039, "lr": 1.7034968942767082e-05, "grad_norm": 5.03125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 2964, "epoch": 1, "loss": 0.0001, "lr": 1.7032839789796712e-05, "grad_norm": 1.28125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 2965, "epoch": 1, "loss": 0.0037, "lr": 1.7030710005803453e-05, "grad_norm": 2.296875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2966, "epoch": 1, "loss": 0.0835, "lr": 1.7028579590978412e-05, "grad_norm": 5.46875, "samples_per_sec": 11.02, "gpu_mem_gb": 6.82} {"step": 2967, "epoch": 1, "loss": 0.1455, "lr": 1.702644854551273e-05, "grad_norm": 4.65625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 2968, "epoch": 1, "loss": 0.001, "lr": 1.702431686959762e-05, "grad_norm": 2.703125, "samples_per_sec": 10.45, "gpu_mem_gb": 6.82} {"step": 2969, "epoch": 1, "loss": 0.0108, "lr": 1.702218456342435e-05, "grad_norm": 5.0, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 2970, "epoch": 1, "loss": 0.0922, "lr": 1.7020051627184234e-05, "grad_norm": 2.484375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2971, "epoch": 1, "loss": 0.0133, "lr": 1.7017918061068657e-05, "grad_norm": 1.0859375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 2972, "epoch": 1, "loss": 0.0261, "lr": 1.7015783865269052e-05, "grad_norm": 6.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2973, "epoch": 1, "loss": 0.0449, "lr": 1.701364903997691e-05, "grad_norm": 5.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 2974, "epoch": 1, "loss": 0.2757, "lr": 1.7011513585383776e-05, "grad_norm": 5.59375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2975, "epoch": 1, "loss": 0.0512, "lr": 1.7009377501681257e-05, "grad_norm": 3.375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 2976, "epoch": 1, "loss": 0.0934, "lr": 1.7007240789061014e-05, "grad_norm": 5.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 2977, "epoch": 1, "loss": 0.1108, "lr": 1.700510344771476e-05, "grad_norm": 6.25, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2978, "epoch": 1, "loss": 0.1791, "lr": 1.7002965477834275e-05, "grad_norm": 5.21875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2979, "epoch": 1, "loss": 0.0487, "lr": 1.7000826879611382e-05, "grad_norm": 3.109375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 2980, "epoch": 1, "loss": 0.0532, "lr": 1.699868765323797e-05, "grad_norm": 6.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 2981, "epoch": 1, "loss": 0.0603, "lr": 1.699654779890598e-05, "grad_norm": 7.625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 2982, "epoch": 1, "loss": 0.0133, "lr": 1.6994407316807412e-05, "grad_norm": 2.78125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2983, "epoch": 1, "loss": 0.0023, "lr": 1.699226620713432e-05, "grad_norm": 2.703125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 2984, "epoch": 1, "loss": 0.023, "lr": 1.699012447007882e-05, "grad_norm": 1.2890625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2985, "epoch": 1, "loss": 0.0726, "lr": 1.698798210583308e-05, "grad_norm": 5.46875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.84} {"step": 2986, "epoch": 1, "loss": 0.2594, "lr": 1.6985839114589313e-05, "grad_norm": 4.53125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 2987, "epoch": 1, "loss": 0.0543, "lr": 1.6983695496539812e-05, "grad_norm": 2.890625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 2988, "epoch": 1, "loss": 0.0039, "lr": 1.6981551251876905e-05, "grad_norm": 2.171875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 2989, "epoch": 1, "loss": 0.028, "lr": 1.697940638079299e-05, "grad_norm": 4.84375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 2990, "epoch": 1, "loss": 0.0429, "lr": 1.6977260883480516e-05, "grad_norm": 3.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 2991, "epoch": 1, "loss": 0.1023, "lr": 1.6975114760131983e-05, "grad_norm": 3.171875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 2992, "epoch": 1, "loss": 0.1215, "lr": 1.6972968010939953e-05, "grad_norm": 5.03125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 2993, "epoch": 1, "loss": 0.0008, "lr": 1.6970820636097052e-05, "grad_norm": 3.421875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 2994, "epoch": 1, "loss": 0.0306, "lr": 1.6968672635795946e-05, "grad_norm": 4.0, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 2995, "epoch": 1, "loss": 0.0005, "lr": 1.6966524010229366e-05, "grad_norm": 4.84375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 2996, "epoch": 1, "loss": 0.1479, "lr": 1.6964374759590096e-05, "grad_norm": 7.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 2997, "epoch": 1, "loss": 0.0372, "lr": 1.6962224884070985e-05, "grad_norm": 4.0625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 2998, "epoch": 1, "loss": 0.1606, "lr": 1.6960074383864923e-05, "grad_norm": 3.53125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 2999, "epoch": 1, "loss": 0.0021, "lr": 1.6957923259164867e-05, "grad_norm": 5.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3000, "epoch": 1, "loss": 0.0004, "lr": 1.6955771510163826e-05, "grad_norm": 4.40625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3001, "epoch": 1, "loss": 0.0072, "lr": 1.695361913705487e-05, "grad_norm": 1.359375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3002, "epoch": 1, "loss": 0.1912, "lr": 1.695146614003112e-05, "grad_norm": 5.75, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 3003, "epoch": 1, "loss": 0.0094, "lr": 1.694931251928575e-05, "grad_norm": 2.953125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3004, "epoch": 1, "loss": 0.009, "lr": 1.6947158275011995e-05, "grad_norm": 4.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3005, "epoch": 1, "loss": 0.003, "lr": 1.694500340740315e-05, "grad_norm": 1.7265625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 3006, "epoch": 1, "loss": 0.0029, "lr": 1.6942847916652555e-05, "grad_norm": 5.0625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3007, "epoch": 1, "loss": 0.0008, "lr": 1.6940691802953614e-05, "grad_norm": 3.328125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3008, "epoch": 1, "loss": 0.0216, "lr": 1.6938535066499784e-05, "grad_norm": 2.890625, "samples_per_sec": 8.71, "gpu_mem_gb": 6.83} {"step": 3009, "epoch": 1, "loss": 0.0013, "lr": 1.6936377707484584e-05, "grad_norm": 2.046875, "samples_per_sec": 11.75, "gpu_mem_gb": 6.82} {"step": 3010, "epoch": 1, "loss": 0.1763, "lr": 1.6934219726101576e-05, "grad_norm": 5.5625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3011, "epoch": 1, "loss": 0.0015, "lr": 1.6932061122544387e-05, "grad_norm": 1.6171875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 3012, "epoch": 1, "loss": 0.2259, "lr": 1.69299018970067e-05, "grad_norm": 5.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 3013, "epoch": 1, "loss": 0.0003, "lr": 1.6927742049682252e-05, "grad_norm": 2.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3014, "epoch": 1, "loss": 0.0006, "lr": 1.6925581580764838e-05, "grad_norm": 1.7421875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3015, "epoch": 1, "loss": 0.0152, "lr": 1.6923420490448298e-05, "grad_norm": 4.8125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3016, "epoch": 1, "loss": 0.0098, "lr": 1.6921258778926547e-05, "grad_norm": 7.1875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3017, "epoch": 1, "loss": 0.1672, "lr": 1.6919096446393534e-05, "grad_norm": 10.875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3018, "epoch": 1, "loss": 0.1501, "lr": 1.6916933493043287e-05, "grad_norm": 5.625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3019, "epoch": 1, "loss": 0.0654, "lr": 1.6914769919069865e-05, "grad_norm": 4.0625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3020, "epoch": 1, "loss": 0.0004, "lr": 1.6912605724667405e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3021, "epoch": 1, "loss": 0.057, "lr": 1.691044091003008e-05, "grad_norm": 2.65625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3022, "epoch": 1, "loss": 0.189, "lr": 1.690827547535214e-05, "grad_norm": 9.5625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.84} {"step": 3023, "epoch": 1, "loss": 0.0285, "lr": 1.690610942082787e-05, "grad_norm": 4.6875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3024, "epoch": 1, "loss": 0.0452, "lr": 1.6903942746651626e-05, "grad_norm": 5.78125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3025, "epoch": 1, "loss": 0.1743, "lr": 1.6901775453017807e-05, "grad_norm": 4.53125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 3026, "epoch": 1, "loss": 0.0438, "lr": 1.6899607540120878e-05, "grad_norm": 4.3125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 3027, "epoch": 1, "loss": 0.0056, "lr": 1.689743900815535e-05, "grad_norm": 3.03125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3028, "epoch": 1, "loss": 0.4373, "lr": 1.6895269857315804e-05, "grad_norm": 5.78125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.84} {"step": 3029, "epoch": 1, "loss": 0.0411, "lr": 1.689310008779686e-05, "grad_norm": 3.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3030, "epoch": 1, "loss": 0.0495, "lr": 1.689092969979321e-05, "grad_norm": 4.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3031, "epoch": 1, "loss": 0.1881, "lr": 1.688875869349958e-05, "grad_norm": 8.5, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3032, "epoch": 1, "loss": 0.0682, "lr": 1.688658706911077e-05, "grad_norm": 3.4375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 3033, "epoch": 1, "loss": 0.0002, "lr": 1.6884414826821632e-05, "grad_norm": 0.87109375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3034, "epoch": 1, "loss": 0.0094, "lr": 1.688224196682707e-05, "grad_norm": 4.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3035, "epoch": 1, "loss": 0.2, "lr": 1.688006848932204e-05, "grad_norm": 3.53125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 3036, "epoch": 1, "loss": 0.0002, "lr": 1.687789439450156e-05, "grad_norm": 6.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3037, "epoch": 1, "loss": 0.0653, "lr": 1.6875719682560698e-05, "grad_norm": 5.09375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 3038, "epoch": 1, "loss": 0.2541, "lr": 1.687354435369459e-05, "grad_norm": 7.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 3039, "epoch": 1, "loss": 0.0799, "lr": 1.687136840809841e-05, "grad_norm": 2.4375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3040, "epoch": 1, "loss": 0.1637, "lr": 1.6869191845967393e-05, "grad_norm": 4.8125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 3041, "epoch": 1, "loss": 0.1463, "lr": 1.6867014667496838e-05, "grad_norm": 7.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3042, "epoch": 1, "loss": 0.0472, "lr": 1.6864836872882086e-05, "grad_norm": 4.71875, "samples_per_sec": 10.18, "gpu_mem_gb": 6.83} {"step": 3043, "epoch": 1, "loss": 0.0068, "lr": 1.6862658462318547e-05, "grad_norm": 3.84375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3044, "epoch": 1, "loss": 0.001, "lr": 1.6860479436001676e-05, "grad_norm": 3.46875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3045, "epoch": 1, "loss": 0.066, "lr": 1.685829979412699e-05, "grad_norm": 3.578125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3046, "epoch": 1, "loss": 0.0137, "lr": 1.685611953689005e-05, "grad_norm": 1.3984375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 3047, "epoch": 1, "loss": 0.0007, "lr": 1.685393866448649e-05, "grad_norm": 2.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3048, "epoch": 1, "loss": 0.1289, "lr": 1.685175717711198e-05, "grad_norm": 3.9375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 3049, "epoch": 1, "loss": 0.1165, "lr": 1.6849575074962257e-05, "grad_norm": 5.125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.84} {"step": 3050, "epoch": 1, "loss": 0.052, "lr": 1.6847392358233112e-05, "grad_norm": 3.21875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3051, "epoch": 1, "loss": 0.0003, "lr": 1.684520902712039e-05, "grad_norm": 5.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3052, "epoch": 1, "loss": 0.2236, "lr": 1.684302508181999e-05, "grad_norm": 3.8125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3053, "epoch": 1, "loss": 0.0054, "lr": 1.684084052252787e-05, "grad_norm": 1.4921875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3054, "epoch": 1, "loss": 0.0013, "lr": 1.6838655349440033e-05, "grad_norm": 4.53125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 3055, "epoch": 1, "loss": 0.0024, "lr": 1.6836469562752548e-05, "grad_norm": 2.21875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.82} {"step": 3056, "epoch": 1, "loss": 0.0358, "lr": 1.6834283162661537e-05, "grad_norm": 5.03125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3057, "epoch": 1, "loss": 0.0022, "lr": 1.6832096149363174e-05, "grad_norm": 3.703125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 3058, "epoch": 1, "loss": 0.0219, "lr": 1.6829908523053685e-05, "grad_norm": 1.84375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 3059, "epoch": 1, "loss": 0.2218, "lr": 1.6827720283929364e-05, "grad_norm": 7.34375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3060, "epoch": 1, "loss": 0.0885, "lr": 1.6825531432186545e-05, "grad_norm": 3.1875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3061, "epoch": 1, "loss": 0.0068, "lr": 1.682334196802162e-05, "grad_norm": 2.046875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3062, "epoch": 1, "loss": 0.0738, "lr": 1.6821151891631045e-05, "grad_norm": 14.9375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3063, "epoch": 1, "loss": 0.0016, "lr": 1.6818961203211325e-05, "grad_norm": 3.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3064, "epoch": 1, "loss": 0.0021, "lr": 1.6816769902959016e-05, "grad_norm": 0.875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3065, "epoch": 1, "loss": 0.004, "lr": 1.6814577991070734e-05, "grad_norm": 8.9375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3066, "epoch": 1, "loss": 0.0012, "lr": 1.681238546774315e-05, "grad_norm": 3.765625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3067, "epoch": 1, "loss": 0.0045, "lr": 1.681019233317299e-05, "grad_norm": 0.87890625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3068, "epoch": 1, "loss": 0.06, "lr": 1.680799858755703e-05, "grad_norm": 5.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3069, "epoch": 1, "loss": 0.3147, "lr": 1.6805804231092108e-05, "grad_norm": 5.78125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 3070, "epoch": 1, "loss": 0.1255, "lr": 1.6803609263975112e-05, "grad_norm": 4.40625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3071, "epoch": 1, "loss": 0.1402, "lr": 1.680141368640298e-05, "grad_norm": 3.5625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3072, "epoch": 1, "loss": 0.1944, "lr": 1.679921749857272e-05, "grad_norm": 5.6875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 3073, "epoch": 1, "loss": 0.018, "lr": 1.6797020700681378e-05, "grad_norm": 5.53125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3074, "epoch": 1, "loss": 0.0073, "lr": 1.6794823292926067e-05, "grad_norm": 4.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3075, "epoch": 1, "loss": 0.0831, "lr": 1.6792625275503945e-05, "grad_norm": 5.71875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 3076, "epoch": 1, "loss": 0.12, "lr": 1.6790426648612236e-05, "grad_norm": 6.25, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3077, "epoch": 1, "loss": 0.1463, "lr": 1.6788227412448203e-05, "grad_norm": 5.0, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3078, "epoch": 1, "loss": 0.0046, "lr": 1.678602756720918e-05, "grad_norm": 8.1875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3079, "epoch": 1, "loss": 0.0032, "lr": 1.6783827113092547e-05, "grad_norm": 0.470703125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3080, "epoch": 1, "loss": 0.0318, "lr": 1.6781626050295742e-05, "grad_norm": 4.1875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3081, "epoch": 1, "loss": 0.0049, "lr": 1.6779424379016252e-05, "grad_norm": 0.4375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3082, "epoch": 1, "loss": 0.0226, "lr": 1.6777222099451625e-05, "grad_norm": 2.84375, "samples_per_sec": 10.31, "gpu_mem_gb": 6.85} {"step": 3083, "epoch": 1, "loss": 0.1424, "lr": 1.677501921179946e-05, "grad_norm": 2.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3084, "epoch": 1, "loss": 0.0002, "lr": 1.6772815716257414e-05, "grad_norm": 2.671875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3085, "epoch": 1, "loss": 0.0395, "lr": 1.6770611613023187e-05, "grad_norm": 3.671875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3086, "epoch": 1, "loss": 0.0006, "lr": 1.6768406902294552e-05, "grad_norm": 6.65625, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 3087, "epoch": 1, "loss": 0.0634, "lr": 1.676620158426933e-05, "grad_norm": 7.03125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3088, "epoch": 1, "loss": 0.0009, "lr": 1.6763995659145383e-05, "grad_norm": 5.6875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 3089, "epoch": 1, "loss": 0.0014, "lr": 1.6761789127120644e-05, "grad_norm": 2.375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 3090, "epoch": 1, "loss": 0.0612, "lr": 1.6759581988393093e-05, "grad_norm": 4.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3091, "epoch": 1, "loss": 0.0004, "lr": 1.6757374243160766e-05, "grad_norm": 0.181640625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 3092, "epoch": 1, "loss": 0.0628, "lr": 1.6755165891621754e-05, "grad_norm": 3.609375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3093, "epoch": 1, "loss": 0.0371, "lr": 1.67529569339742e-05, "grad_norm": 3.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3094, "epoch": 1, "loss": 0.0032, "lr": 1.675074737041631e-05, "grad_norm": 4.09375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.85} {"step": 3095, "epoch": 1, "loss": 0.0664, "lr": 1.6748537201146325e-05, "grad_norm": 5.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3096, "epoch": 1, "loss": 0.1137, "lr": 1.6746326426362566e-05, "grad_norm": 5.75, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3097, "epoch": 1, "loss": 0.1557, "lr": 1.6744115046263387e-05, "grad_norm": 7.46875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3098, "epoch": 1, "loss": 0.2041, "lr": 1.6741903061047204e-05, "grad_norm": 4.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3099, "epoch": 1, "loss": 0.082, "lr": 1.6739690470912495e-05, "grad_norm": 6.28125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.85} {"step": 3100, "epoch": 1, "loss": 0.0022, "lr": 1.6737477276057776e-05, "grad_norm": 0.9921875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3101, "epoch": 1, "loss": 0.0004, "lr": 1.6735263476681636e-05, "grad_norm": 4.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3102, "epoch": 1, "loss": 0.0005, "lr": 1.6733049072982698e-05, "grad_norm": 5.03125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 3103, "epoch": 1, "loss": 0.0082, "lr": 1.673083406515966e-05, "grad_norm": 2.71875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 3104, "epoch": 1, "loss": 0.0614, "lr": 1.672861845341126e-05, "grad_norm": 7.15625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3105, "epoch": 1, "loss": 0.1687, "lr": 1.6726402237936287e-05, "grad_norm": 8.125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 3106, "epoch": 1, "loss": 0.0016, "lr": 1.6724185418933602e-05, "grad_norm": 1.1796875, "samples_per_sec": 10.43, "gpu_mem_gb": 6.82} {"step": 3107, "epoch": 1, "loss": 0.0002, "lr": 1.6721967996602106e-05, "grad_norm": 5.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3108, "epoch": 1, "loss": 0.1141, "lr": 1.6719749971140756e-05, "grad_norm": 2.78125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3109, "epoch": 1, "loss": 0.1862, "lr": 1.6717531342748564e-05, "grad_norm": 4.84375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.84} {"step": 3110, "epoch": 1, "loss": 0.2929, "lr": 1.67153121116246e-05, "grad_norm": 9.25, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3111, "epoch": 1, "loss": 0.072, "lr": 1.6713092277967986e-05, "grad_norm": 2.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3112, "epoch": 1, "loss": 0.1858, "lr": 1.6710871841977892e-05, "grad_norm": 5.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 3113, "epoch": 1, "loss": 0.0003, "lr": 1.670865080385355e-05, "grad_norm": 8.9375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3114, "epoch": 1, "loss": 0.0764, "lr": 1.670642916379424e-05, "grad_norm": 2.921875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3115, "epoch": 1, "loss": 0.0006, "lr": 1.6704206921999305e-05, "grad_norm": 7.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3116, "epoch": 1, "loss": 0.0382, "lr": 1.670198407866813e-05, "grad_norm": 4.125, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 3117, "epoch": 1, "loss": 0.0069, "lr": 1.6699760634000166e-05, "grad_norm": 4.84375, "samples_per_sec": 11.05, "gpu_mem_gb": 6.83} {"step": 3118, "epoch": 1, "loss": 0.0893, "lr": 1.6697536588194907e-05, "grad_norm": 3.859375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3119, "epoch": 1, "loss": 0.0738, "lr": 1.669531194145191e-05, "grad_norm": 5.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3120, "epoch": 1, "loss": 0.0868, "lr": 1.6693086693970776e-05, "grad_norm": 8.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.84} {"step": 3121, "epoch": 1, "loss": 0.0002, "lr": 1.6690860845951168e-05, "grad_norm": 6.28125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3122, "epoch": 1, "loss": 0.0036, "lr": 1.6688634397592808e-05, "grad_norm": 3.015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3123, "epoch": 1, "loss": 0.0012, "lr": 1.6686407349095453e-05, "grad_norm": 5.71875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3124, "epoch": 1, "loss": 0.0285, "lr": 1.6684179700658933e-05, "grad_norm": 8.625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3125, "epoch": 1, "loss": 0.1238, "lr": 1.6681951452483123e-05, "grad_norm": 4.71875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3126, "epoch": 1, "loss": 0.0269, "lr": 1.667972260476795e-05, "grad_norm": 2.8125, "samples_per_sec": 10.38, "gpu_mem_gb": 6.82} {"step": 3127, "epoch": 1, "loss": 0.0014, "lr": 1.66774931577134e-05, "grad_norm": 3.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3128, "epoch": 1, "loss": 0.0526, "lr": 1.6675263111519512e-05, "grad_norm": 4.1875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3129, "epoch": 1, "loss": 0.0014, "lr": 1.6673032466386375e-05, "grad_norm": 3.984375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3130, "epoch": 1, "loss": 0.0006, "lr": 1.6670801222514135e-05, "grad_norm": 4.53125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3131, "epoch": 1, "loss": 0.0087, "lr": 1.666856938010299e-05, "grad_norm": 5.0, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 3132, "epoch": 1, "loss": 0.0029, "lr": 1.666633693935319e-05, "grad_norm": 3.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3133, "epoch": 1, "loss": 0.02, "lr": 1.6664103900465046e-05, "grad_norm": 4.8125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3134, "epoch": 1, "loss": 0.3111, "lr": 1.666187026363892e-05, "grad_norm": 8.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3135, "epoch": 1, "loss": 0.0023, "lr": 1.6659636029075213e-05, "grad_norm": 7.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 3136, "epoch": 1, "loss": 0.0005, "lr": 1.6657401196974405e-05, "grad_norm": 2.21875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3137, "epoch": 1, "loss": 0.0012, "lr": 1.6655165767537015e-05, "grad_norm": 3.140625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3138, "epoch": 1, "loss": 0.1388, "lr": 1.6652929740963616e-05, "grad_norm": 6.1875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3139, "epoch": 1, "loss": 0.1387, "lr": 1.6650693117454827e-05, "grad_norm": 5.09375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3140, "epoch": 1, "loss": 0.0499, "lr": 1.664845589721134e-05, "grad_norm": 6.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3141, "epoch": 1, "loss": 0.023, "lr": 1.664621808043389e-05, "grad_norm": 3.9375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3142, "epoch": 1, "loss": 0.0082, "lr": 1.664397966732326e-05, "grad_norm": 2.78125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 3143, "epoch": 1, "loss": 0.0045, "lr": 1.6641740658080293e-05, "grad_norm": 1.203125, "samples_per_sec": 10.29, "gpu_mem_gb": 6.82} {"step": 3144, "epoch": 1, "loss": 0.001, "lr": 1.663950105290589e-05, "grad_norm": 0.298828125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 3145, "epoch": 1, "loss": 0.0011, "lr": 1.6637260852000997e-05, "grad_norm": 5.34375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3146, "epoch": 1, "loss": 0.0002, "lr": 1.6635020055566612e-05, "grad_norm": 5.375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 3147, "epoch": 1, "loss": 0.0082, "lr": 1.6632778663803794e-05, "grad_norm": 0.40234375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3148, "epoch": 1, "loss": 0.0043, "lr": 1.6630536676913652e-05, "grad_norm": 0.7265625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3149, "epoch": 1, "loss": 0.0497, "lr": 1.6628294095097353e-05, "grad_norm": 2.90625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3150, "epoch": 1, "loss": 0.0106, "lr": 1.6626050918556108e-05, "grad_norm": 4.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3151, "epoch": 1, "loss": 0.0034, "lr": 1.6623807147491188e-05, "grad_norm": 1.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3152, "epoch": 1, "loss": 0.0066, "lr": 1.6621562782103914e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3153, "epoch": 1, "loss": 0.1165, "lr": 1.6619317822595666e-05, "grad_norm": 6.28125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 3154, "epoch": 1, "loss": 0.0555, "lr": 1.661707226916787e-05, "grad_norm": 2.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3155, "epoch": 1, "loss": 0.001, "lr": 1.6614826122022015e-05, "grad_norm": 2.40625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3156, "epoch": 1, "loss": 0.0003, "lr": 1.6612579381359624e-05, "grad_norm": 4.5, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 3157, "epoch": 1, "loss": 0.0495, "lr": 1.6610332047382298e-05, "grad_norm": 6.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3158, "epoch": 1, "loss": 0.0582, "lr": 1.6608084120291676e-05, "grad_norm": 3.515625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3159, "epoch": 1, "loss": 0.021, "lr": 1.6605835600289456e-05, "grad_norm": 1.828125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 3160, "epoch": 1, "loss": 0.0734, "lr": 1.660358648757738e-05, "grad_norm": 5.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3161, "epoch": 1, "loss": 0.0231, "lr": 1.6601336782357254e-05, "grad_norm": 3.484375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 3162, "epoch": 1, "loss": 0.0023, "lr": 1.6599086484830936e-05, "grad_norm": 3.546875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3163, "epoch": 1, "loss": 0.004, "lr": 1.659683559520033e-05, "grad_norm": 2.875, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 3164, "epoch": 1, "loss": 0.019, "lr": 1.6594584113667398e-05, "grad_norm": 3.09375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3165, "epoch": 1, "loss": 0.0024, "lr": 1.659233204043416e-05, "grad_norm": 0.62890625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3166, "epoch": 1, "loss": 0.0019, "lr": 1.6590079375702678e-05, "grad_norm": 8.8125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3167, "epoch": 1, "loss": 0.0045, "lr": 1.6587826119675074e-05, "grad_norm": 3.78125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3168, "epoch": 1, "loss": 0.0014, "lr": 1.658557227255352e-05, "grad_norm": 3.734375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3169, "epoch": 1, "loss": 0.0112, "lr": 1.658331783454025e-05, "grad_norm": 3.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3170, "epoch": 1, "loss": 0.0148, "lr": 1.658106280583754e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 3171, "epoch": 1, "loss": 0.0004, "lr": 1.6578807186647715e-05, "grad_norm": 1.6171875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 3172, "epoch": 1, "loss": 0.0004, "lr": 1.6576550977173173e-05, "grad_norm": 5.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3173, "epoch": 1, "loss": 0.0016, "lr": 1.6574294177616343e-05, "grad_norm": 5.9375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 3174, "epoch": 1, "loss": 0.0965, "lr": 1.6572036788179728e-05, "grad_norm": 6.4375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3175, "epoch": 1, "loss": 0.1589, "lr": 1.656977880906586e-05, "grad_norm": 3.46875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3176, "epoch": 1, "loss": 0.154, "lr": 1.6567520240477344e-05, "grad_norm": 13.0625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 3177, "epoch": 1, "loss": 0.0946, "lr": 1.656526108261683e-05, "grad_norm": 4.75, "samples_per_sec": 10.75, "gpu_mem_gb": 6.84} {"step": 3178, "epoch": 1, "loss": 0.0012, "lr": 1.6563001335687024e-05, "grad_norm": 4.75, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3179, "epoch": 1, "loss": 0.0025, "lr": 1.6560740999890673e-05, "grad_norm": 2.734375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3180, "epoch": 1, "loss": 0.0006, "lr": 1.6558480075430594e-05, "grad_norm": 5.90625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 3181, "epoch": 1, "loss": 0.1089, "lr": 1.655621856250965e-05, "grad_norm": 4.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 3182, "epoch": 1, "loss": 0.084, "lr": 1.6553956461330746e-05, "grad_norm": 4.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3183, "epoch": 1, "loss": 0.0143, "lr": 1.655169377209686e-05, "grad_norm": 1.4609375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 3184, "epoch": 1, "loss": 0.0035, "lr": 1.6549430495011008e-05, "grad_norm": 1.0234375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3185, "epoch": 1, "loss": 0.0055, "lr": 1.654716663027626e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3186, "epoch": 1, "loss": 0.1731, "lr": 1.654490217809575e-05, "grad_norm": 4.59375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 3187, "epoch": 1, "loss": 0.0024, "lr": 1.6542637138672647e-05, "grad_norm": 2.5625, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 3188, "epoch": 1, "loss": 0.0374, "lr": 1.6540371512210187e-05, "grad_norm": 5.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3189, "epoch": 1, "loss": 0.022, "lr": 1.6538105298911653e-05, "grad_norm": 3.984375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3190, "epoch": 1, "loss": 0.022, "lr": 1.653583849898038e-05, "grad_norm": 4.4375, "samples_per_sec": 10.39, "gpu_mem_gb": 6.82} {"step": 3191, "epoch": 1, "loss": 0.0193, "lr": 1.653357111261976e-05, "grad_norm": 1.40625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3192, "epoch": 1, "loss": 0.0031, "lr": 1.6531303140033235e-05, "grad_norm": 0.30078125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3193, "epoch": 1, "loss": 0.0007, "lr": 1.6529034581424293e-05, "grad_norm": 4.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3194, "epoch": 1, "loss": 0.0022, "lr": 1.6526765436996488e-05, "grad_norm": 0.2421875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3195, "epoch": 1, "loss": 0.0891, "lr": 1.6524495706953417e-05, "grad_norm": 6.75, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3196, "epoch": 1, "loss": 0.0044, "lr": 1.652222539149873e-05, "grad_norm": 2.046875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 3197, "epoch": 1, "loss": 0.0094, "lr": 1.6519954490836133e-05, "grad_norm": 4.3125, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 3198, "epoch": 1, "loss": 0.0756, "lr": 1.6517683005169384e-05, "grad_norm": 4.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3199, "epoch": 1, "loss": 0.1737, "lr": 1.651541093470229e-05, "grad_norm": 4.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3200, "epoch": 1, "loss": 0.0088, "lr": 1.6513138279638715e-05, "grad_norm": 0.54296875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 3201, "epoch": 1, "loss": 0.2771, "lr": 1.6510865040182575e-05, "grad_norm": 8.1875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3202, "epoch": 1, "loss": 0.0754, "lr": 1.6508591216537832e-05, "grad_norm": 4.625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3203, "epoch": 1, "loss": 0.0002, "lr": 1.650631680890851e-05, "grad_norm": 4.78125, "samples_per_sec": 10.45, "gpu_mem_gb": 6.82} {"step": 3204, "epoch": 1, "loss": 0.0537, "lr": 1.6504041817498676e-05, "grad_norm": 3.953125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 3205, "epoch": 1, "loss": 0.0555, "lr": 1.650176624251246e-05, "grad_norm": 10.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3206, "epoch": 1, "loss": 0.0488, "lr": 1.6499490084154034e-05, "grad_norm": 5.75, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3207, "epoch": 1, "loss": 0.0297, "lr": 1.649721334262763e-05, "grad_norm": 5.8125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3208, "epoch": 1, "loss": 0.08, "lr": 1.6494936018137525e-05, "grad_norm": 4.71875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3209, "epoch": 1, "loss": 0.1021, "lr": 1.6492658110888055e-05, "grad_norm": 2.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 3210, "epoch": 1, "loss": 0.0387, "lr": 1.64903796210836e-05, "grad_norm": 1.1953125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 3211, "epoch": 1, "loss": 0.0092, "lr": 1.6488100548928608e-05, "grad_norm": 7.8125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3212, "epoch": 1, "loss": 0.0056, "lr": 1.648582089462756e-05, "grad_norm": 4.46875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3213, "epoch": 1, "loss": 0.0005, "lr": 1.6483540658385003e-05, "grad_norm": 0.1640625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3214, "epoch": 1, "loss": 0.0547, "lr": 1.6481259840405533e-05, "grad_norm": 4.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3215, "epoch": 1, "loss": 0.0134, "lr": 1.6478978440893797e-05, "grad_norm": 2.859375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3216, "epoch": 1, "loss": 0.0014, "lr": 1.647669646005449e-05, "grad_norm": 3.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3217, "epoch": 1, "loss": 0.0022, "lr": 1.6474413898092366e-05, "grad_norm": 1.1640625, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 3218, "epoch": 1, "loss": 0.0005, "lr": 1.6472130755212224e-05, "grad_norm": 4.90625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 3219, "epoch": 1, "loss": 0.0003, "lr": 1.6469847031618925e-05, "grad_norm": 4.0625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3220, "epoch": 1, "loss": 0.0001, "lr": 1.6467562727517375e-05, "grad_norm": 3.484375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 3221, "epoch": 1, "loss": 0.0009, "lr": 1.6465277843112533e-05, "grad_norm": 3.375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3222, "epoch": 1, "loss": 0.0738, "lr": 1.646299237860941e-05, "grad_norm": 5.4375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3223, "epoch": 1, "loss": 0.0075, "lr": 1.646070633421307e-05, "grad_norm": 5.21875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3224, "epoch": 1, "loss": 0.2445, "lr": 1.645841971012863e-05, "grad_norm": 7.0, "samples_per_sec": 10.3, "gpu_mem_gb": 6.83} {"step": 3225, "epoch": 1, "loss": 0.0291, "lr": 1.6456132506561258e-05, "grad_norm": 2.0, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3226, "epoch": 1, "loss": 0.1439, "lr": 1.645384472371617e-05, "grad_norm": 4.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 3227, "epoch": 1, "loss": 0.0023, "lr": 1.645155636179864e-05, "grad_norm": 3.6875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 3228, "epoch": 1, "loss": 0.0015, "lr": 1.6449267421013994e-05, "grad_norm": 0.69140625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3229, "epoch": 1, "loss": 0.1211, "lr": 1.6446977901567606e-05, "grad_norm": 2.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3230, "epoch": 1, "loss": 0.0765, "lr": 1.6444687803664904e-05, "grad_norm": 7.96875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3231, "epoch": 1, "loss": 0.1564, "lr": 1.6442397127511366e-05, "grad_norm": 6.15625, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 3232, "epoch": 1, "loss": 0.1336, "lr": 1.6440105873312523e-05, "grad_norm": 4.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 3233, "epoch": 1, "loss": 0.0219, "lr": 1.643781404127396e-05, "grad_norm": 6.3125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 3234, "epoch": 1, "loss": 0.0544, "lr": 1.643552163160131e-05, "grad_norm": 5.6875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 3235, "epoch": 1, "loss": 0.0961, "lr": 1.6433228644500265e-05, "grad_norm": 5.34375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3236, "epoch": 1, "loss": 0.0002, "lr": 1.643093508017656e-05, "grad_norm": 0.37890625, "samples_per_sec": 8.88, "gpu_mem_gb": 6.83} {"step": 3237, "epoch": 1, "loss": 0.0003, "lr": 1.6428640938835984e-05, "grad_norm": 2.875, "samples_per_sec": 11.24, "gpu_mem_gb": 6.83} {"step": 3238, "epoch": 1, "loss": 0.0185, "lr": 1.6426346220684376e-05, "grad_norm": 1.65625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3239, "epoch": 1, "loss": 0.0134, "lr": 1.6424050925927643e-05, "grad_norm": 3.046875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3240, "epoch": 1, "loss": 0.008, "lr": 1.6421755054771718e-05, "grad_norm": 3.375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3241, "epoch": 1, "loss": 0.3049, "lr": 1.6419458607422606e-05, "grad_norm": 8.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3242, "epoch": 1, "loss": 0.0011, "lr": 1.641716158408635e-05, "grad_norm": 3.203125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 3243, "epoch": 1, "loss": 0.0081, "lr": 1.6414863984969054e-05, "grad_norm": 1.8515625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 3244, "epoch": 1, "loss": 0.0046, "lr": 1.641256581027687e-05, "grad_norm": 4.78125, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 3245, "epoch": 1, "loss": 0.1431, "lr": 1.6410267060216007e-05, "grad_norm": 4.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3246, "epoch": 1, "loss": 0.1465, "lr": 1.6407967734992714e-05, "grad_norm": 4.625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3247, "epoch": 1, "loss": 0.2614, "lr": 1.6405667834813302e-05, "grad_norm": 4.78125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3248, "epoch": 1, "loss": 0.0065, "lr": 1.640336735988413e-05, "grad_norm": 5.21875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3249, "epoch": 1, "loss": 0.0203, "lr": 1.6401066310411603e-05, "grad_norm": 2.09375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3250, "epoch": 1, "loss": 0.0027, "lr": 1.6398764686602188e-05, "grad_norm": 0.2255859375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 3251, "epoch": 1, "loss": 0.2123, "lr": 1.6396462488662405e-05, "grad_norm": 6.40625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 3252, "epoch": 1, "loss": 0.0319, "lr": 1.6394159716798807e-05, "grad_norm": 5.09375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3253, "epoch": 1, "loss": 0.3607, "lr": 1.6391856371218014e-05, "grad_norm": 8.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3254, "epoch": 1, "loss": 0.0001, "lr": 1.63895524521267e-05, "grad_norm": 4.28125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3255, "epoch": 1, "loss": 0.0114, "lr": 1.6387247959731582e-05, "grad_norm": 2.546875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3256, "epoch": 1, "loss": 0.0011, "lr": 1.6384942894239426e-05, "grad_norm": 1.8828125, "samples_per_sec": 11.04, "gpu_mem_gb": 6.83} {"step": 3257, "epoch": 1, "loss": 0.065, "lr": 1.638263725585706e-05, "grad_norm": 3.65625, "samples_per_sec": 10.53, "gpu_mem_gb": 6.82} {"step": 3258, "epoch": 1, "loss": 0.0427, "lr": 1.6380331044791354e-05, "grad_norm": 5.65625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 3259, "epoch": 1, "loss": 0.0462, "lr": 1.6378024261249237e-05, "grad_norm": 6.375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3260, "epoch": 1, "loss": 0.0557, "lr": 1.6375716905437685e-05, "grad_norm": 3.046875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3261, "epoch": 1, "loss": 0.0392, "lr": 1.6373408977563726e-05, "grad_norm": 3.5, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3262, "epoch": 1, "loss": 0.0371, "lr": 1.6371100477834433e-05, "grad_norm": 4.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3263, "epoch": 1, "loss": 0.106, "lr": 1.6368791406456942e-05, "grad_norm": 3.578125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.85} {"step": 3264, "epoch": 1, "loss": 0.0005, "lr": 1.636648176363844e-05, "grad_norm": 2.34375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3265, "epoch": 1, "loss": 0.0877, "lr": 1.6364171549586146e-05, "grad_norm": 4.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3266, "epoch": 1, "loss": 0.0991, "lr": 1.636186076450736e-05, "grad_norm": 5.1875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.84} {"step": 3267, "epoch": 1, "loss": 0.1226, "lr": 1.6359549408609407e-05, "grad_norm": 4.4375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3268, "epoch": 1, "loss": 0.0931, "lr": 1.6357237482099682e-05, "grad_norm": 3.40625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3269, "epoch": 1, "loss": 0.0002, "lr": 1.6354924985185614e-05, "grad_norm": 6.84375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3270, "epoch": 1, "loss": 0.008, "lr": 1.63526119180747e-05, "grad_norm": 4.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3271, "epoch": 1, "loss": 0.0683, "lr": 1.6350298280974475e-05, "grad_norm": 6.34375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3272, "epoch": 1, "loss": 0.0024, "lr": 1.6347984074092533e-05, "grad_norm": 3.015625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3273, "epoch": 1, "loss": 0.0951, "lr": 1.6345669297636516e-05, "grad_norm": 4.9375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3274, "epoch": 1, "loss": 0.0125, "lr": 1.634335395181412e-05, "grad_norm": 2.15625, "samples_per_sec": 10.54, "gpu_mem_gb": 6.85} {"step": 3275, "epoch": 1, "loss": 0.0019, "lr": 1.6341038036833084e-05, "grad_norm": 5.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3276, "epoch": 1, "loss": 0.0003, "lr": 1.633872155290121e-05, "grad_norm": 3.109375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3277, "epoch": 1, "loss": 0.0038, "lr": 1.6336404500226345e-05, "grad_norm": 4.78125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 3278, "epoch": 1, "loss": 0.032, "lr": 1.633408687901638e-05, "grad_norm": 5.46875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3279, "epoch": 1, "loss": 0.1377, "lr": 1.6331768689479274e-05, "grad_norm": 10.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3280, "epoch": 1, "loss": 0.1909, "lr": 1.632944993182302e-05, "grad_norm": 5.28125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.84} {"step": 3281, "epoch": 1, "loss": 0.129, "lr": 1.6327130606255667e-05, "grad_norm": 3.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 3282, "epoch": 1, "loss": 0.0027, "lr": 1.6324810712985325e-05, "grad_norm": 2.40625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3283, "epoch": 1, "loss": 0.1944, "lr": 1.632249025222014e-05, "grad_norm": 6.125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3284, "epoch": 1, "loss": 0.0054, "lr": 1.6320169224168322e-05, "grad_norm": 4.28125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3285, "epoch": 1, "loss": 0.0997, "lr": 1.631784762903812e-05, "grad_norm": 5.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3286, "epoch": 1, "loss": 0.126, "lr": 1.631552546703784e-05, "grad_norm": 5.5, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 3287, "epoch": 1, "loss": 0.0166, "lr": 1.6313202738375842e-05, "grad_norm": 2.03125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3288, "epoch": 1, "loss": 0.002, "lr": 1.631087944326053e-05, "grad_norm": 5.28125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3289, "epoch": 1, "loss": 0.0037, "lr": 1.630855558190036e-05, "grad_norm": 1.2578125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3290, "epoch": 1, "loss": 0.0579, "lr": 1.6306231154503852e-05, "grad_norm": 3.84375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3291, "epoch": 1, "loss": 0.0061, "lr": 1.6303906161279554e-05, "grad_norm": 3.0625, "samples_per_sec": 10.21, "gpu_mem_gb": 6.83} {"step": 3292, "epoch": 1, "loss": 0.2728, "lr": 1.630158060243608e-05, "grad_norm": 6.0625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 3293, "epoch": 1, "loss": 0.1559, "lr": 1.629925447818209e-05, "grad_norm": 7.34375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3294, "epoch": 1, "loss": 0.054, "lr": 1.62969277887263e-05, "grad_norm": 3.296875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 3295, "epoch": 1, "loss": 0.0349, "lr": 1.6294600534277468e-05, "grad_norm": 7.5, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3296, "epoch": 1, "loss": 0.0193, "lr": 1.6292272715044414e-05, "grad_norm": 1.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 3297, "epoch": 1, "loss": 0.339, "lr": 1.6289944331235993e-05, "grad_norm": 6.78125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 3298, "epoch": 1, "loss": 0.0509, "lr": 1.6287615383061127e-05, "grad_norm": 2.140625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3299, "epoch": 1, "loss": 0.1169, "lr": 1.628528587072878e-05, "grad_norm": 8.3125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 3300, "epoch": 1, "loss": 0.174, "lr": 1.628295579444796e-05, "grad_norm": 4.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3301, "epoch": 1, "loss": 0.0778, "lr": 1.6280625154427743e-05, "grad_norm": 3.484375, "samples_per_sec": 10.34, "gpu_mem_gb": 6.83} {"step": 3302, "epoch": 1, "loss": 0.2334, "lr": 1.6278293950877246e-05, "grad_norm": 10.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3303, "epoch": 1, "loss": 0.0273, "lr": 1.627596218400563e-05, "grad_norm": 7.0, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3304, "epoch": 1, "loss": 0.0002, "lr": 1.627362985402212e-05, "grad_norm": 2.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3305, "epoch": 1, "loss": 0.0041, "lr": 1.627129696113598e-05, "grad_norm": 1.0625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3306, "epoch": 1, "loss": 0.0031, "lr": 1.6268963505556532e-05, "grad_norm": 2.359375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3307, "epoch": 1, "loss": 0.003, "lr": 1.6266629487493144e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 3308, "epoch": 1, "loss": 0.0047, "lr": 1.6264294907155237e-05, "grad_norm": 3.640625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 3309, "epoch": 1, "loss": 0.1381, "lr": 1.626195976475228e-05, "grad_norm": 4.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3310, "epoch": 1, "loss": 0.0002, "lr": 1.62596240604938e-05, "grad_norm": 5.90625, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 3311, "epoch": 1, "loss": 0.0005, "lr": 1.6257287794589362e-05, "grad_norm": 1.8828125, "samples_per_sec": 10.56, "gpu_mem_gb": 6.82} {"step": 3312, "epoch": 1, "loss": 0.0668, "lr": 1.6254950967248592e-05, "grad_norm": 6.65625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3313, "epoch": 1, "loss": 0.0107, "lr": 1.625261357868116e-05, "grad_norm": 1.5390625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3314, "epoch": 1, "loss": 0.0115, "lr": 1.625027562909679e-05, "grad_norm": 5.3125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3315, "epoch": 1, "loss": 0.168, "lr": 1.624793711870525e-05, "grad_norm": 5.40625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3316, "epoch": 1, "loss": 0.0052, "lr": 1.6245598047716368e-05, "grad_norm": 3.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3317, "epoch": 1, "loss": 0.0032, "lr": 1.6243258416340017e-05, "grad_norm": 3.390625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 3318, "epoch": 1, "loss": 0.0739, "lr": 1.6240918224786123e-05, "grad_norm": 7.375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3319, "epoch": 1, "loss": 0.1244, "lr": 1.6238577473264658e-05, "grad_norm": 4.375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3320, "epoch": 1, "loss": 0.0002, "lr": 1.6236236161985646e-05, "grad_norm": 5.625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3321, "epoch": 1, "loss": 0.1187, "lr": 1.623389429115916e-05, "grad_norm": 3.515625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.84} {"step": 3322, "epoch": 1, "loss": 0.0604, "lr": 1.6231551860995328e-05, "grad_norm": 4.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3323, "epoch": 1, "loss": 0.1352, "lr": 1.6229208871704326e-05, "grad_norm": 8.1875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3324, "epoch": 1, "loss": 0.0096, "lr": 1.6226865323496373e-05, "grad_norm": 4.40625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 3325, "epoch": 1, "loss": 0.0008, "lr": 1.6224521216581748e-05, "grad_norm": 3.25, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 3326, "epoch": 1, "loss": 0.065, "lr": 1.622217655117078e-05, "grad_norm": 1.359375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3327, "epoch": 1, "loss": 0.1898, "lr": 1.621983132747384e-05, "grad_norm": 4.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3328, "epoch": 1, "loss": 0.0024, "lr": 1.6217485545701353e-05, "grad_norm": 13.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3329, "epoch": 1, "loss": 0.1037, "lr": 1.6215139206063797e-05, "grad_norm": 4.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3330, "epoch": 1, "loss": 0.0693, "lr": 1.62127923087717e-05, "grad_norm": 3.546875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3331, "epoch": 1, "loss": 0.0085, "lr": 1.6210444854035633e-05, "grad_norm": 4.65625, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 3332, "epoch": 1, "loss": 0.3272, "lr": 1.620809684206622e-05, "grad_norm": 6.90625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3333, "epoch": 1, "loss": 0.0007, "lr": 1.6205748273074147e-05, "grad_norm": 6.0625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3334, "epoch": 1, "loss": 0.0514, "lr": 1.620339914727013e-05, "grad_norm": 2.4375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3335, "epoch": 1, "loss": 0.0686, "lr": 1.620104946486495e-05, "grad_norm": 5.90625, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 3336, "epoch": 1, "loss": 0.0013, "lr": 1.619869922606943e-05, "grad_norm": 5.0625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3337, "epoch": 1, "loss": 0.0218, "lr": 1.6196348431094448e-05, "grad_norm": 1.484375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 3338, "epoch": 1, "loss": 0.0263, "lr": 1.619399708015093e-05, "grad_norm": 4.125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3339, "epoch": 1, "loss": 0.0515, "lr": 1.6191645173449843e-05, "grad_norm": 1.9765625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3340, "epoch": 1, "loss": 0.0025, "lr": 1.6189292711202225e-05, "grad_norm": 0.8203125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3341, "epoch": 1, "loss": 0.1683, "lr": 1.618693969361914e-05, "grad_norm": 3.15625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3342, "epoch": 1, "loss": 0.0444, "lr": 1.618458612091172e-05, "grad_norm": 6.21875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3343, "epoch": 1, "loss": 0.0281, "lr": 1.618223199329114e-05, "grad_norm": 2.515625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3344, "epoch": 1, "loss": 0.0471, "lr": 1.617987731096862e-05, "grad_norm": 4.15625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3345, "epoch": 1, "loss": 0.0191, "lr": 1.6177522074155436e-05, "grad_norm": 5.3125, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 3346, "epoch": 1, "loss": 0.1579, "lr": 1.6175166283062915e-05, "grad_norm": 5.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3347, "epoch": 1, "loss": 0.0168, "lr": 1.6172809937902425e-05, "grad_norm": 4.125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 3348, "epoch": 1, "loss": 0.0709, "lr": 1.6170453038885394e-05, "grad_norm": 4.59375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 3349, "epoch": 1, "loss": 0.0015, "lr": 1.6168095586223295e-05, "grad_norm": 4.46875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3350, "epoch": 1, "loss": 0.0295, "lr": 1.616573758012765e-05, "grad_norm": 5.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3351, "epoch": 1, "loss": 0.0009, "lr": 1.6163379020810032e-05, "grad_norm": 5.71875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 3352, "epoch": 1, "loss": 0.1186, "lr": 1.616101990848206e-05, "grad_norm": 2.828125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3353, "epoch": 1, "loss": 0.0004, "lr": 1.615866024335541e-05, "grad_norm": 2.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 3354, "epoch": 1, "loss": 0.0248, "lr": 1.6156300025641805e-05, "grad_norm": 3.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3355, "epoch": 1, "loss": 0.0076, "lr": 1.6153939255553006e-05, "grad_norm": 3.875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3356, "epoch": 1, "loss": 0.1692, "lr": 1.6151577933300847e-05, "grad_norm": 5.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3357, "epoch": 1, "loss": 0.158, "lr": 1.614921605909719e-05, "grad_norm": 6.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3358, "epoch": 1, "loss": 0.0472, "lr": 1.6146853633153954e-05, "grad_norm": 3.734375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 3359, "epoch": 1, "loss": 0.23, "lr": 1.6144490655683112e-05, "grad_norm": 5.875, "samples_per_sec": 11.04, "gpu_mem_gb": 6.83} {"step": 3360, "epoch": 1, "loss": 0.0012, "lr": 1.6142127126896682e-05, "grad_norm": 3.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3361, "epoch": 1, "loss": 0.0005, "lr": 1.6139763047006727e-05, "grad_norm": 2.796875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3362, "epoch": 1, "loss": 0.0009, "lr": 1.613739841622537e-05, "grad_norm": 2.59375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 3363, "epoch": 1, "loss": 0.0303, "lr": 1.6135033234764777e-05, "grad_norm": 3.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3364, "epoch": 1, "loss": 0.0183, "lr": 1.6132667502837164e-05, "grad_norm": 5.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3365, "epoch": 1, "loss": 0.0115, "lr": 1.61303012206548e-05, "grad_norm": 1.453125, "samples_per_sec": 10.36, "gpu_mem_gb": 6.83} {"step": 3366, "epoch": 1, "loss": 0.0297, "lr": 1.612793438842999e-05, "grad_norm": 3.4375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3367, "epoch": 1, "loss": 0.0574, "lr": 1.6125567006375105e-05, "grad_norm": 3.328125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3368, "epoch": 1, "loss": 0.2043, "lr": 1.6123199074702565e-05, "grad_norm": 5.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3369, "epoch": 1, "loss": 0.0043, "lr": 1.612083059362482e-05, "grad_norm": 4.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3370, "epoch": 1, "loss": 0.1335, "lr": 1.6118461563354398e-05, "grad_norm": 6.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3371, "epoch": 1, "loss": 0.001, "lr": 1.6116091984103846e-05, "grad_norm": 1.0390625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 3372, "epoch": 1, "loss": 0.0292, "lr": 1.6113721856085783e-05, "grad_norm": 1.8984375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 3373, "epoch": 1, "loss": 0.0106, "lr": 1.6111351179512867e-05, "grad_norm": 1.3515625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3374, "epoch": 1, "loss": 0.0065, "lr": 1.6108979954597804e-05, "grad_norm": 5.375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 3375, "epoch": 1, "loss": 0.0468, "lr": 1.610660818155336e-05, "grad_norm": 3.140625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 3376, "epoch": 1, "loss": 0.0822, "lr": 1.6104235860592342e-05, "grad_norm": 5.28125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3377, "epoch": 1, "loss": 0.0436, "lr": 1.6101862991927598e-05, "grad_norm": 3.75, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3378, "epoch": 1, "loss": 0.015, "lr": 1.6099489575772046e-05, "grad_norm": 5.34375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 3379, "epoch": 1, "loss": 0.0011, "lr": 1.609711561233863e-05, "grad_norm": 0.9375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 3380, "epoch": 1, "loss": 0.1298, "lr": 1.609474110184036e-05, "grad_norm": 5.625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 3381, "epoch": 1, "loss": 0.0654, "lr": 1.6092366044490293e-05, "grad_norm": 3.265625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3382, "epoch": 1, "loss": 0.2963, "lr": 1.6089990440501523e-05, "grad_norm": 5.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3383, "epoch": 1, "loss": 0.2416, "lr": 1.608761429008721e-05, "grad_norm": 7.65625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3384, "epoch": 1, "loss": 0.0008, "lr": 1.6085237593460544e-05, "grad_norm": 4.40625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 3385, "epoch": 1, "loss": 0.1179, "lr": 1.6082860350834788e-05, "grad_norm": 4.84375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3386, "epoch": 1, "loss": 0.1269, "lr": 1.6080482562423233e-05, "grad_norm": 2.375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3387, "epoch": 1, "loss": 0.0947, "lr": 1.6078104228439225e-05, "grad_norm": 2.953125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3388, "epoch": 1, "loss": 0.0025, "lr": 1.607572534909616e-05, "grad_norm": 4.5625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3389, "epoch": 1, "loss": 0.0066, "lr": 1.607334592460749e-05, "grad_norm": 1.453125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 3390, "epoch": 1, "loss": 0.2055, "lr": 1.6070965955186705e-05, "grad_norm": 6.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3391, "epoch": 1, "loss": 0.0618, "lr": 1.6068585441047345e-05, "grad_norm": 5.84375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3392, "epoch": 1, "loss": 0.0328, "lr": 1.606620438240301e-05, "grad_norm": 5.53125, "samples_per_sec": 10.24, "gpu_mem_gb": 6.82} {"step": 3393, "epoch": 1, "loss": 0.0071, "lr": 1.606382277946734e-05, "grad_norm": 2.46875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3394, "epoch": 1, "loss": 0.0025, "lr": 1.6061440632454014e-05, "grad_norm": 3.578125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3395, "epoch": 1, "loss": 0.0005, "lr": 1.605905794157678e-05, "grad_norm": 5.0, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3396, "epoch": 1, "loss": 0.0004, "lr": 1.6056674707049423e-05, "grad_norm": 5.8125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3397, "epoch": 1, "loss": 0.0232, "lr": 1.605429092908578e-05, "grad_norm": 2.546875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3398, "epoch": 1, "loss": 0.0273, "lr": 1.6051906607899735e-05, "grad_norm": 1.9609375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 3399, "epoch": 1, "loss": 0.0007, "lr": 1.6049521743705224e-05, "grad_norm": 3.15625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 3400, "epoch": 1, "loss": 0.0948, "lr": 1.6047136336716226e-05, "grad_norm": 4.9375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 3401, "epoch": 1, "loss": 0.0502, "lr": 1.6044750387146772e-05, "grad_norm": 1.921875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3402, "epoch": 1, "loss": 0.2407, "lr": 1.6042363895210948e-05, "grad_norm": 4.28125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.84} {"step": 3403, "epoch": 1, "loss": 0.004, "lr": 1.6039976861122873e-05, "grad_norm": 1.4453125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 3404, "epoch": 1, "loss": 0.0459, "lr": 1.603758928509673e-05, "grad_norm": 3.171875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3405, "epoch": 1, "loss": 0.0003, "lr": 1.6035201167346746e-05, "grad_norm": 5.21875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3406, "epoch": 1, "loss": 0.0005, "lr": 1.603281250808719e-05, "grad_norm": 2.125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3407, "epoch": 1, "loss": 0.0144, "lr": 1.603042330753239e-05, "grad_norm": 5.875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3408, "epoch": 1, "loss": 0.1358, "lr": 1.6028033565896715e-05, "grad_norm": 8.9375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 3409, "epoch": 1, "loss": 0.0116, "lr": 1.6025643283394587e-05, "grad_norm": 3.328125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3410, "epoch": 1, "loss": 0.0025, "lr": 1.602325246024047e-05, "grad_norm": 2.96875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3411, "epoch": 1, "loss": 0.0623, "lr": 1.6020861096648883e-05, "grad_norm": 3.15625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3412, "epoch": 1, "loss": 0.1579, "lr": 1.6018469192834394e-05, "grad_norm": 2.765625, "samples_per_sec": 10.39, "gpu_mem_gb": 6.84} {"step": 3413, "epoch": 1, "loss": 0.0021, "lr": 1.6016076749011616e-05, "grad_norm": 3.4375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3414, "epoch": 1, "loss": 0.0082, "lr": 1.601368376539521e-05, "grad_norm": 4.53125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3415, "epoch": 1, "loss": 0.0053, "lr": 1.601129024219989e-05, "grad_norm": 2.671875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3416, "epoch": 1, "loss": 0.0001, "lr": 1.600889617964041e-05, "grad_norm": 5.40625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3417, "epoch": 1, "loss": 0.0086, "lr": 1.6006501577931582e-05, "grad_norm": 2.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3418, "epoch": 1, "loss": 0.0834, "lr": 1.600410643728826e-05, "grad_norm": 6.3125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 3419, "epoch": 1, "loss": 0.0012, "lr": 1.6001710757925356e-05, "grad_norm": 3.890625, "samples_per_sec": 10.25, "gpu_mem_gb": 6.83} {"step": 3420, "epoch": 1, "loss": 0.1394, "lr": 1.599931454005781e-05, "grad_norm": 8.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3421, "epoch": 1, "loss": 0.0216, "lr": 1.5996917783900633e-05, "grad_norm": 5.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3422, "epoch": 1, "loss": 0.1798, "lr": 1.5994520489668865e-05, "grad_norm": 6.78125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3423, "epoch": 1, "loss": 0.0731, "lr": 1.5992122657577612e-05, "grad_norm": 4.71875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3424, "epoch": 1, "loss": 0.1567, "lr": 1.598972428784202e-05, "grad_norm": 6.09375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3425, "epoch": 1, "loss": 0.122, "lr": 1.5987325380677278e-05, "grad_norm": 3.40625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3426, "epoch": 1, "loss": 0.0554, "lr": 1.598492593629863e-05, "grad_norm": 5.65625, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 3427, "epoch": 1, "loss": 0.0002, "lr": 1.598252595492137e-05, "grad_norm": 4.1875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3428, "epoch": 1, "loss": 0.2186, "lr": 1.5980125436760835e-05, "grad_norm": 6.15625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3429, "epoch": 1, "loss": 0.1883, "lr": 1.597772438203241e-05, "grad_norm": 3.078125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3430, "epoch": 1, "loss": 0.0001, "lr": 1.5975322790951532e-05, "grad_norm": 4.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3431, "epoch": 1, "loss": 0.0475, "lr": 1.597292066373368e-05, "grad_norm": 4.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3432, "epoch": 1, "loss": 0.01, "lr": 1.5970518000594392e-05, "grad_norm": 4.4375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 3433, "epoch": 1, "loss": 0.0009, "lr": 1.5968114801749247e-05, "grad_norm": 3.46875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 3434, "epoch": 1, "loss": 0.014, "lr": 1.5965711067413864e-05, "grad_norm": 1.765625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3435, "epoch": 1, "loss": 0.0623, "lr": 1.596330679780393e-05, "grad_norm": 3.15625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3436, "epoch": 1, "loss": 0.004, "lr": 1.5960901993135158e-05, "grad_norm": 4.875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3437, "epoch": 1, "loss": 0.002, "lr": 1.5958496653623327e-05, "grad_norm": 0.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3438, "epoch": 1, "loss": 0.0008, "lr": 1.5956090779484252e-05, "grad_norm": 4.5625, "samples_per_sec": 11.08, "gpu_mem_gb": 6.83} {"step": 3439, "epoch": 1, "loss": 0.0174, "lr": 1.5953684370933806e-05, "grad_norm": 4.5625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 3440, "epoch": 1, "loss": 0.004, "lr": 1.59512774281879e-05, "grad_norm": 4.53125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3441, "epoch": 1, "loss": 0.2028, "lr": 1.59488699514625e-05, "grad_norm": 5.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3442, "epoch": 1, "loss": 0.0024, "lr": 1.594646194097361e-05, "grad_norm": 5.59375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 3443, "epoch": 1, "loss": 0.0083, "lr": 1.5944053396937298e-05, "grad_norm": 3.890625, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 3444, "epoch": 1, "loss": 0.0004, "lr": 1.5941644319569665e-05, "grad_norm": 0.59375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3445, "epoch": 1, "loss": 0.0404, "lr": 1.593923470908687e-05, "grad_norm": 4.625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 3446, "epoch": 1, "loss": 0.0044, "lr": 1.5936824565705118e-05, "grad_norm": 3.546875, "samples_per_sec": 10.43, "gpu_mem_gb": 6.85} {"step": 3447, "epoch": 1, "loss": 0.0015, "lr": 1.5934413889640653e-05, "grad_norm": 1.7578125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3448, "epoch": 1, "loss": 0.0357, "lr": 1.593200268110978e-05, "grad_norm": 2.640625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 3449, "epoch": 1, "loss": 0.2036, "lr": 1.592959094032884e-05, "grad_norm": 5.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3450, "epoch": 1, "loss": 0.0162, "lr": 1.5927178667514226e-05, "grad_norm": 8.1875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3451, "epoch": 1, "loss": 0.0776, "lr": 1.5924765862882385e-05, "grad_norm": 3.484375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3452, "epoch": 1, "loss": 0.0349, "lr": 1.5922352526649803e-05, "grad_norm": 3.09375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3453, "epoch": 1, "loss": 0.0166, "lr": 1.5919938659033016e-05, "grad_norm": 2.8125, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 3454, "epoch": 1, "loss": 0.1028, "lr": 1.591752426024861e-05, "grad_norm": 5.0, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3455, "epoch": 1, "loss": 0.0125, "lr": 1.5915109330513222e-05, "grad_norm": 4.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3456, "epoch": 1, "loss": 0.052, "lr": 1.5912693870043524e-05, "grad_norm": 4.28125, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 3457, "epoch": 1, "loss": 0.002, "lr": 1.5910277879056247e-05, "grad_norm": 1.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3458, "epoch": 1, "loss": 0.0109, "lr": 1.590786135776817e-05, "grad_norm": 6.78125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3459, "epoch": 1, "loss": 0.0204, "lr": 1.590544430639611e-05, "grad_norm": 4.8125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3460, "epoch": 1, "loss": 0.0047, "lr": 1.5903026725156942e-05, "grad_norm": 13.625, "samples_per_sec": 10.49, "gpu_mem_gb": 6.85} {"step": 3461, "epoch": 1, "loss": 0.0356, "lr": 1.590060861426758e-05, "grad_norm": 1.6171875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3462, "epoch": 1, "loss": 0.0019, "lr": 1.5898189973944988e-05, "grad_norm": 2.046875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 3463, "epoch": 1, "loss": 0.0209, "lr": 1.5895770804406186e-05, "grad_norm": 7.5625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3464, "epoch": 1, "loss": 0.1291, "lr": 1.589335110586823e-05, "grad_norm": 3.28125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3465, "epoch": 1, "loss": 0.0146, "lr": 1.589093087854823e-05, "grad_norm": 1.578125, "samples_per_sec": 8.89, "gpu_mem_gb": 6.82} {"step": 3466, "epoch": 1, "loss": 0.0896, "lr": 1.588851012266334e-05, "grad_norm": 5.65625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3467, "epoch": 1, "loss": 0.0241, "lr": 1.5886088838430763e-05, "grad_norm": 2.359375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 3468, "epoch": 1, "loss": 0.3053, "lr": 1.5883667026067745e-05, "grad_norm": 6.09375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 3469, "epoch": 1, "loss": 0.0025, "lr": 1.588124468579159e-05, "grad_norm": 4.21875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 3470, "epoch": 1, "loss": 0.0923, "lr": 1.587882181781964e-05, "grad_norm": 6.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3471, "epoch": 1, "loss": 0.0115, "lr": 1.587639842236929e-05, "grad_norm": 4.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3472, "epoch": 1, "loss": 0.0691, "lr": 1.5873974499657973e-05, "grad_norm": 7.15625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3473, "epoch": 1, "loss": 0.0024, "lr": 1.5871550049903185e-05, "grad_norm": 6.84375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 3474, "epoch": 1, "loss": 0.0325, "lr": 1.586912507332245e-05, "grad_norm": 4.96875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3475, "epoch": 1, "loss": 0.0011, "lr": 1.586669957013336e-05, "grad_norm": 2.1875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 3476, "epoch": 1, "loss": 0.0564, "lr": 1.5864273540553536e-05, "grad_norm": 4.5625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3477, "epoch": 1, "loss": 0.0411, "lr": 1.5861846984800658e-05, "grad_norm": 5.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3478, "epoch": 1, "loss": 0.0003, "lr": 1.585941990309245e-05, "grad_norm": 6.59375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 3479, "epoch": 1, "loss": 0.0002, "lr": 1.5856992295646675e-05, "grad_norm": 3.21875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3480, "epoch": 1, "loss": 0.2259, "lr": 1.585456416268116e-05, "grad_norm": 5.25, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3481, "epoch": 1, "loss": 0.0847, "lr": 1.585213550441376e-05, "grad_norm": 2.609375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3482, "epoch": 1, "loss": 0.0117, "lr": 1.5849706321062395e-05, "grad_norm": 1.6796875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3483, "epoch": 1, "loss": 0.0003, "lr": 1.5847276612845025e-05, "grad_norm": 2.140625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 3484, "epoch": 1, "loss": 0.0003, "lr": 1.584484637997965e-05, "grad_norm": 3.78125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3485, "epoch": 1, "loss": 0.0005, "lr": 1.584241562268432e-05, "grad_norm": 3.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3486, "epoch": 1, "loss": 0.0073, "lr": 1.5839984341177147e-05, "grad_norm": 4.46875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3487, "epoch": 1, "loss": 0.007, "lr": 1.583755253567627e-05, "grad_norm": 4.0, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 3488, "epoch": 1, "loss": 0.0121, "lr": 1.5835120206399884e-05, "grad_norm": 2.53125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 3489, "epoch": 1, "loss": 0.0038, "lr": 1.5832687353566232e-05, "grad_norm": 5.0, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3490, "epoch": 1, "loss": 0.1664, "lr": 1.58302539773936e-05, "grad_norm": 5.0625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3491, "epoch": 1, "loss": 0.0347, "lr": 1.5827820078100326e-05, "grad_norm": 5.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3492, "epoch": 1, "loss": 0.0014, "lr": 1.582538565590479e-05, "grad_norm": 1.484375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3493, "epoch": 1, "loss": 0.1695, "lr": 1.582295071102542e-05, "grad_norm": 7.46875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3494, "epoch": 1, "loss": 0.0914, "lr": 1.5820515243680693e-05, "grad_norm": 4.21875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3495, "epoch": 1, "loss": 0.0334, "lr": 1.5818079254089135e-05, "grad_norm": 5.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3496, "epoch": 1, "loss": 0.1756, "lr": 1.581564274246931e-05, "grad_norm": 4.96875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.84} {"step": 3497, "epoch": 1, "loss": 0.0179, "lr": 1.5813205709039842e-05, "grad_norm": 4.90625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3498, "epoch": 1, "loss": 0.1532, "lr": 1.5810768154019386e-05, "grad_norm": 5.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3499, "epoch": 1, "loss": 0.0006, "lr": 1.5808330077626654e-05, "grad_norm": 4.0625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3500, "epoch": 1, "loss": 0.0239, "lr": 1.5805891480080405e-05, "grad_norm": 4.09375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3501, "epoch": 1, "loss": 0.113, "lr": 1.5803452361599447e-05, "grad_norm": 6.71875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 3502, "epoch": 1, "loss": 0.269, "lr": 1.580101272240262e-05, "grad_norm": 10.1875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3503, "epoch": 1, "loss": 0.0017, "lr": 1.5798572562708827e-05, "grad_norm": 3.359375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 3504, "epoch": 1, "loss": 0.0515, "lr": 1.5796131882737013e-05, "grad_norm": 2.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 3505, "epoch": 1, "loss": 0.0831, "lr": 1.5793690682706167e-05, "grad_norm": 6.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3506, "epoch": 1, "loss": 0.0023, "lr": 1.5791248962835324e-05, "grad_norm": 7.9375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 3507, "epoch": 1, "loss": 0.0004, "lr": 1.5788806723343572e-05, "grad_norm": 6.53125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3508, "epoch": 1, "loss": 0.0607, "lr": 1.5786363964450034e-05, "grad_norm": 6.4375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3509, "epoch": 1, "loss": 0.0074, "lr": 1.57839206863739e-05, "grad_norm": 1.9375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3510, "epoch": 1, "loss": 0.0521, "lr": 1.578147688933438e-05, "grad_norm": 3.390625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 3511, "epoch": 1, "loss": 0.0192, "lr": 1.5779032573550756e-05, "grad_norm": 1.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3512, "epoch": 1, "loss": 0.002, "lr": 1.577658773924233e-05, "grad_norm": 1.3203125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3513, "epoch": 1, "loss": 0.1148, "lr": 1.577414238662848e-05, "grad_norm": 4.25, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 3514, "epoch": 1, "loss": 0.2701, "lr": 1.5771696515928608e-05, "grad_norm": 4.53125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3515, "epoch": 1, "loss": 0.2326, "lr": 1.5769250127362173e-05, "grad_norm": 4.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3516, "epoch": 1, "loss": 0.0003, "lr": 1.5766803221148676e-05, "grad_norm": 2.625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 3517, "epoch": 1, "loss": 0.0401, "lr": 1.576435579750766e-05, "grad_norm": 4.1875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3518, "epoch": 1, "loss": 0.0006, "lr": 1.5761907856658733e-05, "grad_norm": 4.125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3519, "epoch": 1, "loss": 0.0001, "lr": 1.575945939882153e-05, "grad_norm": 1.734375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3520, "epoch": 1, "loss": 0.1467, "lr": 1.5757010424215742e-05, "grad_norm": 4.25, "samples_per_sec": 10.23, "gpu_mem_gb": 6.83} {"step": 3521, "epoch": 1, "loss": 0.0119, "lr": 1.5754560933061104e-05, "grad_norm": 3.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3522, "epoch": 1, "loss": 0.0906, "lr": 1.575211092557739e-05, "grad_norm": 8.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3523, "epoch": 1, "loss": 0.0052, "lr": 1.5749660401984435e-05, "grad_norm": 1.984375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3524, "epoch": 1, "loss": 0.0055, "lr": 1.574720936250211e-05, "grad_norm": 1.078125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3525, "epoch": 1, "loss": 0.0377, "lr": 1.5744757807350336e-05, "grad_norm": 3.421875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3526, "epoch": 1, "loss": 0.1394, "lr": 1.574230573674908e-05, "grad_norm": 5.84375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3527, "epoch": 1, "loss": 0.0996, "lr": 1.5739853150918347e-05, "grad_norm": 4.90625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.84} {"step": 3528, "epoch": 1, "loss": 0.0282, "lr": 1.5737400050078203e-05, "grad_norm": 6.96875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3529, "epoch": 1, "loss": 0.0004, "lr": 1.5734946434448755e-05, "grad_norm": 0.138671875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3530, "epoch": 1, "loss": 0.0005, "lr": 1.5732492304250152e-05, "grad_norm": 6.3125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3531, "epoch": 1, "loss": 0.0007, "lr": 1.573003765970259e-05, "grad_norm": 5.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3532, "epoch": 1, "loss": 0.045, "lr": 1.5727582501026306e-05, "grad_norm": 3.96875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3533, "epoch": 1, "loss": 0.0265, "lr": 1.57251268284416e-05, "grad_norm": 4.78125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 3534, "epoch": 1, "loss": 0.0233, "lr": 1.5722670642168805e-05, "grad_norm": 3.390625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3535, "epoch": 1, "loss": 0.0102, "lr": 1.57202139424283e-05, "grad_norm": 5.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3536, "epoch": 1, "loss": 0.0013, "lr": 1.5717756729440515e-05, "grad_norm": 2.875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3537, "epoch": 1, "loss": 0.0002, "lr": 1.5715299003425925e-05, "grad_norm": 1.9453125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3538, "epoch": 1, "loss": 0.0011, "lr": 1.5712840764605046e-05, "grad_norm": 2.953125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3539, "epoch": 1, "loss": 0.0001, "lr": 1.5710382013198446e-05, "grad_norm": 3.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3540, "epoch": 1, "loss": 0.0209, "lr": 1.5707922749426735e-05, "grad_norm": 7.71875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3541, "epoch": 1, "loss": 0.0173, "lr": 1.570546297351058e-05, "grad_norm": 6.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3542, "epoch": 1, "loss": 0.0046, "lr": 1.570300268567067e-05, "grad_norm": 4.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3543, "epoch": 1, "loss": 0.0892, "lr": 1.570054188612777e-05, "grad_norm": 4.28125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 3544, "epoch": 1, "loss": 0.0736, "lr": 1.5698080575102662e-05, "grad_norm": 2.109375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3545, "epoch": 1, "loss": 0.0045, "lr": 1.56956187528162e-05, "grad_norm": 3.328125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 3546, "epoch": 1, "loss": 0.0231, "lr": 1.5693156419489263e-05, "grad_norm": 3.96875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3547, "epoch": 1, "loss": 0.0004, "lr": 1.5690693575342782e-05, "grad_norm": 5.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 3548, "epoch": 1, "loss": 0.0007, "lr": 1.5688230220597746e-05, "grad_norm": 4.6875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 3549, "epoch": 1, "loss": 0.0007, "lr": 1.5685766355475173e-05, "grad_norm": 6.15625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3550, "epoch": 1, "loss": 0.005, "lr": 1.5683301980196137e-05, "grad_norm": 0.6953125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3551, "epoch": 1, "loss": 0.1949, "lr": 1.568083709498175e-05, "grad_norm": 3.6875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3552, "epoch": 1, "loss": 0.0004, "lr": 1.5678371700053184e-05, "grad_norm": 5.5, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3553, "epoch": 1, "loss": 0.1269, "lr": 1.5675905795631634e-05, "grad_norm": 7.625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3554, "epoch": 1, "loss": 0.0083, "lr": 1.5673439381938365e-05, "grad_norm": 1.59375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3555, "epoch": 1, "loss": 0.0012, "lr": 1.567097245919467e-05, "grad_norm": 3.28125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3556, "epoch": 1, "loss": 0.0718, "lr": 1.5668505027621895e-05, "grad_norm": 4.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3557, "epoch": 1, "loss": 0.2375, "lr": 1.5666037087441434e-05, "grad_norm": 7.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.84} {"step": 3558, "epoch": 1, "loss": 0.3369, "lr": 1.566356863887472e-05, "grad_norm": 5.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3559, "epoch": 1, "loss": 0.0685, "lr": 1.566109968214323e-05, "grad_norm": 7.03125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3560, "epoch": 1, "loss": 0.1598, "lr": 1.5658630217468504e-05, "grad_norm": 9.25, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 3561, "epoch": 1, "loss": 0.0493, "lr": 1.565616024507211e-05, "grad_norm": 3.984375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3562, "epoch": 1, "loss": 0.0897, "lr": 1.5653689765175665e-05, "grad_norm": 3.890625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3563, "epoch": 1, "loss": 0.0095, "lr": 1.565121877800083e-05, "grad_norm": 6.4375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3564, "epoch": 1, "loss": 0.0907, "lr": 1.564874728376932e-05, "grad_norm": 3.5625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3565, "epoch": 1, "loss": 0.336, "lr": 1.564627528270289e-05, "grad_norm": 7.875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3566, "epoch": 1, "loss": 0.1396, "lr": 1.564380277502334e-05, "grad_norm": 5.125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 3567, "epoch": 1, "loss": 0.0039, "lr": 1.5641329760952514e-05, "grad_norm": 2.78125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3568, "epoch": 1, "loss": 0.004, "lr": 1.5638856240712303e-05, "grad_norm": 2.78125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3569, "epoch": 1, "loss": 0.0137, "lr": 1.563638221452465e-05, "grad_norm": 6.4375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 3570, "epoch": 1, "loss": 0.0288, "lr": 1.5633907682611535e-05, "grad_norm": 5.5, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 3571, "epoch": 1, "loss": 0.0004, "lr": 1.5631432645194982e-05, "grad_norm": 6.53125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3572, "epoch": 1, "loss": 0.0034, "lr": 1.5628957102497073e-05, "grad_norm": 2.90625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3573, "epoch": 1, "loss": 0.0134, "lr": 1.5626481054739916e-05, "grad_norm": 11.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3574, "epoch": 1, "loss": 0.0653, "lr": 1.562400450214568e-05, "grad_norm": 4.90625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3575, "epoch": 1, "loss": 0.0048, "lr": 1.562152744493657e-05, "grad_norm": 2.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3576, "epoch": 1, "loss": 0.1361, "lr": 1.5619049883334853e-05, "grad_norm": 5.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3577, "epoch": 1, "loss": 0.0251, "lr": 1.5616571817562817e-05, "grad_norm": 3.1875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3578, "epoch": 1, "loss": 0.0002, "lr": 1.5614093247842807e-05, "grad_norm": 5.28125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3579, "epoch": 1, "loss": 0.2807, "lr": 1.5611614174397212e-05, "grad_norm": 6.65625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3580, "epoch": 1, "loss": 0.0251, "lr": 1.5609134597448478e-05, "grad_norm": 2.796875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3581, "epoch": 1, "loss": 0.0007, "lr": 1.560665451721908e-05, "grad_norm": 3.0, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3582, "epoch": 1, "loss": 0.0011, "lr": 1.5604173933931536e-05, "grad_norm": 1.609375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3583, "epoch": 1, "loss": 0.1096, "lr": 1.5601692847808424e-05, "grad_norm": 4.375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3584, "epoch": 1, "loss": 0.0002, "lr": 1.559921125907236e-05, "grad_norm": 5.1875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3585, "epoch": 1, "loss": 0.0034, "lr": 1.559672916794601e-05, "grad_norm": 2.578125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 3586, "epoch": 1, "loss": 0.0038, "lr": 1.5594246574652064e-05, "grad_norm": 3.203125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 3587, "epoch": 1, "loss": 0.0006, "lr": 1.559176347941329e-05, "grad_norm": 4.8125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3588, "epoch": 1, "loss": 0.0004, "lr": 1.5589279882452476e-05, "grad_norm": 4.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3589, "epoch": 1, "loss": 0.008, "lr": 1.5586795783992467e-05, "grad_norm": 5.59375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3590, "epoch": 1, "loss": 0.0024, "lr": 1.5584311184256144e-05, "grad_norm": 2.078125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3591, "epoch": 1, "loss": 0.0021, "lr": 1.5581826083466442e-05, "grad_norm": 5.3125, "samples_per_sec": 10.35, "gpu_mem_gb": 6.82} {"step": 3592, "epoch": 1, "loss": 0.0079, "lr": 1.5579340481846338e-05, "grad_norm": 2.703125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3593, "epoch": 1, "loss": 0.0048, "lr": 1.557685437961885e-05, "grad_norm": 3.203125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3594, "epoch": 1, "loss": 0.1343, "lr": 1.557436777700704e-05, "grad_norm": 4.46875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 3595, "epoch": 1, "loss": 0.0003, "lr": 1.557188067423403e-05, "grad_norm": 3.015625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 3596, "epoch": 1, "loss": 0.0032, "lr": 1.556939307152297e-05, "grad_norm": 4.0625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3597, "epoch": 1, "loss": 0.1381, "lr": 1.5566904969097062e-05, "grad_norm": 3.796875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 3598, "epoch": 1, "loss": 0.027, "lr": 1.5564416367179543e-05, "grad_norm": 4.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 3599, "epoch": 1, "loss": 0.0124, "lr": 1.5561927265993717e-05, "grad_norm": 7.6875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3600, "epoch": 1, "loss": 0.0045, "lr": 1.5559437665762913e-05, "grad_norm": 3.375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3601, "epoch": 1, "loss": 0.0014, "lr": 1.555694756671051e-05, "grad_norm": 4.3125, "samples_per_sec": 10.35, "gpu_mem_gb": 6.83} {"step": 3602, "epoch": 1, "loss": 0.1758, "lr": 1.5554456969059927e-05, "grad_norm": 6.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3603, "epoch": 1, "loss": 0.0005, "lr": 1.5551965873034643e-05, "grad_norm": 5.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3604, "epoch": 1, "loss": 0.0234, "lr": 1.554947427885817e-05, "grad_norm": 3.09375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"epoch_summary": 1, "epoch_avg_loss": 0.17578376803551654, "val_loss": 0.049536630291272775} {"step": 3605, "epoch": 2, "loss": 0.0019, "lr": 1.5546982186754064e-05, "grad_norm": 5.1875, "samples_per_sec": 8.09, "gpu_mem_gb": 6.83} {"step": 3606, "epoch": 2, "loss": 0.0008, "lr": 1.5544489596945927e-05, "grad_norm": 1.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3607, "epoch": 2, "loss": 0.0025, "lr": 1.5541996509657414e-05, "grad_norm": 1.140625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3608, "epoch": 2, "loss": 0.1418, "lr": 1.5539502925112212e-05, "grad_norm": 4.78125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3609, "epoch": 2, "loss": 0.0003, "lr": 1.553700884353406e-05, "grad_norm": 8.875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3610, "epoch": 2, "loss": 0.0812, "lr": 1.5534514265146737e-05, "grad_norm": 3.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3611, "epoch": 2, "loss": 0.0181, "lr": 1.5532019190174074e-05, "grad_norm": 6.96875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3612, "epoch": 2, "loss": 0.0029, "lr": 1.5529523618839937e-05, "grad_norm": 1.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3613, "epoch": 2, "loss": 0.0001, "lr": 1.552702755136825e-05, "grad_norm": 2.40625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3614, "epoch": 2, "loss": 0.0439, "lr": 1.5524530987982966e-05, "grad_norm": 4.65625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 3615, "epoch": 2, "loss": 0.0004, "lr": 1.5522033928908092e-05, "grad_norm": 0.22265625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3616, "epoch": 2, "loss": 0.0149, "lr": 1.5519536374367676e-05, "grad_norm": 4.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3617, "epoch": 2, "loss": 0.0443, "lr": 1.5517038324585812e-05, "grad_norm": 5.3125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3618, "epoch": 2, "loss": 0.1445, "lr": 1.5514539779786638e-05, "grad_norm": 4.59375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.84} {"step": 3619, "epoch": 2, "loss": 0.0007, "lr": 1.5512040740194336e-05, "grad_norm": 6.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3620, "epoch": 2, "loss": 0.0012, "lr": 1.5509541206033138e-05, "grad_norm": 1.5078125, "samples_per_sec": 10.55, "gpu_mem_gb": 6.82} {"step": 3621, "epoch": 2, "loss": 0.0202, "lr": 1.5507041177527306e-05, "grad_norm": 2.09375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3622, "epoch": 2, "loss": 0.0362, "lr": 1.5504540654901158e-05, "grad_norm": 2.0, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3623, "epoch": 2, "loss": 0.0002, "lr": 1.550203963837906e-05, "grad_norm": 2.734375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3624, "epoch": 2, "loss": 0.0036, "lr": 1.5499538128185413e-05, "grad_norm": 3.546875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3625, "epoch": 2, "loss": 0.0002, "lr": 1.549703612454466e-05, "grad_norm": 3.09375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 3626, "epoch": 2, "loss": 0.0021, "lr": 1.5494533627681306e-05, "grad_norm": 4.0625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3627, "epoch": 2, "loss": 0.0127, "lr": 1.5492030637819874e-05, "grad_norm": 3.078125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 3628, "epoch": 2, "loss": 0.0712, "lr": 1.5489527155184956e-05, "grad_norm": 3.625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3629, "epoch": 2, "loss": 0.059, "lr": 1.5487023180001168e-05, "grad_norm": 6.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3630, "epoch": 2, "loss": 0.0004, "lr": 1.5484518712493188e-05, "grad_norm": 6.15625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3631, "epoch": 2, "loss": 0.0276, "lr": 1.5482013752885727e-05, "grad_norm": 6.40625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 3632, "epoch": 2, "loss": 0.2552, "lr": 1.547950830140354e-05, "grad_norm": 8.125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 3633, "epoch": 2, "loss": 0.1563, "lr": 1.5477002358271436e-05, "grad_norm": 5.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3634, "epoch": 2, "loss": 0.0019, "lr": 1.5474495923714252e-05, "grad_norm": 0.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3635, "epoch": 2, "loss": 0.0005, "lr": 1.5471988997956886e-05, "grad_norm": 4.15625, "samples_per_sec": 8.19, "gpu_mem_gb": 6.83} {"step": 3636, "epoch": 2, "loss": 0.06, "lr": 1.5469481581224274e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3637, "epoch": 2, "loss": 0.0696, "lr": 1.5466973673741383e-05, "grad_norm": 3.390625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 3638, "epoch": 2, "loss": 0.1121, "lr": 1.546446527573324e-05, "grad_norm": 6.3125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3639, "epoch": 2, "loss": 0.1206, "lr": 1.5461956387424923e-05, "grad_norm": 5.34375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3640, "epoch": 2, "loss": 0.0055, "lr": 1.545944700904153e-05, "grad_norm": 6.21875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3641, "epoch": 2, "loss": 0.0006, "lr": 1.545693714080822e-05, "grad_norm": 3.015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3642, "epoch": 2, "loss": 0.0219, "lr": 1.5454426782950185e-05, "grad_norm": 4.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3643, "epoch": 2, "loss": 0.0054, "lr": 1.5451915935692677e-05, "grad_norm": 2.859375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3644, "epoch": 2, "loss": 0.0001, "lr": 1.5449404599260976e-05, "grad_norm": 0.6640625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3645, "epoch": 2, "loss": 0.0148, "lr": 1.5446892773880415e-05, "grad_norm": 5.28125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3646, "epoch": 2, "loss": 0.0535, "lr": 1.5444380459776368e-05, "grad_norm": 3.8125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3647, "epoch": 2, "loss": 0.0085, "lr": 1.544186765717425e-05, "grad_norm": 8.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 3648, "epoch": 2, "loss": 0.075, "lr": 1.5439354366299528e-05, "grad_norm": 2.359375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3649, "epoch": 2, "loss": 0.0039, "lr": 1.54368405873777e-05, "grad_norm": 4.25, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 3650, "epoch": 2, "loss": 0.061, "lr": 1.5434326320634324e-05, "grad_norm": 2.59375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 3651, "epoch": 2, "loss": 0.0001, "lr": 1.5431811566294987e-05, "grad_norm": 1.6484375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3652, "epoch": 2, "loss": 0.0787, "lr": 1.5429296324585323e-05, "grad_norm": 2.703125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 3653, "epoch": 2, "loss": 0.0014, "lr": 1.5426780595731023e-05, "grad_norm": 6.21875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3654, "epoch": 2, "loss": 0.0732, "lr": 1.54242643799578e-05, "grad_norm": 2.09375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3655, "epoch": 2, "loss": 0.0398, "lr": 1.542174767749143e-05, "grad_norm": 5.8125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 3656, "epoch": 2, "loss": 0.0049, "lr": 1.541923048855772e-05, "grad_norm": 0.453125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3657, "epoch": 2, "loss": 0.066, "lr": 1.5416712813382528e-05, "grad_norm": 3.921875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 3658, "epoch": 2, "loss": 0.0004, "lr": 1.541419465219175e-05, "grad_norm": 3.71875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3659, "epoch": 2, "loss": 0.09, "lr": 1.541167600521133e-05, "grad_norm": 4.0625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 3660, "epoch": 2, "loss": 0.0027, "lr": 1.540915687266726e-05, "grad_norm": 6.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3661, "epoch": 2, "loss": 0.0542, "lr": 1.5406637254785556e-05, "grad_norm": 2.578125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3662, "epoch": 2, "loss": 0.1245, "lr": 1.5404117151792302e-05, "grad_norm": 5.9375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3663, "epoch": 2, "loss": 0.0955, "lr": 1.5401596563913615e-05, "grad_norm": 4.09375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 3664, "epoch": 2, "loss": 0.0468, "lr": 1.539907549137565e-05, "grad_norm": 5.28125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3665, "epoch": 2, "loss": 0.1143, "lr": 1.539655393440461e-05, "grad_norm": 3.921875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3666, "epoch": 2, "loss": 0.0006, "lr": 1.539403189322675e-05, "grad_norm": 2.78125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3667, "epoch": 2, "loss": 0.2249, "lr": 1.5391509368068353e-05, "grad_norm": 5.625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.84} {"step": 3668, "epoch": 2, "loss": 0.0209, "lr": 1.538898635915576e-05, "grad_norm": 6.125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 3669, "epoch": 2, "loss": 0.001, "lr": 1.538646286671534e-05, "grad_norm": 6.0625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 3670, "epoch": 2, "loss": 0.1462, "lr": 1.5383938890973514e-05, "grad_norm": 3.46875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3671, "epoch": 2, "loss": 0.1522, "lr": 1.5381414432156755e-05, "grad_norm": 4.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3672, "epoch": 2, "loss": 0.0552, "lr": 1.5378889490491565e-05, "grad_norm": 5.71875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.84} {"step": 3673, "epoch": 2, "loss": 0.0189, "lr": 1.5376364066204494e-05, "grad_norm": 9.8125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3674, "epoch": 2, "loss": 0.0329, "lr": 1.5373838159522137e-05, "grad_norm": 5.84375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 3675, "epoch": 2, "loss": 0.0005, "lr": 1.5371311770671138e-05, "grad_norm": 5.1875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3676, "epoch": 2, "loss": 0.0777, "lr": 1.5368784899878168e-05, "grad_norm": 4.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3677, "epoch": 2, "loss": 0.0198, "lr": 1.5366257547369954e-05, "grad_norm": 3.28125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3678, "epoch": 2, "loss": 0.0003, "lr": 1.536372971337327e-05, "grad_norm": 3.46875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 3679, "epoch": 2, "loss": 0.0269, "lr": 1.536120139811491e-05, "grad_norm": 4.34375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 3680, "epoch": 2, "loss": 0.0037, "lr": 1.5358672601821743e-05, "grad_norm": 2.234375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3681, "epoch": 2, "loss": 0.0006, "lr": 1.5356143324720663e-05, "grad_norm": 2.203125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3682, "epoch": 2, "loss": 0.1663, "lr": 1.5353613567038607e-05, "grad_norm": 6.59375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 3683, "epoch": 2, "loss": 0.0605, "lr": 1.535108332900256e-05, "grad_norm": 4.25, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3684, "epoch": 2, "loss": 0.0022, "lr": 1.534855261083954e-05, "grad_norm": 1.875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3685, "epoch": 2, "loss": 0.0004, "lr": 1.5346021412776627e-05, "grad_norm": 3.671875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3686, "epoch": 2, "loss": 0.0017, "lr": 1.534348973504093e-05, "grad_norm": 7.125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3687, "epoch": 2, "loss": 0.0716, "lr": 1.5340957577859605e-05, "grad_norm": 6.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3688, "epoch": 2, "loss": 0.0019, "lr": 1.5338424941459842e-05, "grad_norm": 5.46875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3689, "epoch": 2, "loss": 0.0971, "lr": 1.5335891826068894e-05, "grad_norm": 3.6875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3690, "epoch": 2, "loss": 0.0597, "lr": 1.5333358231914043e-05, "grad_norm": 6.84375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3691, "epoch": 2, "loss": 0.0846, "lr": 1.5330824159222605e-05, "grad_norm": 4.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3692, "epoch": 2, "loss": 0.2486, "lr": 1.5328289608221966e-05, "grad_norm": 7.03125, "samples_per_sec": 10.43, "gpu_mem_gb": 6.84} {"step": 3693, "epoch": 2, "loss": 0.0271, "lr": 1.532575457913953e-05, "grad_norm": 2.46875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3694, "epoch": 2, "loss": 0.0586, "lr": 1.532321907220275e-05, "grad_norm": 3.828125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 3695, "epoch": 2, "loss": 0.0012, "lr": 1.5320683087639135e-05, "grad_norm": 2.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3696, "epoch": 2, "loss": 0.0774, "lr": 1.531814662567622e-05, "grad_norm": 6.1875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3697, "epoch": 2, "loss": 0.009, "lr": 1.5315609686541587e-05, "grad_norm": 5.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3698, "epoch": 2, "loss": 0.0103, "lr": 1.5313072270462873e-05, "grad_norm": 6.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3699, "epoch": 2, "loss": 0.1132, "lr": 1.5310534377667737e-05, "grad_norm": 4.78125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 3700, "epoch": 2, "loss": 0.0002, "lr": 1.5307996008383897e-05, "grad_norm": 4.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 3701, "epoch": 2, "loss": 0.0382, "lr": 1.530545716283911e-05, "grad_norm": 4.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3702, "epoch": 2, "loss": 0.0027, "lr": 1.5302917841261172e-05, "grad_norm": 3.578125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 3703, "epoch": 2, "loss": 0.2072, "lr": 1.530037804387793e-05, "grad_norm": 7.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3704, "epoch": 2, "loss": 0.0001, "lr": 1.5297837770917258e-05, "grad_norm": 3.984375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3705, "epoch": 2, "loss": 0.029, "lr": 1.529529702260709e-05, "grad_norm": 4.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3706, "epoch": 2, "loss": 0.0719, "lr": 1.529275579917539e-05, "grad_norm": 4.09375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.84} {"step": 3707, "epoch": 2, "loss": 0.001, "lr": 1.5290214100850175e-05, "grad_norm": 3.296875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3708, "epoch": 2, "loss": 0.0029, "lr": 1.5287671927859494e-05, "grad_norm": 1.71875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3709, "epoch": 2, "loss": 0.0213, "lr": 1.5285129280431446e-05, "grad_norm": 6.40625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 3710, "epoch": 2, "loss": 0.0288, "lr": 1.5282586158794178e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3711, "epoch": 2, "loss": 0.1067, "lr": 1.528004256317586e-05, "grad_norm": 3.8125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 3712, "epoch": 2, "loss": 0.0278, "lr": 1.5277498493804723e-05, "grad_norm": 3.71875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 3713, "epoch": 2, "loss": 0.0398, "lr": 1.5274953950909033e-05, "grad_norm": 8.6875, "samples_per_sec": 10.35, "gpu_mem_gb": 6.85} {"step": 3714, "epoch": 2, "loss": 0.0039, "lr": 1.5272408934717098e-05, "grad_norm": 2.84375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3715, "epoch": 2, "loss": 0.0488, "lr": 1.5269863445457273e-05, "grad_norm": 1.875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3716, "epoch": 2, "loss": 0.0197, "lr": 1.526731748335795e-05, "grad_norm": 2.6875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3717, "epoch": 2, "loss": 0.1084, "lr": 1.526477104864757e-05, "grad_norm": 4.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3718, "epoch": 2, "loss": 0.0011, "lr": 1.5262224141554605e-05, "grad_norm": 0.466796875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3719, "epoch": 2, "loss": 0.0619, "lr": 1.5259676762307582e-05, "grad_norm": 6.34375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 3720, "epoch": 2, "loss": 0.0064, "lr": 1.5257128911135069e-05, "grad_norm": 4.5625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3721, "epoch": 2, "loss": 0.009, "lr": 1.5254580588265664e-05, "grad_norm": 1.84375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3722, "epoch": 2, "loss": 0.0002, "lr": 1.5252031793928019e-05, "grad_norm": 2.40625, "samples_per_sec": 11.01, "gpu_mem_gb": 6.82} {"step": 3723, "epoch": 2, "loss": 0.0003, "lr": 1.5249482528350827e-05, "grad_norm": 5.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3724, "epoch": 2, "loss": 0.0264, "lr": 1.5246932791762818e-05, "grad_norm": 4.5625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3725, "epoch": 2, "loss": 0.1032, "lr": 1.5244382584392772e-05, "grad_norm": 5.5625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3726, "epoch": 2, "loss": 0.0068, "lr": 1.5241831906469502e-05, "grad_norm": 4.03125, "samples_per_sec": 10.29, "gpu_mem_gb": 6.82} {"step": 3727, "epoch": 2, "loss": 0.0164, "lr": 1.523928075822187e-05, "grad_norm": 5.625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3728, "epoch": 2, "loss": 0.0176, "lr": 1.523672913987878e-05, "grad_norm": 7.65625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 3729, "epoch": 2, "loss": 0.0361, "lr": 1.5234177051669175e-05, "grad_norm": 4.625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3730, "epoch": 2, "loss": 0.0017, "lr": 1.5231624493822042e-05, "grad_norm": 4.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3731, "epoch": 2, "loss": 0.0029, "lr": 1.5229071466566405e-05, "grad_norm": 3.125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3732, "epoch": 2, "loss": 0.0071, "lr": 1.5226517970131345e-05, "grad_norm": 4.0625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3733, "epoch": 2, "loss": 0.0532, "lr": 1.5223964004745966e-05, "grad_norm": 3.890625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 3734, "epoch": 2, "loss": 0.0001, "lr": 1.5221409570639428e-05, "grad_norm": 6.71875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3735, "epoch": 2, "loss": 0.0168, "lr": 1.5218854668040923e-05, "grad_norm": 4.84375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3736, "epoch": 2, "loss": 0.0008, "lr": 1.5216299297179694e-05, "grad_norm": 2.65625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 3737, "epoch": 2, "loss": 0.0004, "lr": 1.5213743458285027e-05, "grad_norm": 0.408203125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3738, "epoch": 2, "loss": 0.0152, "lr": 1.5211187151586233e-05, "grad_norm": 1.21875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3739, "epoch": 2, "loss": 0.0079, "lr": 1.5208630377312686e-05, "grad_norm": 7.6875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 3740, "epoch": 2, "loss": 0.0027, "lr": 1.520607313569379e-05, "grad_norm": 0.322265625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3741, "epoch": 2, "loss": 0.0012, "lr": 1.5203515426958997e-05, "grad_norm": 1.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3742, "epoch": 2, "loss": 0.2076, "lr": 1.5200957251337796e-05, "grad_norm": 5.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3743, "epoch": 2, "loss": 0.0047, "lr": 1.5198398609059715e-05, "grad_norm": 3.453125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3744, "epoch": 2, "loss": 0.1861, "lr": 1.5195839500354337e-05, "grad_norm": 3.5625, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 3745, "epoch": 2, "loss": 0.2762, "lr": 1.5193279925451276e-05, "grad_norm": 6.3125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3746, "epoch": 2, "loss": 0.1637, "lr": 1.5190719884580185e-05, "grad_norm": 3.546875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.85} {"step": 3747, "epoch": 2, "loss": 0.0068, "lr": 1.5188159377970771e-05, "grad_norm": 8.75, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3748, "epoch": 2, "loss": 0.3864, "lr": 1.5185598405852769e-05, "grad_norm": 5.625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3749, "epoch": 2, "loss": 0.1485, "lr": 1.5183036968455971e-05, "grad_norm": 3.453125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3750, "epoch": 2, "loss": 0.0014, "lr": 1.5180475066010195e-05, "grad_norm": 6.96875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3751, "epoch": 2, "loss": 0.0005, "lr": 1.5177912698745314e-05, "grad_norm": 1.5390625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3752, "epoch": 2, "loss": 0.0062, "lr": 1.5175349866891234e-05, "grad_norm": 0.94140625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 3753, "epoch": 2, "loss": 0.2357, "lr": 1.5172786570677906e-05, "grad_norm": 6.5, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 3754, "epoch": 2, "loss": 0.3376, "lr": 1.5170222810335322e-05, "grad_norm": 5.59375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3755, "epoch": 2, "loss": 0.0004, "lr": 1.5167658586093517e-05, "grad_norm": 1.5546875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 3756, "epoch": 2, "loss": 0.0024, "lr": 1.516509389818256e-05, "grad_norm": 6.0625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 3757, "epoch": 2, "loss": 0.121, "lr": 1.516252874683258e-05, "grad_norm": 6.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3758, "epoch": 2, "loss": 0.0211, "lr": 1.5159963132273727e-05, "grad_norm": 3.09375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3759, "epoch": 2, "loss": 0.0013, "lr": 1.5157397054736204e-05, "grad_norm": 0.66796875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3760, "epoch": 2, "loss": 0.0961, "lr": 1.515483051445025e-05, "grad_norm": 4.5, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3761, "epoch": 2, "loss": 0.0014, "lr": 1.5152263511646155e-05, "grad_norm": 4.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3762, "epoch": 2, "loss": 0.0258, "lr": 1.5149696046554236e-05, "grad_norm": 6.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3763, "epoch": 2, "loss": 0.0006, "lr": 1.5147128119404863e-05, "grad_norm": 4.5, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 3764, "epoch": 2, "loss": 0.0001, "lr": 1.5144559730428445e-05, "grad_norm": 5.15625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3765, "epoch": 2, "loss": 0.0009, "lr": 1.5141990879855431e-05, "grad_norm": 4.15625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 3766, "epoch": 2, "loss": 0.0021, "lr": 1.5139421567916306e-05, "grad_norm": 8.4375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 3767, "epoch": 2, "loss": 0.1054, "lr": 1.5136851794841611e-05, "grad_norm": 3.546875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 3768, "epoch": 2, "loss": 0.0227, "lr": 1.5134281560861912e-05, "grad_norm": 4.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3769, "epoch": 2, "loss": 0.0832, "lr": 1.5131710866207829e-05, "grad_norm": 4.0625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3770, "epoch": 2, "loss": 0.002, "lr": 1.512913971111001e-05, "grad_norm": 4.1875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3771, "epoch": 2, "loss": 0.0003, "lr": 1.5126568095799163e-05, "grad_norm": 5.75, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3772, "epoch": 2, "loss": 0.0053, "lr": 1.5123996020506022e-05, "grad_norm": 4.71875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 3773, "epoch": 2, "loss": 0.0708, "lr": 1.5121423485461364e-05, "grad_norm": 3.1875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 3774, "epoch": 2, "loss": 0.0187, "lr": 1.5118850490896012e-05, "grad_norm": 5.6875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3775, "epoch": 2, "loss": 0.0413, "lr": 1.511627703704083e-05, "grad_norm": 5.375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3776, "epoch": 2, "loss": 0.0746, "lr": 1.5113703124126723e-05, "grad_norm": 12.3125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3777, "epoch": 2, "loss": 0.0013, "lr": 1.511112875238463e-05, "grad_norm": 3.625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3778, "epoch": 2, "loss": 0.0003, "lr": 1.510855392204554e-05, "grad_norm": 3.75, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3779, "epoch": 2, "loss": 0.0135, "lr": 1.5105978633340486e-05, "grad_norm": 3.328125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3780, "epoch": 2, "loss": 0.0003, "lr": 1.5103402886500526e-05, "grad_norm": 0.96875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 3781, "epoch": 2, "loss": 0.0035, "lr": 1.5100826681756777e-05, "grad_norm": 2.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3782, "epoch": 2, "loss": 0.0566, "lr": 1.5098250019340385e-05, "grad_norm": 2.796875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3783, "epoch": 2, "loss": 0.0087, "lr": 1.5095672899482546e-05, "grad_norm": 4.28125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 3784, "epoch": 2, "loss": 0.0001, "lr": 1.5093095322414488e-05, "grad_norm": 2.015625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3785, "epoch": 2, "loss": 0.0722, "lr": 1.5090517288367485e-05, "grad_norm": 7.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3786, "epoch": 2, "loss": 0.0004, "lr": 1.5087938797572854e-05, "grad_norm": 0.52734375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3787, "epoch": 2, "loss": 0.0833, "lr": 1.5085359850261951e-05, "grad_norm": 4.75, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 3788, "epoch": 2, "loss": 0.0005, "lr": 1.508278044666617e-05, "grad_norm": 8.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 3789, "epoch": 2, "loss": 0.0003, "lr": 1.5080200587016953e-05, "grad_norm": 2.359375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 3790, "epoch": 2, "loss": 0.0066, "lr": 1.507762027154577e-05, "grad_norm": 2.34375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 3791, "epoch": 2, "loss": 0.1248, "lr": 1.5075039500484147e-05, "grad_norm": 3.234375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 3792, "epoch": 2, "loss": 0.072, "lr": 1.5072458274063645e-05, "grad_norm": 3.015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3793, "epoch": 2, "loss": 0.1148, "lr": 1.506987659251586e-05, "grad_norm": 4.53125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3794, "epoch": 2, "loss": 0.0103, "lr": 1.5067294456072437e-05, "grad_norm": 13.6875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 3795, "epoch": 2, "loss": 0.0169, "lr": 1.5064711864965058e-05, "grad_norm": 1.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3796, "epoch": 2, "loss": 0.0006, "lr": 1.5062128819425447e-05, "grad_norm": 1.8046875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 3797, "epoch": 2, "loss": 0.0184, "lr": 1.505954531968537e-05, "grad_norm": 3.515625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3798, "epoch": 2, "loss": 0.0387, "lr": 1.5056961365976626e-05, "grad_norm": 2.796875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3799, "epoch": 2, "loss": 0.2259, "lr": 1.5054376958531066e-05, "grad_norm": 4.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3800, "epoch": 2, "loss": 0.0017, "lr": 1.5051792097580576e-05, "grad_norm": 2.984375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 3801, "epoch": 2, "loss": 0.0567, "lr": 1.5049206783357082e-05, "grad_norm": 4.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3802, "epoch": 2, "loss": 0.0305, "lr": 1.5046621016092549e-05, "grad_norm": 2.65625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3803, "epoch": 2, "loss": 0.0002, "lr": 1.5044034796018992e-05, "grad_norm": 3.859375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3804, "epoch": 2, "loss": 0.0601, "lr": 1.5041448123368454e-05, "grad_norm": 3.125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3805, "epoch": 2, "loss": 0.0192, "lr": 1.5038860998373029e-05, "grad_norm": 1.265625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3806, "epoch": 2, "loss": 0.0418, "lr": 1.5036273421264845e-05, "grad_norm": 1.671875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3807, "epoch": 2, "loss": 0.108, "lr": 1.5033685392276071e-05, "grad_norm": 5.75, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 3808, "epoch": 2, "loss": 0.0168, "lr": 1.5031096911638927e-05, "grad_norm": 5.5625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3809, "epoch": 2, "loss": 0.0804, "lr": 1.5028507979585654e-05, "grad_norm": 3.609375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3810, "epoch": 2, "loss": 0.0595, "lr": 1.5025918596348548e-05, "grad_norm": 6.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3811, "epoch": 2, "loss": 0.2433, "lr": 1.5023328762159945e-05, "grad_norm": 5.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3812, "epoch": 2, "loss": 0.1263, "lr": 1.5020738477252217e-05, "grad_norm": 8.375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3813, "epoch": 2, "loss": 0.0001, "lr": 1.5018147741857776e-05, "grad_norm": 2.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 3814, "epoch": 2, "loss": 0.0087, "lr": 1.5015556556209076e-05, "grad_norm": 3.453125, "samples_per_sec": 10.52, "gpu_mem_gb": 6.85} {"step": 3815, "epoch": 2, "loss": 0.1811, "lr": 1.5012964920538617e-05, "grad_norm": 6.65625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 3816, "epoch": 2, "loss": 0.0215, "lr": 1.501037283507893e-05, "grad_norm": 3.390625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 3817, "epoch": 2, "loss": 0.0003, "lr": 1.5007780300062585e-05, "grad_norm": 6.0625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 3818, "epoch": 2, "loss": 0.0013, "lr": 1.5005187315722208e-05, "grad_norm": 4.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3819, "epoch": 2, "loss": 0.0722, "lr": 1.5002593882290447e-05, "grad_norm": 7.84375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 3820, "epoch": 2, "loss": 0.0836, "lr": 1.5000000000000002e-05, "grad_norm": 5.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3821, "epoch": 2, "loss": 0.0348, "lr": 1.4997405669083609e-05, "grad_norm": 4.75, "samples_per_sec": 10.43, "gpu_mem_gb": 6.82} {"step": 3822, "epoch": 2, "loss": 0.002, "lr": 1.4994810889774045e-05, "grad_norm": 0.375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3823, "epoch": 2, "loss": 0.0305, "lr": 1.4992215662304128e-05, "grad_norm": 2.75, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3824, "epoch": 2, "loss": 0.0061, "lr": 1.4989619986906711e-05, "grad_norm": 5.6875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 3825, "epoch": 2, "loss": 0.0029, "lr": 1.4987023863814698e-05, "grad_norm": 3.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3826, "epoch": 2, "loss": 0.0005, "lr": 1.4984427293261022e-05, "grad_norm": 7.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3827, "epoch": 2, "loss": 0.0782, "lr": 1.4981830275478662e-05, "grad_norm": 4.84375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 3828, "epoch": 2, "loss": 0.0001, "lr": 1.4979232810700638e-05, "grad_norm": 5.5, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 3829, "epoch": 2, "loss": 0.0007, "lr": 1.4976634899160002e-05, "grad_norm": 1.15625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 3830, "epoch": 2, "loss": 0.0222, "lr": 1.4974036541089855e-05, "grad_norm": 4.15625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 3831, "epoch": 2, "loss": 0.0954, "lr": 1.497143773672334e-05, "grad_norm": 4.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3832, "epoch": 2, "loss": 0.004, "lr": 1.4968838486293633e-05, "grad_norm": 2.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3833, "epoch": 2, "loss": 0.0781, "lr": 1.4966238790033947e-05, "grad_norm": 3.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3834, "epoch": 2, "loss": 0.0006, "lr": 1.4963638648177544e-05, "grad_norm": 0.8125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 3835, "epoch": 2, "loss": 0.0545, "lr": 1.4961038060957724e-05, "grad_norm": 2.453125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 3836, "epoch": 2, "loss": 0.0061, "lr": 1.4958437028607821e-05, "grad_norm": 3.828125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3837, "epoch": 2, "loss": 0.01, "lr": 1.4955835551361217e-05, "grad_norm": 5.90625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 3838, "epoch": 2, "loss": 0.0048, "lr": 1.4953233629451332e-05, "grad_norm": 3.875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3839, "epoch": 2, "loss": 0.0001, "lr": 1.4950631263111615e-05, "grad_norm": 4.9375, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 3840, "epoch": 2, "loss": 0.1296, "lr": 1.4948028452575572e-05, "grad_norm": 3.953125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.84} {"step": 3841, "epoch": 2, "loss": 0.0422, "lr": 1.4945425198076735e-05, "grad_norm": 6.40625, "samples_per_sec": 10.37, "gpu_mem_gb": 6.83} {"step": 3842, "epoch": 2, "loss": 0.0444, "lr": 1.4942821499848687e-05, "grad_norm": 5.03125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.85} {"step": 3843, "epoch": 2, "loss": 0.0033, "lr": 1.4940217358125042e-05, "grad_norm": 5.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3844, "epoch": 2, "loss": 0.0023, "lr": 1.4937612773139458e-05, "grad_norm": 4.84375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3845, "epoch": 2, "loss": 0.0005, "lr": 1.4935007745125632e-05, "grad_norm": 2.59375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3846, "epoch": 2, "loss": 0.0622, "lr": 1.4932402274317297e-05, "grad_norm": 4.90625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3847, "epoch": 2, "loss": 0.1175, "lr": 1.4929796360948237e-05, "grad_norm": 4.28125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 3848, "epoch": 2, "loss": 0.0001, "lr": 1.4927190005252263e-05, "grad_norm": 5.375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 3849, "epoch": 2, "loss": 0.0252, "lr": 1.492458320746323e-05, "grad_norm": 4.5625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3850, "epoch": 2, "loss": 0.0451, "lr": 1.4921975967815036e-05, "grad_norm": 3.859375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 3851, "epoch": 2, "loss": 0.002, "lr": 1.4919368286541616e-05, "grad_norm": 1.84375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 3852, "epoch": 2, "loss": 0.0557, "lr": 1.491676016387694e-05, "grad_norm": 3.984375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3853, "epoch": 2, "loss": 0.0277, "lr": 1.4914151600055031e-05, "grad_norm": 6.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3854, "epoch": 2, "loss": 0.059, "lr": 1.4911542595309935e-05, "grad_norm": 6.3125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 3855, "epoch": 2, "loss": 0.0004, "lr": 1.490893314987575e-05, "grad_norm": 3.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 3856, "epoch": 2, "loss": 0.0182, "lr": 1.4906323263986605e-05, "grad_norm": 2.734375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3857, "epoch": 2, "loss": 0.0707, "lr": 1.4903712937876676e-05, "grad_norm": 3.140625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3858, "epoch": 2, "loss": 0.3461, "lr": 1.4901102171780175e-05, "grad_norm": 8.8125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 3859, "epoch": 2, "loss": 0.0211, "lr": 1.4898490965931349e-05, "grad_norm": 1.7734375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 3860, "epoch": 2, "loss": 0.0163, "lr": 1.4895879320564494e-05, "grad_norm": 2.46875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3861, "epoch": 2, "loss": 0.1106, "lr": 1.489326723591394e-05, "grad_norm": 7.1875, "samples_per_sec": 10.45, "gpu_mem_gb": 6.82} {"step": 3862, "epoch": 2, "loss": 0.0062, "lr": 1.4890654712214056e-05, "grad_norm": 4.71875, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 3863, "epoch": 2, "loss": 0.0461, "lr": 1.4888041749699249e-05, "grad_norm": 2.390625, "samples_per_sec": 8.99, "gpu_mem_gb": 6.83} {"step": 3864, "epoch": 2, "loss": 0.0245, "lr": 1.4885428348603968e-05, "grad_norm": 4.1875, "samples_per_sec": 11.68, "gpu_mem_gb": 6.83} {"step": 3865, "epoch": 2, "loss": 0.0906, "lr": 1.4882814509162705e-05, "grad_norm": 2.875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 3866, "epoch": 2, "loss": 0.0038, "lr": 1.4880200231609982e-05, "grad_norm": 5.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3867, "epoch": 2, "loss": 0.049, "lr": 1.487758551618037e-05, "grad_norm": 2.796875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3868, "epoch": 2, "loss": 0.0057, "lr": 1.487497036310847e-05, "grad_norm": 3.5, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 3869, "epoch": 2, "loss": 0.1577, "lr": 1.487235477262893e-05, "grad_norm": 7.71875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3870, "epoch": 2, "loss": 0.005, "lr": 1.4869738744976436e-05, "grad_norm": 2.828125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3871, "epoch": 2, "loss": 0.0006, "lr": 1.486712228038571e-05, "grad_norm": 4.25, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 3872, "epoch": 2, "loss": 0.0345, "lr": 1.4864505379091512e-05, "grad_norm": 5.78125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3873, "epoch": 2, "loss": 0.0626, "lr": 1.486188804132865e-05, "grad_norm": 3.578125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3874, "epoch": 2, "loss": 0.081, "lr": 1.485927026733196e-05, "grad_norm": 3.375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 3875, "epoch": 2, "loss": 0.0036, "lr": 1.4856652057336324e-05, "grad_norm": 3.671875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3876, "epoch": 2, "loss": 0.0035, "lr": 1.4854033411576659e-05, "grad_norm": 3.3125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 3877, "epoch": 2, "loss": 0.0003, "lr": 1.485141433028793e-05, "grad_norm": 4.65625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3878, "epoch": 2, "loss": 0.0037, "lr": 1.4848794813705128e-05, "grad_norm": 2.9375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 3879, "epoch": 2, "loss": 0.1861, "lr": 1.4846174862063292e-05, "grad_norm": 6.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3880, "epoch": 2, "loss": 0.0325, "lr": 1.4843554475597498e-05, "grad_norm": 3.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 3881, "epoch": 2, "loss": 0.0178, "lr": 1.4840933654542857e-05, "grad_norm": 6.46875, "samples_per_sec": 10.4, "gpu_mem_gb": 6.82} {"step": 3882, "epoch": 2, "loss": 0.0011, "lr": 1.4838312399134531e-05, "grad_norm": 4.90625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3883, "epoch": 2, "loss": 0.0001, "lr": 1.4835690709607705e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 3884, "epoch": 2, "loss": 0.0001, "lr": 1.4833068586197612e-05, "grad_norm": 2.78125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 3885, "epoch": 2, "loss": 0.0754, "lr": 1.4830446029139526e-05, "grad_norm": 3.890625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3886, "epoch": 2, "loss": 0.0275, "lr": 1.4827823038668753e-05, "grad_norm": 2.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3887, "epoch": 2, "loss": 0.0006, "lr": 1.4825199615020644e-05, "grad_norm": 0.93359375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 3888, "epoch": 2, "loss": 0.038, "lr": 1.4822575758430581e-05, "grad_norm": 4.5625, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 3889, "epoch": 2, "loss": 0.0004, "lr": 1.4819951469133997e-05, "grad_norm": 0.5234375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3890, "epoch": 2, "loss": 0.1018, "lr": 1.4817326747366354e-05, "grad_norm": 5.59375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3891, "epoch": 2, "loss": 0.0041, "lr": 1.4814701593363154e-05, "grad_norm": 2.453125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3892, "epoch": 2, "loss": 0.0016, "lr": 1.4812076007359941e-05, "grad_norm": 5.46875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 3893, "epoch": 2, "loss": 0.0012, "lr": 1.4809449989592295e-05, "grad_norm": 4.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 3894, "epoch": 2, "loss": 0.0007, "lr": 1.4806823540295839e-05, "grad_norm": 5.4375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3895, "epoch": 2, "loss": 0.019, "lr": 1.4804196659706224e-05, "grad_norm": 5.75, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 3896, "epoch": 2, "loss": 0.002, "lr": 1.4801569348059158e-05, "grad_norm": 1.3671875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3897, "epoch": 2, "loss": 0.1223, "lr": 1.4798941605590372e-05, "grad_norm": 3.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3898, "epoch": 2, "loss": 0.0024, "lr": 1.4796313432535636e-05, "grad_norm": 2.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3899, "epoch": 2, "loss": 0.1998, "lr": 1.479368482913077e-05, "grad_norm": 5.3125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3900, "epoch": 2, "loss": 0.0156, "lr": 1.4791055795611623e-05, "grad_norm": 3.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3901, "epoch": 2, "loss": 0.2965, "lr": 1.478842633221409e-05, "grad_norm": 4.59375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 3902, "epoch": 2, "loss": 0.0216, "lr": 1.4785796439174093e-05, "grad_norm": 1.046875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 3903, "epoch": 2, "loss": 0.0253, "lr": 1.4783166116727603e-05, "grad_norm": 2.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3904, "epoch": 2, "loss": 0.0015, "lr": 1.4780535365110629e-05, "grad_norm": 3.796875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3905, "epoch": 2, "loss": 0.0535, "lr": 1.4777904184559208e-05, "grad_norm": 3.96875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.85} {"step": 3906, "epoch": 2, "loss": 0.1384, "lr": 1.4775272575309434e-05, "grad_norm": 3.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3907, "epoch": 2, "loss": 0.1657, "lr": 1.477264053759742e-05, "grad_norm": 3.265625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 3908, "epoch": 2, "loss": 0.0603, "lr": 1.4770008071659327e-05, "grad_norm": 4.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3909, "epoch": 2, "loss": 0.0061, "lr": 1.4767375177731358e-05, "grad_norm": 3.796875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 3910, "epoch": 2, "loss": 0.1002, "lr": 1.4764741856049747e-05, "grad_norm": 3.421875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 3911, "epoch": 2, "loss": 0.0338, "lr": 1.4762108106850768e-05, "grad_norm": 3.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 3912, "epoch": 2, "loss": 0.0004, "lr": 1.4759473930370738e-05, "grad_norm": 6.25, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 3913, "epoch": 2, "loss": 0.2048, "lr": 1.4756839326846003e-05, "grad_norm": 5.6875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 3914, "epoch": 2, "loss": 0.0107, "lr": 1.4754204296512964e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 3915, "epoch": 2, "loss": 0.0036, "lr": 1.4751568839608036e-05, "grad_norm": 0.96875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 3916, "epoch": 2, "loss": 0.001, "lr": 1.4748932956367699e-05, "grad_norm": 0.171875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 3917, "epoch": 2, "loss": 0.005, "lr": 1.4746296647028448e-05, "grad_norm": 2.90625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3918, "epoch": 2, "loss": 0.0901, "lr": 1.4743659911826833e-05, "grad_norm": 7.375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3919, "epoch": 2, "loss": 0.0065, "lr": 1.4741022750999432e-05, "grad_norm": 3.234375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 3920, "epoch": 2, "loss": 0.2846, "lr": 1.4738385164782865e-05, "grad_norm": 5.6875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 3921, "epoch": 2, "loss": 0.0209, "lr": 1.4735747153413793e-05, "grad_norm": 6.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3922, "epoch": 2, "loss": 0.0023, "lr": 1.4733108717128906e-05, "grad_norm": 4.46875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.85} {"step": 3923, "epoch": 2, "loss": 0.0122, "lr": 1.4730469856164944e-05, "grad_norm": 2.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3924, "epoch": 2, "loss": 0.1341, "lr": 1.472783057075868e-05, "grad_norm": 6.9375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3925, "epoch": 2, "loss": 0.0002, "lr": 1.4725190861146917e-05, "grad_norm": 3.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3926, "epoch": 2, "loss": 0.155, "lr": 1.472255072756651e-05, "grad_norm": 5.8125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3927, "epoch": 2, "loss": 0.0403, "lr": 1.471991017025434e-05, "grad_norm": 9.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3928, "epoch": 2, "loss": 0.1258, "lr": 1.471726918944734e-05, "grad_norm": 4.21875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 3929, "epoch": 2, "loss": 0.137, "lr": 1.4714627785382468e-05, "grad_norm": 6.59375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 3930, "epoch": 2, "loss": 0.0075, "lr": 1.4711985958296722e-05, "grad_norm": 1.2578125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 3931, "epoch": 2, "loss": 0.1106, "lr": 1.470934370842714e-05, "grad_norm": 4.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3932, "epoch": 2, "loss": 0.0837, "lr": 1.4706701036010801e-05, "grad_norm": 11.4375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 3933, "epoch": 2, "loss": 0.0845, "lr": 1.4704057941284823e-05, "grad_norm": 6.0625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3934, "epoch": 2, "loss": 0.0001, "lr": 1.4701414424486353e-05, "grad_norm": 4.1875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 3935, "epoch": 2, "loss": 0.001, "lr": 1.4698770485852581e-05, "grad_norm": 5.875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 3936, "epoch": 2, "loss": 0.0472, "lr": 1.4696126125620738e-05, "grad_norm": 3.75, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 3937, "epoch": 2, "loss": 0.016, "lr": 1.4693481344028087e-05, "grad_norm": 4.65625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 3938, "epoch": 2, "loss": 0.0474, "lr": 1.4690836141311933e-05, "grad_norm": 4.96875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 3939, "epoch": 2, "loss": 0.1665, "lr": 1.4688190517709616e-05, "grad_norm": 4.8125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3940, "epoch": 2, "loss": 0.0028, "lr": 1.4685544473458514e-05, "grad_norm": 6.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3941, "epoch": 2, "loss": 0.0088, "lr": 1.468289800879605e-05, "grad_norm": 3.0, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 3942, "epoch": 2, "loss": 0.0002, "lr": 1.468025112395967e-05, "grad_norm": 4.40625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 3943, "epoch": 2, "loss": 0.0003, "lr": 1.4677603819186873e-05, "grad_norm": 6.34375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 3944, "epoch": 2, "loss": 0.0924, "lr": 1.4674956094715184e-05, "grad_norm": 5.65625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 3945, "epoch": 2, "loss": 0.1821, "lr": 1.4672307950782179e-05, "grad_norm": 6.28125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 3946, "epoch": 2, "loss": 0.0012, "lr": 1.466965938762545e-05, "grad_norm": 4.25, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3947, "epoch": 2, "loss": 0.0291, "lr": 1.4667010405482649e-05, "grad_norm": 2.703125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3948, "epoch": 2, "loss": 0.2287, "lr": 1.4664361004591459e-05, "grad_norm": 6.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3949, "epoch": 2, "loss": 0.1132, "lr": 1.4661711185189589e-05, "grad_norm": 4.875, "samples_per_sec": 10.24, "gpu_mem_gb": 6.82} {"step": 3950, "epoch": 2, "loss": 0.2115, "lr": 1.4659060947514798e-05, "grad_norm": 5.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3951, "epoch": 2, "loss": 0.2014, "lr": 1.4656410291804883e-05, "grad_norm": 6.53125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 3952, "epoch": 2, "loss": 0.0075, "lr": 1.465375921829767e-05, "grad_norm": 2.1875, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 3953, "epoch": 2, "loss": 0.0004, "lr": 1.4651107727231032e-05, "grad_norm": 4.75, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3954, "epoch": 2, "loss": 0.0011, "lr": 1.4648455818842866e-05, "grad_norm": 0.185546875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3955, "epoch": 2, "loss": 0.0013, "lr": 1.4645803493371123e-05, "grad_norm": 3.21875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 3956, "epoch": 2, "loss": 0.0002, "lr": 1.4643150751053781e-05, "grad_norm": 4.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 3957, "epoch": 2, "loss": 0.0356, "lr": 1.4640497592128858e-05, "grad_norm": 2.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3958, "epoch": 2, "loss": 0.0335, "lr": 1.4637844016834407e-05, "grad_norm": 3.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 3959, "epoch": 2, "loss": 0.0007, "lr": 1.4635190025408521e-05, "grad_norm": 4.0625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3960, "epoch": 2, "loss": 0.046, "lr": 1.4632535618089334e-05, "grad_norm": 4.5, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 3961, "epoch": 2, "loss": 0.1016, "lr": 1.462988079511501e-05, "grad_norm": 9.3125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 3962, "epoch": 2, "loss": 0.0178, "lr": 1.4627225556723753e-05, "grad_norm": 1.578125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 3963, "epoch": 2, "loss": 0.0394, "lr": 1.4624569903153805e-05, "grad_norm": 1.40625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 3964, "epoch": 2, "loss": 0.0008, "lr": 1.4621913834643446e-05, "grad_norm": 3.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 3965, "epoch": 2, "loss": 0.075, "lr": 1.4619257351430993e-05, "grad_norm": 3.328125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 3966, "epoch": 2, "loss": 0.0787, "lr": 1.4616600453754796e-05, "grad_norm": 3.25, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 3967, "epoch": 2, "loss": 0.0276, "lr": 1.4613943141853248e-05, "grad_norm": 4.125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.84} {"step": 3968, "epoch": 2, "loss": 0.0211, "lr": 1.461128541596478e-05, "grad_norm": 3.453125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 3969, "epoch": 2, "loss": 0.0002, "lr": 1.4608627276327851e-05, "grad_norm": 4.78125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 3970, "epoch": 2, "loss": 0.2353, "lr": 1.4605968723180968e-05, "grad_norm": 8.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 3971, "epoch": 2, "loss": 0.0406, "lr": 1.4603309756762665e-05, "grad_norm": 2.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3972, "epoch": 2, "loss": 0.0008, "lr": 1.4600650377311523e-05, "grad_norm": 5.59375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 3973, "epoch": 2, "loss": 0.0619, "lr": 1.4597990585066156e-05, "grad_norm": 2.390625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 3974, "epoch": 2, "loss": 0.1177, "lr": 1.4595330380265204e-05, "grad_norm": 8.1875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 3975, "epoch": 2, "loss": 0.0842, "lr": 1.4592669763147368e-05, "grad_norm": 1.984375, "samples_per_sec": 11.0, "gpu_mem_gb": 6.85} {"step": 3976, "epoch": 2, "loss": 0.0179, "lr": 1.4590008733951367e-05, "grad_norm": 2.015625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 3977, "epoch": 2, "loss": 0.0993, "lr": 1.4587347292915964e-05, "grad_norm": 2.453125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 3978, "epoch": 2, "loss": 0.0879, "lr": 1.458468544027995e-05, "grad_norm": 4.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3979, "epoch": 2, "loss": 0.04, "lr": 1.4582023176282168e-05, "grad_norm": 5.46875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 3980, "epoch": 2, "loss": 0.0237, "lr": 1.4579360501161489e-05, "grad_norm": 3.15625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 3981, "epoch": 2, "loss": 0.0692, "lr": 1.4576697415156818e-05, "grad_norm": 5.375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 3982, "epoch": 2, "loss": 0.0389, "lr": 1.4574033918507105e-05, "grad_norm": 4.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 3983, "epoch": 2, "loss": 0.0005, "lr": 1.4571370011451334e-05, "grad_norm": 3.46875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 3984, "epoch": 2, "loss": 0.1059, "lr": 1.4568705694228517e-05, "grad_norm": 9.1875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 3985, "epoch": 2, "loss": 0.1935, "lr": 1.456604096707772e-05, "grad_norm": 6.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 3986, "epoch": 2, "loss": 0.0006, "lr": 1.4563375830238027e-05, "grad_norm": 1.859375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 3987, "epoch": 2, "loss": 0.0004, "lr": 1.4560710283948573e-05, "grad_norm": 1.453125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 3988, "epoch": 2, "loss": 0.0012, "lr": 1.4558044328448529e-05, "grad_norm": 5.125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 3989, "epoch": 2, "loss": 0.0131, "lr": 1.4555377963977086e-05, "grad_norm": 4.15625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 3990, "epoch": 2, "loss": 0.0006, "lr": 1.4552711190773496e-05, "grad_norm": 3.8125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3991, "epoch": 2, "loss": 0.0011, "lr": 1.455004400907703e-05, "grad_norm": 4.4375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 3992, "epoch": 2, "loss": 0.2541, "lr": 1.4547376419127007e-05, "grad_norm": 8.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 3993, "epoch": 2, "loss": 0.0344, "lr": 1.4544708421162766e-05, "grad_norm": 3.703125, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 3994, "epoch": 2, "loss": 0.1788, "lr": 1.4542040015423701e-05, "grad_norm": 7.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 3995, "epoch": 2, "loss": 0.0017, "lr": 1.4539371202149237e-05, "grad_norm": 3.859375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3996, "epoch": 2, "loss": 0.0038, "lr": 1.453670198157883e-05, "grad_norm": 5.78125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 3997, "epoch": 2, "loss": 0.0172, "lr": 1.453403235395198e-05, "grad_norm": 1.9140625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 3998, "epoch": 2, "loss": 0.0038, "lr": 1.4531362319508216e-05, "grad_norm": 3.96875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 3999, "epoch": 2, "loss": 0.0069, "lr": 1.4528691878487107e-05, "grad_norm": 2.953125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4000, "epoch": 2, "loss": 0.0006, "lr": 1.4526021031128261e-05, "grad_norm": 1.5546875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.85} {"step": 4001, "epoch": 2, "loss": 0.0226, "lr": 1.4523349777671318e-05, "grad_norm": 2.96875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4002, "epoch": 2, "loss": 0.0043, "lr": 1.4520678118355962e-05, "grad_norm": 2.796875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4003, "epoch": 2, "loss": 0.1956, "lr": 1.4518006053421902e-05, "grad_norm": 7.78125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4004, "epoch": 2, "loss": 0.0582, "lr": 1.4515333583108896e-05, "grad_norm": 5.5, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4005, "epoch": 2, "loss": 0.0009, "lr": 1.4512660707656725e-05, "grad_norm": 5.96875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4006, "epoch": 2, "loss": 0.0023, "lr": 1.4509987427305216e-05, "grad_norm": 5.21875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4007, "epoch": 2, "loss": 0.0099, "lr": 1.4507313742294233e-05, "grad_norm": 5.0, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4008, "epoch": 2, "loss": 0.0152, "lr": 1.4504639652863666e-05, "grad_norm": 1.0234375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4009, "epoch": 2, "loss": 0.0594, "lr": 1.4501965159253453e-05, "grad_norm": 3.796875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4010, "epoch": 2, "loss": 0.0049, "lr": 1.4499290261703565e-05, "grad_norm": 2.46875, "samples_per_sec": 10.38, "gpu_mem_gb": 6.82} {"step": 4011, "epoch": 2, "loss": 0.0005, "lr": 1.4496614960454002e-05, "grad_norm": 2.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4012, "epoch": 2, "loss": 0.1126, "lr": 1.4493939255744814e-05, "grad_norm": 5.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4013, "epoch": 2, "loss": 0.0136, "lr": 1.4491263147816065e-05, "grad_norm": 3.46875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4014, "epoch": 2, "loss": 0.0067, "lr": 1.4488586636907887e-05, "grad_norm": 3.46875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4015, "epoch": 2, "loss": 0.0509, "lr": 1.4485909723260421e-05, "grad_norm": 4.59375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4016, "epoch": 2, "loss": 0.0003, "lr": 1.448323240711385e-05, "grad_norm": 0.3203125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4017, "epoch": 2, "loss": 0.0009, "lr": 1.4480554688708404e-05, "grad_norm": 4.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4018, "epoch": 2, "loss": 0.1145, "lr": 1.4477876568284338e-05, "grad_norm": 6.3125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4019, "epoch": 2, "loss": 0.0007, "lr": 1.4475198046081951e-05, "grad_norm": 0.314453125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4020, "epoch": 2, "loss": 0.0453, "lr": 1.4472519122341566e-05, "grad_norm": 5.5, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 4021, "epoch": 2, "loss": 0.0229, "lr": 1.4469839797303555e-05, "grad_norm": 5.4375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4022, "epoch": 2, "loss": 0.0519, "lr": 1.4467160071208327e-05, "grad_norm": 7.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4023, "epoch": 2, "loss": 0.0006, "lr": 1.4464479944296308e-05, "grad_norm": 5.875, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 4024, "epoch": 2, "loss": 0.0647, "lr": 1.4461799416807983e-05, "grad_norm": 3.265625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4025, "epoch": 2, "loss": 0.0071, "lr": 1.4459118488983856e-05, "grad_norm": 1.5234375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4026, "epoch": 2, "loss": 0.0041, "lr": 1.4456437161064481e-05, "grad_norm": 4.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4027, "epoch": 2, "loss": 0.0998, "lr": 1.4453755433290435e-05, "grad_norm": 6.03125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4028, "epoch": 2, "loss": 0.0714, "lr": 1.4451073305902336e-05, "grad_norm": 2.453125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4029, "epoch": 2, "loss": 0.241, "lr": 1.4448390779140844e-05, "grad_norm": 5.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4030, "epoch": 2, "loss": 0.0141, "lr": 1.4445707853246644e-05, "grad_norm": 3.9375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4031, "epoch": 2, "loss": 0.2922, "lr": 1.4443024528460465e-05, "grad_norm": 4.65625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 4032, "epoch": 2, "loss": 0.0593, "lr": 1.4440340805023068e-05, "grad_norm": 4.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4033, "epoch": 2, "loss": 0.0028, "lr": 1.4437656683175249e-05, "grad_norm": 3.0625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4034, "epoch": 2, "loss": 0.0147, "lr": 1.4434972163157843e-05, "grad_norm": 5.71875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4035, "epoch": 2, "loss": 0.1304, "lr": 1.443228724521172e-05, "grad_norm": 7.5625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4036, "epoch": 2, "loss": 0.0606, "lr": 1.4429601929577781e-05, "grad_norm": 3.28125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4037, "epoch": 2, "loss": 0.0016, "lr": 1.442691621649697e-05, "grad_norm": 5.9375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 4038, "epoch": 2, "loss": 0.0005, "lr": 1.4424230106210265e-05, "grad_norm": 5.15625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4039, "epoch": 2, "loss": 0.0023, "lr": 1.4421543598958674e-05, "grad_norm": 4.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4040, "epoch": 2, "loss": 0.0037, "lr": 1.4418856694983242e-05, "grad_norm": 1.75, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 4041, "epoch": 2, "loss": 0.144, "lr": 1.4416169394525056e-05, "grad_norm": 5.65625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4042, "epoch": 2, "loss": 0.0005, "lr": 1.4413481697825236e-05, "grad_norm": 1.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4043, "epoch": 2, "loss": 0.0293, "lr": 1.441079360512493e-05, "grad_norm": 3.59375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 4044, "epoch": 2, "loss": 0.0025, "lr": 1.4408105116665336e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 4045, "epoch": 2, "loss": 0.1963, "lr": 1.440541623268767e-05, "grad_norm": 4.40625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4046, "epoch": 2, "loss": 0.0045, "lr": 1.4402726953433202e-05, "grad_norm": 3.015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4047, "epoch": 2, "loss": 0.0036, "lr": 1.440003727914322e-05, "grad_norm": 3.625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 4048, "epoch": 2, "loss": 0.1566, "lr": 1.4397347210059059e-05, "grad_norm": 6.84375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4049, "epoch": 2, "loss": 0.0015, "lr": 1.4394656746422087e-05, "grad_norm": 3.671875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4050, "epoch": 2, "loss": 0.0422, "lr": 1.4391965888473705e-05, "grad_norm": 5.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4051, "epoch": 2, "loss": 0.0157, "lr": 1.4389274636455351e-05, "grad_norm": 4.1875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4052, "epoch": 2, "loss": 0.0026, "lr": 1.4386582990608497e-05, "grad_norm": 0.87109375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 4053, "epoch": 2, "loss": 0.0448, "lr": 1.4383890951174653e-05, "grad_norm": 4.0625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.85} {"step": 4054, "epoch": 2, "loss": 0.0079, "lr": 1.4381198518395363e-05, "grad_norm": 1.359375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4055, "epoch": 2, "loss": 0.0003, "lr": 1.4378505692512203e-05, "grad_norm": 2.09375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4056, "epoch": 2, "loss": 0.2075, "lr": 1.437581247376679e-05, "grad_norm": 7.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4057, "epoch": 2, "loss": 0.036, "lr": 1.4373118862400775e-05, "grad_norm": 3.921875, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4058, "epoch": 2, "loss": 0.0707, "lr": 1.437042485865584e-05, "grad_norm": 3.515625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4059, "epoch": 2, "loss": 0.0009, "lr": 1.4367730462773708e-05, "grad_norm": 0.640625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4060, "epoch": 2, "loss": 0.1801, "lr": 1.4365035674996127e-05, "grad_norm": 6.6875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4061, "epoch": 2, "loss": 0.2082, "lr": 1.4362340495564897e-05, "grad_norm": 5.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4062, "epoch": 2, "loss": 0.0001, "lr": 1.4359644924721836e-05, "grad_norm": 1.953125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4063, "epoch": 2, "loss": 0.002, "lr": 1.4356948962708809e-05, "grad_norm": 3.625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4064, "epoch": 2, "loss": 0.0519, "lr": 1.4354252609767709e-05, "grad_norm": 4.40625, "samples_per_sec": 10.3, "gpu_mem_gb": 6.83} {"step": 4065, "epoch": 2, "loss": 0.0576, "lr": 1.4351555866140467e-05, "grad_norm": 5.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4066, "epoch": 2, "loss": 0.0014, "lr": 1.4348858732069056e-05, "grad_norm": 3.796875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4067, "epoch": 2, "loss": 0.0001, "lr": 1.4346161207795464e-05, "grad_norm": 3.078125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4068, "epoch": 2, "loss": 0.0732, "lr": 1.4343463293561734e-05, "grad_norm": 5.25, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 4069, "epoch": 2, "loss": 0.0374, "lr": 1.434076498960994e-05, "grad_norm": 2.96875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4070, "epoch": 2, "loss": 0.0204, "lr": 1.4338066296182183e-05, "grad_norm": 3.34375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 4071, "epoch": 2, "loss": 0.0213, "lr": 1.4335367213520605e-05, "grad_norm": 4.15625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4072, "epoch": 2, "loss": 0.0046, "lr": 1.4332667741867378e-05, "grad_norm": 0.52734375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4073, "epoch": 2, "loss": 0.0064, "lr": 1.4329967881464721e-05, "grad_norm": 4.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4074, "epoch": 2, "loss": 0.0324, "lr": 1.4327267632554873e-05, "grad_norm": 5.5625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4075, "epoch": 2, "loss": 0.0037, "lr": 1.4324566995380118e-05, "grad_norm": 3.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4076, "epoch": 2, "loss": 0.0021, "lr": 1.4321865970182766e-05, "grad_norm": 0.796875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4077, "epoch": 2, "loss": 0.0001, "lr": 1.4319164557205173e-05, "grad_norm": 5.125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4078, "epoch": 2, "loss": 0.0035, "lr": 1.4316462756689721e-05, "grad_norm": 1.1796875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4079, "epoch": 2, "loss": 0.1427, "lr": 1.4313760568878829e-05, "grad_norm": 4.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 4080, "epoch": 2, "loss": 0.0971, "lr": 1.4311057994014953e-05, "grad_norm": 8.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 4081, "epoch": 2, "loss": 0.0978, "lr": 1.4308355032340579e-05, "grad_norm": 7.59375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 4082, "epoch": 2, "loss": 0.0074, "lr": 1.4305651684098234e-05, "grad_norm": 2.75, "samples_per_sec": 11.0, "gpu_mem_gb": 6.85} {"step": 4083, "epoch": 2, "loss": 0.1913, "lr": 1.4302947949530475e-05, "grad_norm": 6.625, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 4084, "epoch": 2, "loss": 0.0003, "lr": 1.4300243828879896e-05, "grad_norm": 2.28125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4085, "epoch": 2, "loss": 0.0004, "lr": 1.4297539322389122e-05, "grad_norm": 9.125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4086, "epoch": 2, "loss": 0.0237, "lr": 1.4294834430300822e-05, "grad_norm": 7.28125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4087, "epoch": 2, "loss": 0.163, "lr": 1.4292129152857682e-05, "grad_norm": 7.21875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4088, "epoch": 2, "loss": 0.0895, "lr": 1.4289423490302445e-05, "grad_norm": 3.1875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4089, "epoch": 2, "loss": 0.0099, "lr": 1.4286717442877871e-05, "grad_norm": 0.79296875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4090, "epoch": 2, "loss": 0.0209, "lr": 1.428401101082676e-05, "grad_norm": 6.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4091, "epoch": 2, "loss": 0.0011, "lr": 1.428130419439195e-05, "grad_norm": 3.140625, "samples_per_sec": 8.49, "gpu_mem_gb": 6.82} {"step": 4092, "epoch": 2, "loss": 0.0033, "lr": 1.427859699381631e-05, "grad_norm": 2.734375, "samples_per_sec": 11.76, "gpu_mem_gb": 6.82} {"step": 4093, "epoch": 2, "loss": 0.0003, "lr": 1.4275889409342743e-05, "grad_norm": 5.78125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4094, "epoch": 2, "loss": 0.0001, "lr": 1.4273181441214186e-05, "grad_norm": 5.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4095, "epoch": 2, "loss": 0.1457, "lr": 1.4270473089673616e-05, "grad_norm": 4.96875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4096, "epoch": 2, "loss": 0.0182, "lr": 1.4267764354964038e-05, "grad_norm": 5.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4097, "epoch": 2, "loss": 0.0, "lr": 1.4265055237328493e-05, "grad_norm": 2.609375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 4098, "epoch": 2, "loss": 0.1972, "lr": 1.4262345737010061e-05, "grad_norm": 5.65625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4099, "epoch": 2, "loss": 0.0373, "lr": 1.4259635854251845e-05, "grad_norm": 2.875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4100, "epoch": 2, "loss": 0.0041, "lr": 1.4256925589296997e-05, "grad_norm": 4.5625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.85} {"step": 4101, "epoch": 2, "loss": 0.0004, "lr": 1.4254214942388696e-05, "grad_norm": 2.6875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4102, "epoch": 2, "loss": 0.1254, "lr": 1.4251503913770149e-05, "grad_norm": 4.5625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.84} {"step": 4103, "epoch": 2, "loss": 0.1117, "lr": 1.4248792503684608e-05, "grad_norm": 3.4375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4104, "epoch": 2, "loss": 0.0022, "lr": 1.4246080712375352e-05, "grad_norm": 2.984375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 4105, "epoch": 2, "loss": 0.0042, "lr": 1.4243368540085702e-05, "grad_norm": 4.53125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4106, "epoch": 2, "loss": 0.1088, "lr": 1.4240655987059004e-05, "grad_norm": 4.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4107, "epoch": 2, "loss": 0.0012, "lr": 1.4237943053538643e-05, "grad_norm": 0.78515625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 4108, "epoch": 2, "loss": 0.0011, "lr": 1.4235229739768039e-05, "grad_norm": 4.96875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4109, "epoch": 2, "loss": 0.0031, "lr": 1.4232516045990643e-05, "grad_norm": 1.4453125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4110, "epoch": 2, "loss": 0.1051, "lr": 1.4229801972449946e-05, "grad_norm": 11.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4111, "epoch": 2, "loss": 0.0085, "lr": 1.422708751938946e-05, "grad_norm": 5.125, "samples_per_sec": 10.36, "gpu_mem_gb": 6.82} {"step": 4112, "epoch": 2, "loss": 0.1309, "lr": 1.4224372687052746e-05, "grad_norm": 7.09375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4113, "epoch": 2, "loss": 0.0098, "lr": 1.4221657475683392e-05, "grad_norm": 8.5, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4114, "epoch": 2, "loss": 0.0003, "lr": 1.4218941885525019e-05, "grad_norm": 4.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 4115, "epoch": 2, "loss": 0.0945, "lr": 1.4216225916821287e-05, "grad_norm": 3.390625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4116, "epoch": 2, "loss": 0.3609, "lr": 1.4213509569815884e-05, "grad_norm": 7.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4117, "epoch": 2, "loss": 0.0452, "lr": 1.4210792844752539e-05, "grad_norm": 6.375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4118, "epoch": 2, "loss": 0.0003, "lr": 1.4208075741875007e-05, "grad_norm": 3.484375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4119, "epoch": 2, "loss": 0.0014, "lr": 1.4205358261427076e-05, "grad_norm": 3.34375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4120, "epoch": 2, "loss": 0.0372, "lr": 1.4202640403652583e-05, "grad_norm": 5.34375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4121, "epoch": 2, "loss": 0.0845, "lr": 1.4199922168795382e-05, "grad_norm": 5.96875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4122, "epoch": 2, "loss": 0.1584, "lr": 1.4197203557099367e-05, "grad_norm": 6.375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 4123, "epoch": 2, "loss": 0.0535, "lr": 1.4194484568808466e-05, "grad_norm": 3.4375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4124, "epoch": 2, "loss": 0.0037, "lr": 1.4191765204166643e-05, "grad_norm": 2.3125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 4125, "epoch": 2, "loss": 0.2413, "lr": 1.4189045463417892e-05, "grad_norm": 5.03125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4126, "epoch": 2, "loss": 0.0909, "lr": 1.418632534680624e-05, "grad_norm": 3.953125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4127, "epoch": 2, "loss": 0.171, "lr": 1.4183604854575759e-05, "grad_norm": 5.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4128, "epoch": 2, "loss": 0.0402, "lr": 1.4180883986970536e-05, "grad_norm": 5.03125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4129, "epoch": 2, "loss": 0.0248, "lr": 1.4178162744234703e-05, "grad_norm": 3.359375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4130, "epoch": 2, "loss": 0.0026, "lr": 1.4175441126612425e-05, "grad_norm": 2.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4131, "epoch": 2, "loss": 0.0817, "lr": 1.41727191343479e-05, "grad_norm": 4.75, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83} {"step": 4132, "epoch": 2, "loss": 0.0002, "lr": 1.4169996767685363e-05, "grad_norm": 2.421875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4133, "epoch": 2, "loss": 0.1187, "lr": 1.4167274026869074e-05, "grad_norm": 3.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4134, "epoch": 2, "loss": 0.0009, "lr": 1.4164550912143331e-05, "grad_norm": 1.890625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4135, "epoch": 2, "loss": 0.0696, "lr": 1.416182742375247e-05, "grad_norm": 4.625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4136, "epoch": 2, "loss": 0.003, "lr": 1.4159103561940852e-05, "grad_norm": 10.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4137, "epoch": 2, "loss": 0.006, "lr": 1.4156379326952881e-05, "grad_norm": 1.2578125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4138, "epoch": 2, "loss": 0.433, "lr": 1.4153654719032985e-05, "grad_norm": 7.34375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 4139, "epoch": 2, "loss": 0.0075, "lr": 1.4150929738425632e-05, "grad_norm": 4.125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4140, "epoch": 2, "loss": 0.1467, "lr": 1.414820438537532e-05, "grad_norm": 4.46875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4141, "epoch": 2, "loss": 0.0134, "lr": 1.4145478660126586e-05, "grad_norm": 1.265625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4142, "epoch": 2, "loss": 0.1573, "lr": 1.4142752562923988e-05, "grad_norm": 6.78125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4143, "epoch": 2, "loss": 0.0151, "lr": 1.4140026094012136e-05, "grad_norm": 4.59375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4144, "epoch": 2, "loss": 0.0017, "lr": 1.4137299253635657e-05, "grad_norm": 4.78125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4145, "epoch": 2, "loss": 0.0283, "lr": 1.4134572042039218e-05, "grad_norm": 2.5, "samples_per_sec": 10.49, "gpu_mem_gb": 6.85} {"step": 4146, "epoch": 2, "loss": 0.0003, "lr": 1.4131844459467514e-05, "grad_norm": 3.578125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4147, "epoch": 2, "loss": 0.0014, "lr": 1.4129116506165289e-05, "grad_norm": 0.9609375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4148, "epoch": 2, "loss": 0.0005, "lr": 1.4126388182377298e-05, "grad_norm": 5.28125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4149, "epoch": 2, "loss": 0.001, "lr": 1.4123659488348346e-05, "grad_norm": 0.79296875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4150, "epoch": 2, "loss": 0.0378, "lr": 1.4120930424323267e-05, "grad_norm": 3.390625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4151, "epoch": 2, "loss": 0.095, "lr": 1.411820099054692e-05, "grad_norm": 4.71875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4152, "epoch": 2, "loss": 0.0527, "lr": 1.4115471187264212e-05, "grad_norm": 6.65625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 4153, "epoch": 2, "loss": 0.2684, "lr": 1.4112741014720069e-05, "grad_norm": 4.8125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4154, "epoch": 2, "loss": 0.0002, "lr": 1.411001047315946e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4155, "epoch": 2, "loss": 0.0173, "lr": 1.410727956282738e-05, "grad_norm": 3.09375, "samples_per_sec": 10.4, "gpu_mem_gb": 6.83} {"step": 4156, "epoch": 2, "loss": 0.0148, "lr": 1.4104548283968864e-05, "grad_norm": 6.53125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4157, "epoch": 2, "loss": 0.0077, "lr": 1.4101816636828978e-05, "grad_norm": 2.28125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 4158, "epoch": 2, "loss": 0.1642, "lr": 1.409908462165281e-05, "grad_norm": 5.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4159, "epoch": 2, "loss": 0.0002, "lr": 1.4096352238685502e-05, "grad_norm": 2.109375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 4160, "epoch": 2, "loss": 0.085, "lr": 1.409361948817221e-05, "grad_norm": 3.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4161, "epoch": 2, "loss": 0.0634, "lr": 1.4090886370358137e-05, "grad_norm": 6.15625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4162, "epoch": 2, "loss": 0.0056, "lr": 1.4088152885488504e-05, "grad_norm": 6.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4163, "epoch": 2, "loss": 0.0006, "lr": 1.4085419033808579e-05, "grad_norm": 4.09375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 4164, "epoch": 2, "loss": 0.0001, "lr": 1.408268481556366e-05, "grad_norm": 2.203125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4165, "epoch": 2, "loss": 0.1139, "lr": 1.4079950230999069e-05, "grad_norm": 3.765625, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 4166, "epoch": 2, "loss": 0.0044, "lr": 1.407721528036017e-05, "grad_norm": 5.53125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4167, "epoch": 2, "loss": 0.0165, "lr": 1.4074479963892358e-05, "grad_norm": 1.859375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4168, "epoch": 2, "loss": 0.0273, "lr": 1.4071744281841058e-05, "grad_norm": 4.75, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 4169, "epoch": 2, "loss": 0.0031, "lr": 1.4069008234451729e-05, "grad_norm": 0.310546875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4170, "epoch": 2, "loss": 0.0603, "lr": 1.4066271821969866e-05, "grad_norm": 4.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4171, "epoch": 2, "loss": 0.0003, "lr": 1.4063535044640995e-05, "grad_norm": 5.875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4172, "epoch": 2, "loss": 0.1351, "lr": 1.406079790271067e-05, "grad_norm": 4.25, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 4173, "epoch": 2, "loss": 0.0983, "lr": 1.405806039642448e-05, "grad_norm": 5.28125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 4174, "epoch": 2, "loss": 0.0265, "lr": 1.4055322526028054e-05, "grad_norm": 3.75, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4175, "epoch": 2, "loss": 0.0039, "lr": 1.4052584291767046e-05, "grad_norm": 0.34765625, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4176, "epoch": 2, "loss": 0.0164, "lr": 1.4049845693887146e-05, "grad_norm": 5.15625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 4177, "epoch": 2, "loss": 0.015, "lr": 1.4047106732634068e-05, "grad_norm": 3.671875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4178, "epoch": 2, "loss": 0.003, "lr": 1.4044367408253574e-05, "grad_norm": 1.2734375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4179, "epoch": 2, "loss": 0.1328, "lr": 1.4041627720991448e-05, "grad_norm": 5.375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.83} {"step": 4180, "epoch": 2, "loss": 0.1002, "lr": 1.4038887671093506e-05, "grad_norm": 8.25, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4181, "epoch": 2, "loss": 0.0005, "lr": 1.4036147258805604e-05, "grad_norm": 3.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4182, "epoch": 2, "loss": 0.0705, "lr": 1.4033406484373624e-05, "grad_norm": 6.0, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4183, "epoch": 2, "loss": 0.0033, "lr": 1.4030665348043482e-05, "grad_norm": 4.5625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4184, "epoch": 2, "loss": 0.0149, "lr": 1.4027923850061127e-05, "grad_norm": 2.328125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4185, "epoch": 2, "loss": 0.1001, "lr": 1.4025181990672541e-05, "grad_norm": 6.09375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4186, "epoch": 2, "loss": 0.1521, "lr": 1.4022439770123736e-05, "grad_norm": 9.8125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 4187, "epoch": 2, "loss": 0.1179, "lr": 1.4019697188660764e-05, "grad_norm": 4.3125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.84} {"step": 4188, "epoch": 2, "loss": 0.0286, "lr": 1.4016954246529697e-05, "grad_norm": 2.78125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4189, "epoch": 2, "loss": 0.0028, "lr": 1.4014210943976647e-05, "grad_norm": 0.91796875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 4190, "epoch": 2, "loss": 0.0088, "lr": 1.4011467281247762e-05, "grad_norm": 5.78125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4191, "epoch": 2, "loss": 0.0499, "lr": 1.4008723258589214e-05, "grad_norm": 3.421875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4192, "epoch": 2, "loss": 0.0003, "lr": 1.4005978876247208e-05, "grad_norm": 4.75, "samples_per_sec": 10.23, "gpu_mem_gb": 6.83} {"step": 4193, "epoch": 2, "loss": 0.0046, "lr": 1.4003234134467991e-05, "grad_norm": 1.3984375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4194, "epoch": 2, "loss": 0.0907, "lr": 1.4000489033497833e-05, "grad_norm": 8.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4195, "epoch": 2, "loss": 0.005, "lr": 1.3997743573583035e-05, "grad_norm": 3.40625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 4196, "epoch": 2, "loss": 0.0032, "lr": 1.3994997754969942e-05, "grad_norm": 4.875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4197, "epoch": 2, "loss": 0.1148, "lr": 1.3992251577904912e-05, "grad_norm": 6.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4198, "epoch": 2, "loss": 0.0001, "lr": 1.3989505042634356e-05, "grad_norm": 6.46875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.85} {"step": 4199, "epoch": 2, "loss": 0.001, "lr": 1.3986758149404702e-05, "grad_norm": 4.4375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 4200, "epoch": 2, "loss": 0.001, "lr": 1.3984010898462417e-05, "grad_norm": 4.25, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4201, "epoch": 2, "loss": 0.0002, "lr": 1.3981263290053998e-05, "grad_norm": 1.1796875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4202, "epoch": 2, "loss": 0.2013, "lr": 1.3978515324425976e-05, "grad_norm": 9.125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4203, "epoch": 2, "loss": 0.0042, "lr": 1.3975767001824913e-05, "grad_norm": 4.09375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 4204, "epoch": 2, "loss": 0.0452, "lr": 1.39730183224974e-05, "grad_norm": 4.59375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4205, "epoch": 2, "loss": 0.0036, "lr": 1.3970269286690066e-05, "grad_norm": 4.40625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 4206, "epoch": 2, "loss": 0.0002, "lr": 1.396751989464957e-05, "grad_norm": 3.671875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4207, "epoch": 2, "loss": 0.0008, "lr": 1.3964770146622594e-05, "grad_norm": 2.640625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4208, "epoch": 2, "loss": 0.0029, "lr": 1.396202004285587e-05, "grad_norm": 5.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4209, "epoch": 2, "loss": 0.0003, "lr": 1.3959269583596145e-05, "grad_norm": 0.2578125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4210, "epoch": 2, "loss": 0.011, "lr": 1.3956518769090204e-05, "grad_norm": 7.4375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 4211, "epoch": 2, "loss": 0.0014, "lr": 1.3953767599584869e-05, "grad_norm": 4.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4212, "epoch": 2, "loss": 0.0042, "lr": 1.395101607532698e-05, "grad_norm": 4.40625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4213, "epoch": 2, "loss": 0.2218, "lr": 1.3948264196563433e-05, "grad_norm": 5.65625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4214, "epoch": 2, "loss": 0.0292, "lr": 1.3945511963541127e-05, "grad_norm": 1.7421875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4215, "epoch": 2, "loss": 0.0048, "lr": 1.3942759376507013e-05, "grad_norm": 0.66796875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4216, "epoch": 2, "loss": 0.0007, "lr": 1.3940006435708066e-05, "grad_norm": 4.28125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 4217, "epoch": 2, "loss": 0.0242, "lr": 1.3937253141391296e-05, "grad_norm": 2.421875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4218, "epoch": 2, "loss": 0.0242, "lr": 1.3934499493803741e-05, "grad_norm": 1.96875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4219, "epoch": 2, "loss": 0.1693, "lr": 1.3931745493192473e-05, "grad_norm": 5.375, "samples_per_sec": 10.24, "gpu_mem_gb": 6.83} {"step": 4220, "epoch": 2, "loss": 0.0834, "lr": 1.3928991139804592e-05, "grad_norm": 4.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4221, "epoch": 2, "loss": 0.0879, "lr": 1.392623643388724e-05, "grad_norm": 4.21875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4222, "epoch": 2, "loss": 0.1444, "lr": 1.3923481375687577e-05, "grad_norm": 4.21875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4223, "epoch": 2, "loss": 0.0878, "lr": 1.3920725965452806e-05, "grad_norm": 4.9375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.84} {"step": 4224, "epoch": 2, "loss": 0.0196, "lr": 1.3917970203430154e-05, "grad_norm": 6.34375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4225, "epoch": 2, "loss": 0.0015, "lr": 1.3915214089866882e-05, "grad_norm": 5.28125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4226, "epoch": 2, "loss": 0.0013, "lr": 1.3912457625010282e-05, "grad_norm": 6.0625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 4227, "epoch": 2, "loss": 0.1309, "lr": 1.3909700809107682e-05, "grad_norm": 8.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4228, "epoch": 2, "loss": 0.011, "lr": 1.3906943642406435e-05, "grad_norm": 4.71875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4229, "epoch": 2, "loss": 0.1456, "lr": 1.3904186125153927e-05, "grad_norm": 5.84375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4230, "epoch": 2, "loss": 0.0061, "lr": 1.390142825759758e-05, "grad_norm": 0.921875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4231, "epoch": 2, "loss": 0.0591, "lr": 1.3898670039984845e-05, "grad_norm": 4.875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4232, "epoch": 2, "loss": 0.0019, "lr": 1.3895911472563195e-05, "grad_norm": 3.09375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4233, "epoch": 2, "loss": 0.0001, "lr": 1.3893152555580157e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 4234, "epoch": 2, "loss": 0.0005, "lr": 1.3890393289283262e-05, "grad_norm": 0.16796875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4235, "epoch": 2, "loss": 0.0165, "lr": 1.3887633673920095e-05, "grad_norm": 2.296875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4236, "epoch": 2, "loss": 0.0658, "lr": 1.3884873709738259e-05, "grad_norm": 6.09375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4237, "epoch": 2, "loss": 0.001, "lr": 1.388211339698539e-05, "grad_norm": 1.765625, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 4238, "epoch": 2, "loss": 0.018, "lr": 1.3879352735909163e-05, "grad_norm": 1.9375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4239, "epoch": 2, "loss": 0.0001, "lr": 1.3876591726757276e-05, "grad_norm": 4.15625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4240, "epoch": 2, "loss": 0.0013, "lr": 1.387383036977746e-05, "grad_norm": 4.0, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 4241, "epoch": 2, "loss": 0.0572, "lr": 1.3871068665217482e-05, "grad_norm": 2.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4242, "epoch": 2, "loss": 0.4051, "lr": 1.3868306613325131e-05, "grad_norm": 10.875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4243, "epoch": 2, "loss": 0.128, "lr": 1.3865544214348238e-05, "grad_norm": 3.109375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4244, "epoch": 2, "loss": 0.0145, "lr": 1.3862781468534654e-05, "grad_norm": 5.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4245, "epoch": 2, "loss": 0.0388, "lr": 1.3860018376132275e-05, "grad_norm": 3.078125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4246, "epoch": 2, "loss": 0.0124, "lr": 1.3857254937389012e-05, "grad_norm": 8.0625, "samples_per_sec": 10.41, "gpu_mem_gb": 6.85} {"step": 4247, "epoch": 2, "loss": 0.0228, "lr": 1.3854491152552822e-05, "grad_norm": 1.3125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4248, "epoch": 2, "loss": 0.1563, "lr": 1.3851727021871679e-05, "grad_norm": 5.8125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.84} {"step": 4249, "epoch": 2, "loss": 0.0002, "lr": 1.3848962545593598e-05, "grad_norm": 3.875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4250, "epoch": 2, "loss": 0.002, "lr": 1.3846197723966626e-05, "grad_norm": 5.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4251, "epoch": 2, "loss": 0.056, "lr": 1.3843432557238832e-05, "grad_norm": 6.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4252, "epoch": 2, "loss": 0.0003, "lr": 1.3840667045658323e-05, "grad_norm": 5.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4253, "epoch": 2, "loss": 0.036, "lr": 1.3837901189473235e-05, "grad_norm": 6.0, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 4254, "epoch": 2, "loss": 0.001, "lr": 1.3835134988931736e-05, "grad_norm": 4.96875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 4255, "epoch": 2, "loss": 0.0005, "lr": 1.3832368444282022e-05, "grad_norm": 5.78125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4256, "epoch": 2, "loss": 0.1353, "lr": 1.3829601555772323e-05, "grad_norm": 5.5, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4257, "epoch": 2, "loss": 0.0059, "lr": 1.3826834323650899e-05, "grad_norm": 4.46875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4258, "epoch": 2, "loss": 0.0005, "lr": 1.3824066748166041e-05, "grad_norm": 4.21875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4259, "epoch": 2, "loss": 0.0001, "lr": 1.3821298829566063e-05, "grad_norm": 4.5625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4260, "epoch": 2, "loss": 0.0576, "lr": 1.3818530568099328e-05, "grad_norm": 3.078125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 4261, "epoch": 2, "loss": 0.2667, "lr": 1.3815761964014214e-05, "grad_norm": 5.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4262, "epoch": 2, "loss": 0.1179, "lr": 1.3812993017559132e-05, "grad_norm": 5.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4263, "epoch": 2, "loss": 0.0029, "lr": 1.381022372898253e-05, "grad_norm": 0.416015625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4264, "epoch": 2, "loss": 0.2284, "lr": 1.380745409853288e-05, "grad_norm": 6.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4265, "epoch": 2, "loss": 0.0002, "lr": 1.3804684126458692e-05, "grad_norm": 4.78125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 4266, "epoch": 2, "loss": 0.0721, "lr": 1.3801913813008496e-05, "grad_norm": 5.90625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.84} {"step": 4267, "epoch": 2, "loss": 0.0001, "lr": 1.3799143158430863e-05, "grad_norm": 5.1875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 4268, "epoch": 2, "loss": 0.0006, "lr": 1.3796372162974395e-05, "grad_norm": 8.4375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4269, "epoch": 2, "loss": 0.0551, "lr": 1.3793600826887708e-05, "grad_norm": 8.625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4270, "epoch": 2, "loss": 0.0256, "lr": 1.3790829150419471e-05, "grad_norm": 2.15625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4271, "epoch": 2, "loss": 0.0142, "lr": 1.3788057133818365e-05, "grad_norm": 3.5, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4272, "epoch": 2, "loss": 0.0029, "lr": 1.378528477733312e-05, "grad_norm": 0.4921875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4273, "epoch": 2, "loss": 0.0768, "lr": 1.3782512081212476e-05, "grad_norm": 5.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4274, "epoch": 2, "loss": 0.0379, "lr": 1.3779739045705223e-05, "grad_norm": 5.53125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4275, "epoch": 2, "loss": 0.0671, "lr": 1.3776965671060165e-05, "grad_norm": 5.6875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.84} {"step": 4276, "epoch": 2, "loss": 0.2369, "lr": 1.3774191957526144e-05, "grad_norm": 8.4375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4277, "epoch": 2, "loss": 0.0195, "lr": 1.3771417905352038e-05, "grad_norm": 1.5234375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4278, "epoch": 2, "loss": 0.0924, "lr": 1.3768643514786744e-05, "grad_norm": 4.71875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4279, "epoch": 2, "loss": 0.1088, "lr": 1.3765868786079195e-05, "grad_norm": 5.34375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 4280, "epoch": 2, "loss": 0.0905, "lr": 1.3763093719478357e-05, "grad_norm": 2.8125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 4281, "epoch": 2, "loss": 0.0298, "lr": 1.3760318315233218e-05, "grad_norm": 7.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4282, "epoch": 2, "loss": 0.001, "lr": 1.375754257359281e-05, "grad_norm": 2.71875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4283, "epoch": 2, "loss": 0.1187, "lr": 1.3754766494806178e-05, "grad_norm": 7.03125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 4284, "epoch": 2, "loss": 0.0446, "lr": 1.3751990079122412e-05, "grad_norm": 4.875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4285, "epoch": 2, "loss": 0.0012, "lr": 1.3749213326790626e-05, "grad_norm": 6.0625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4286, "epoch": 2, "loss": 0.0549, "lr": 1.3746436238059961e-05, "grad_norm": 2.234375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 4287, "epoch": 2, "loss": 0.0053, "lr": 1.3743658813179594e-05, "grad_norm": 4.90625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 4288, "epoch": 2, "loss": 0.0064, "lr": 1.3740881052398731e-05, "grad_norm": 2.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4289, "epoch": 2, "loss": 0.1388, "lr": 1.3738102955966607e-05, "grad_norm": 3.453125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4290, "epoch": 2, "loss": 0.0612, "lr": 1.3735324524132484e-05, "grad_norm": 4.8125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4291, "epoch": 2, "loss": 0.1105, "lr": 1.3732545757145663e-05, "grad_norm": 3.515625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4292, "epoch": 2, "loss": 0.0035, "lr": 1.3729766655255468e-05, "grad_norm": 6.65625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4293, "epoch": 2, "loss": 0.0377, "lr": 1.3726987218711248e-05, "grad_norm": 5.03125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4294, "epoch": 2, "loss": 0.0214, "lr": 1.3724207447762394e-05, "grad_norm": 4.25, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4295, "epoch": 2, "loss": 0.0027, "lr": 1.3721427342658322e-05, "grad_norm": 5.0, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4296, "epoch": 2, "loss": 0.013, "lr": 1.3718646903648476e-05, "grad_norm": 5.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4297, "epoch": 2, "loss": 0.0027, "lr": 1.3715866130982331e-05, "grad_norm": 3.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4298, "epoch": 2, "loss": 0.0109, "lr": 1.3713085024909392e-05, "grad_norm": 0.9375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4299, "epoch": 2, "loss": 0.1532, "lr": 1.3710303585679195e-05, "grad_norm": 4.28125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4300, "epoch": 2, "loss": 0.0728, "lr": 1.3707521813541305e-05, "grad_norm": 7.90625, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83} {"step": 4301, "epoch": 2, "loss": 0.0016, "lr": 1.3704739708745321e-05, "grad_norm": 8.0625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4302, "epoch": 2, "loss": 0.0005, "lr": 1.3701957271540863e-05, "grad_norm": 2.125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4303, "epoch": 2, "loss": 0.0047, "lr": 1.369917450217758e-05, "grad_norm": 3.578125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 4304, "epoch": 2, "loss": 0.0138, "lr": 1.3696391400905172e-05, "grad_norm": 6.40625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 4305, "epoch": 2, "loss": 0.0214, "lr": 1.369360796797334e-05, "grad_norm": 7.71875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4306, "epoch": 2, "loss": 0.0978, "lr": 1.3690824203631833e-05, "grad_norm": 3.65625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4307, "epoch": 2, "loss": 0.0157, "lr": 1.368804010813043e-05, "grad_norm": 1.796875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4308, "epoch": 2, "loss": 0.0088, "lr": 1.3685255681718922e-05, "grad_norm": 3.515625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4309, "epoch": 2, "loss": 0.0002, "lr": 1.3682470924647153e-05, "grad_norm": 5.34375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4310, "epoch": 2, "loss": 0.138, "lr": 1.367968583716498e-05, "grad_norm": 6.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4311, "epoch": 2, "loss": 0.0088, "lr": 1.36769004195223e-05, "grad_norm": 3.671875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4312, "epoch": 2, "loss": 0.1806, "lr": 1.3674114671969032e-05, "grad_norm": 3.890625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 4313, "epoch": 2, "loss": 0.0001, "lr": 1.3671328594755125e-05, "grad_norm": 4.625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4314, "epoch": 2, "loss": 0.0037, "lr": 1.3668542188130567e-05, "grad_norm": 3.5, "samples_per_sec": 10.4, "gpu_mem_gb": 6.85} {"step": 4315, "epoch": 2, "loss": 0.0018, "lr": 1.3665755452345364e-05, "grad_norm": 0.7109375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4316, "epoch": 2, "loss": 0.1532, "lr": 1.366296838764956e-05, "grad_norm": 3.59375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4317, "epoch": 2, "loss": 0.0015, "lr": 1.3660180994293222e-05, "grad_norm": 3.703125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4318, "epoch": 2, "loss": 0.0923, "lr": 1.3657393272526447e-05, "grad_norm": 2.765625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4319, "epoch": 2, "loss": 0.0246, "lr": 1.3654605222599373e-05, "grad_norm": 3.78125, "samples_per_sec": 8.97, "gpu_mem_gb": 6.83} {"step": 4320, "epoch": 2, "loss": 0.1314, "lr": 1.3651816844762148e-05, "grad_norm": 4.78125, "samples_per_sec": 11.77, "gpu_mem_gb": 6.83} {"step": 4321, "epoch": 2, "loss": 0.0103, "lr": 1.3649028139264969e-05, "grad_norm": 1.7421875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 4322, "epoch": 2, "loss": 0.0016, "lr": 1.3646239106358041e-05, "grad_norm": 1.953125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4323, "epoch": 2, "loss": 0.0551, "lr": 1.3643449746291624e-05, "grad_norm": 4.40625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4324, "epoch": 2, "loss": 0.0054, "lr": 1.3640660059315986e-05, "grad_norm": 2.9375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4325, "epoch": 2, "loss": 0.1194, "lr": 1.3637870045681432e-05, "grad_norm": 6.59375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4326, "epoch": 2, "loss": 0.0006, "lr": 1.3635079705638298e-05, "grad_norm": 4.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4327, "epoch": 2, "loss": 0.0011, "lr": 1.3632289039436953e-05, "grad_norm": 1.53125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 4328, "epoch": 2, "loss": 0.0043, "lr": 1.362949804732778e-05, "grad_norm": 4.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4329, "epoch": 2, "loss": 0.0885, "lr": 1.3626706729561211e-05, "grad_norm": 5.96875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4330, "epoch": 2, "loss": 0.001, "lr": 1.3623915086387687e-05, "grad_norm": 3.984375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4331, "epoch": 2, "loss": 0.0348, "lr": 1.36211231180577e-05, "grad_norm": 5.125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4332, "epoch": 2, "loss": 0.0033, "lr": 1.3618330824821753e-05, "grad_norm": 3.53125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4333, "epoch": 2, "loss": 0.004, "lr": 1.3615538206930387e-05, "grad_norm": 5.34375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4334, "epoch": 2, "loss": 0.0005, "lr": 1.3612745264634171e-05, "grad_norm": 4.03125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 4335, "epoch": 2, "loss": 0.0029, "lr": 1.36099519981837e-05, "grad_norm": 4.28125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4336, "epoch": 2, "loss": 0.0022, "lr": 1.3607158407829602e-05, "grad_norm": 1.40625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4337, "epoch": 2, "loss": 0.0039, "lr": 1.3604364493822529e-05, "grad_norm": 4.3125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 4338, "epoch": 2, "loss": 0.0006, "lr": 1.360157025641317e-05, "grad_norm": 5.59375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4339, "epoch": 2, "loss": 0.0806, "lr": 1.3598775695852238e-05, "grad_norm": 5.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4340, "epoch": 2, "loss": 0.0003, "lr": 1.3595980812390472e-05, "grad_norm": 8.0, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 4341, "epoch": 2, "loss": 0.0965, "lr": 1.359318560627865e-05, "grad_norm": 8.8125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.85} {"step": 4342, "epoch": 2, "loss": 0.1475, "lr": 1.3590390077767564e-05, "grad_norm": 6.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4343, "epoch": 2, "loss": 0.0042, "lr": 1.358759422710805e-05, "grad_norm": 5.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4344, "epoch": 2, "loss": 0.0032, "lr": 1.3584798054550963e-05, "grad_norm": 5.84375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 4345, "epoch": 2, "loss": 0.0012, "lr": 1.3582001560347191e-05, "grad_norm": 5.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4346, "epoch": 2, "loss": 0.2367, "lr": 1.3579204744747647e-05, "grad_norm": 6.46875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.84} {"step": 4347, "epoch": 2, "loss": 0.0016, "lr": 1.3576407608003282e-05, "grad_norm": 3.859375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 4348, "epoch": 2, "loss": 0.0118, "lr": 1.3573610150365064e-05, "grad_norm": 1.359375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4349, "epoch": 2, "loss": 0.0008, "lr": 1.3570812372084e-05, "grad_norm": 0.279296875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 4350, "epoch": 2, "loss": 0.1422, "lr": 1.3568014273411117e-05, "grad_norm": 6.59375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4351, "epoch": 2, "loss": 0.0371, "lr": 1.3565215854597478e-05, "grad_norm": 2.734375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4352, "epoch": 2, "loss": 0.0968, "lr": 1.356241711589417e-05, "grad_norm": 5.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4353, "epoch": 2, "loss": 0.002, "lr": 1.355961805755231e-05, "grad_norm": 4.6875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4354, "epoch": 2, "loss": 0.0121, "lr": 1.3556818679823048e-05, "grad_norm": 4.0625, "samples_per_sec": 10.38, "gpu_mem_gb": 6.83} {"step": 4355, "epoch": 2, "loss": 0.0056, "lr": 1.3554018982957551e-05, "grad_norm": 5.46875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4356, "epoch": 2, "loss": 0.0018, "lr": 1.3551218967207032e-05, "grad_norm": 3.3125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4357, "epoch": 2, "loss": 0.0462, "lr": 1.354841863282271e-05, "grad_norm": 4.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 4358, "epoch": 2, "loss": 0.0422, "lr": 1.3545617980055861e-05, "grad_norm": 3.328125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4359, "epoch": 2, "loss": 0.0559, "lr": 1.3542817009157763e-05, "grad_norm": 5.0625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4360, "epoch": 2, "loss": 0.0016, "lr": 1.3540015720379738e-05, "grad_norm": 2.75, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4361, "epoch": 2, "loss": 0.0027, "lr": 1.353721411397313e-05, "grad_norm": 2.625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 4362, "epoch": 2, "loss": 0.0401, "lr": 1.3534412190189317e-05, "grad_norm": 2.796875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4363, "epoch": 2, "loss": 0.2318, "lr": 1.3531609949279698e-05, "grad_norm": 5.53125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4364, "epoch": 2, "loss": 0.1758, "lr": 1.3528807391495708e-05, "grad_norm": 17.125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4365, "epoch": 2, "loss": 0.0681, "lr": 1.3526004517088805e-05, "grad_norm": 7.21875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4366, "epoch": 2, "loss": 0.0197, "lr": 1.352320132631048e-05, "grad_norm": 4.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4367, "epoch": 2, "loss": 0.0021, "lr": 1.3520397819412245e-05, "grad_norm": 4.8125, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 4368, "epoch": 2, "loss": 0.054, "lr": 1.351759399664565e-05, "grad_norm": 6.0625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4369, "epoch": 2, "loss": 0.0074, "lr": 1.3514789858262264e-05, "grad_norm": 2.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4370, "epoch": 2, "loss": 0.1836, "lr": 1.3511985404513697e-05, "grad_norm": 4.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4371, "epoch": 2, "loss": 0.0417, "lr": 1.350918063565157e-05, "grad_norm": 2.484375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 4372, "epoch": 2, "loss": 0.0115, "lr": 1.3506375551927546e-05, "grad_norm": 5.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4373, "epoch": 2, "loss": 0.1245, "lr": 1.3503570153593311e-05, "grad_norm": 9.0625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.84} {"step": 4374, "epoch": 2, "loss": 0.0025, "lr": 1.3500764440900577e-05, "grad_norm": 4.4375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 4375, "epoch": 2, "loss": 0.0012, "lr": 1.3497958414101093e-05, "grad_norm": 2.703125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 4376, "epoch": 2, "loss": 0.0006, "lr": 1.3495152073446624e-05, "grad_norm": 3.890625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4377, "epoch": 2, "loss": 0.0236, "lr": 1.3492345419188975e-05, "grad_norm": 1.9140625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4378, "epoch": 2, "loss": 0.0002, "lr": 1.348953845157997e-05, "grad_norm": 0.8984375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4379, "epoch": 2, "loss": 0.0068, "lr": 1.3486731170871466e-05, "grad_norm": 4.34375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4380, "epoch": 2, "loss": 0.1295, "lr": 1.3483923577315347e-05, "grad_norm": 4.96875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4381, "epoch": 2, "loss": 0.0345, "lr": 1.3481115671163525e-05, "grad_norm": 3.359375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.84} {"step": 4382, "epoch": 2, "loss": 0.1015, "lr": 1.3478307452667938e-05, "grad_norm": 6.125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 4383, "epoch": 2, "loss": 0.0003, "lr": 1.3475498922080557e-05, "grad_norm": 3.15625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4384, "epoch": 2, "loss": 0.0506, "lr": 1.3472690079653374e-05, "grad_norm": 5.15625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4385, "epoch": 2, "loss": 0.0368, "lr": 1.3469880925638414e-05, "grad_norm": 6.28125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4386, "epoch": 2, "loss": 0.0228, "lr": 1.3467071460287733e-05, "grad_norm": 8.4375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4387, "epoch": 2, "loss": 0.0012, "lr": 1.3464261683853408e-05, "grad_norm": 4.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4388, "epoch": 2, "loss": 0.0003, "lr": 1.3461451596587547e-05, "grad_norm": 5.90625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 4389, "epoch": 2, "loss": 0.2224, "lr": 1.3458641198742282e-05, "grad_norm": 6.84375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4390, "epoch": 2, "loss": 0.0034, "lr": 1.3455830490569782e-05, "grad_norm": 4.875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4391, "epoch": 2, "loss": 0.0063, "lr": 1.3453019472322234e-05, "grad_norm": 9.6875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4392, "epoch": 2, "loss": 0.0, "lr": 1.3450208144251861e-05, "grad_norm": 14.6875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4393, "epoch": 2, "loss": 0.0096, "lr": 1.3447396506610907e-05, "grad_norm": 3.890625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4394, "epoch": 2, "loss": 0.0432, "lr": 1.3444584559651647e-05, "grad_norm": 7.4375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4395, "epoch": 2, "loss": 0.0861, "lr": 1.3441772303626387e-05, "grad_norm": 5.15625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4396, "epoch": 2, "loss": 0.0011, "lr": 1.3438959738787452e-05, "grad_norm": 3.859375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4397, "epoch": 2, "loss": 0.0033, "lr": 1.3436146865387203e-05, "grad_norm": 0.58203125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 4398, "epoch": 2, "loss": 0.0008, "lr": 1.3433333683678026e-05, "grad_norm": 6.9375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4399, "epoch": 2, "loss": 0.1457, "lr": 1.3430520193912333e-05, "grad_norm": 8.8125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4400, "epoch": 2, "loss": 0.0009, "lr": 1.3427706396342564e-05, "grad_norm": 3.53125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4401, "epoch": 2, "loss": 0.0002, "lr": 1.3424892291221188e-05, "grad_norm": 2.6875, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 4402, "epoch": 2, "loss": 0.0001, "lr": 1.3422077878800706e-05, "grad_norm": 7.25, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4403, "epoch": 2, "loss": 0.1338, "lr": 1.3419263159333634e-05, "grad_norm": 5.21875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4404, "epoch": 2, "loss": 0.0002, "lr": 1.3416448133072525e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.99, "gpu_mem_gb": 6.85} {"step": 4405, "epoch": 2, "loss": 0.1491, "lr": 1.3413632800269963e-05, "grad_norm": 4.40625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4406, "epoch": 2, "loss": 0.0003, "lr": 1.3410817161178549e-05, "grad_norm": 1.5703125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 4407, "epoch": 2, "loss": 0.142, "lr": 1.340800121605092e-05, "grad_norm": 6.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4408, "epoch": 2, "loss": 0.0007, "lr": 1.3405184965139735e-05, "grad_norm": 4.46875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4409, "epoch": 2, "loss": 0.0015, "lr": 1.3402368408697681e-05, "grad_norm": 0.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4410, "epoch": 2, "loss": 0.0025, "lr": 1.3399551546977481e-05, "grad_norm": 3.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4411, "epoch": 2, "loss": 0.0245, "lr": 1.339673438023187e-05, "grad_norm": 1.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4412, "epoch": 2, "loss": 0.0314, "lr": 1.3393916908713622e-05, "grad_norm": 3.140625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4413, "epoch": 2, "loss": 0.0324, "lr": 1.3391099132675538e-05, "grad_norm": 4.0, "samples_per_sec": 10.71, "gpu_mem_gb": 6.84} {"step": 4414, "epoch": 2, "loss": 0.0032, "lr": 1.3388281052370443e-05, "grad_norm": 4.5, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4415, "epoch": 2, "loss": 0.001, "lr": 1.3385462668051186e-05, "grad_norm": 1.15625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4416, "epoch": 2, "loss": 0.0001, "lr": 1.3382643979970649e-05, "grad_norm": 7.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4417, "epoch": 2, "loss": 0.1609, "lr": 1.337982498838174e-05, "grad_norm": 6.875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4418, "epoch": 2, "loss": 0.07, "lr": 1.3377005693537393e-05, "grad_norm": 3.640625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4419, "epoch": 2, "loss": 0.0171, "lr": 1.337418609569057e-05, "grad_norm": 4.53125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 4420, "epoch": 2, "loss": 0.0202, "lr": 1.337136619509426e-05, "grad_norm": 3.921875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4421, "epoch": 2, "loss": 0.3997, "lr": 1.3368545992001476e-05, "grad_norm": 9.75, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4422, "epoch": 2, "loss": 0.015, "lr": 1.3365725486665268e-05, "grad_norm": 4.71875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4423, "epoch": 2, "loss": 0.0004, "lr": 1.3362904679338702e-05, "grad_norm": 4.8125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 4424, "epoch": 2, "loss": 0.1114, "lr": 1.3360083570274874e-05, "grad_norm": 6.78125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4425, "epoch": 2, "loss": 0.0028, "lr": 1.3357262159726911e-05, "grad_norm": 5.0625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4426, "epoch": 2, "loss": 0.1485, "lr": 1.3354440447947965e-05, "grad_norm": 4.4375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4427, "epoch": 2, "loss": 0.0602, "lr": 1.3351618435191215e-05, "grad_norm": 6.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4428, "epoch": 2, "loss": 0.0781, "lr": 1.3348796121709862e-05, "grad_norm": 3.015625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 4429, "epoch": 2, "loss": 0.0711, "lr": 1.3345973507757144e-05, "grad_norm": 5.84375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.85} {"step": 4430, "epoch": 2, "loss": 0.0489, "lr": 1.3343150593586318e-05, "grad_norm": 2.84375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4431, "epoch": 2, "loss": 0.028, "lr": 1.3340327379450674e-05, "grad_norm": 4.0625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4432, "epoch": 2, "loss": 0.0042, "lr": 1.333750386560352e-05, "grad_norm": 0.5390625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4433, "epoch": 2, "loss": 0.001, "lr": 1.3334680052298196e-05, "grad_norm": 1.2578125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 4434, "epoch": 2, "loss": 0.0042, "lr": 1.3331855939788078e-05, "grad_norm": 4.875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4435, "epoch": 2, "loss": 0.0002, "lr": 1.332903152832655e-05, "grad_norm": 5.34375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 4436, "epoch": 2, "loss": 0.0028, "lr": 1.3326206818167038e-05, "grad_norm": 4.875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 4437, "epoch": 2, "loss": 0.0223, "lr": 1.332338180956299e-05, "grad_norm": 2.28125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4438, "epoch": 2, "loss": 0.1153, "lr": 1.3320556502767877e-05, "grad_norm": 6.3125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4439, "epoch": 2, "loss": 0.0702, "lr": 1.3317730898035201e-05, "grad_norm": 3.0625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4440, "epoch": 2, "loss": 0.0005, "lr": 1.3314904995618494e-05, "grad_norm": 5.59375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 4441, "epoch": 2, "loss": 0.0039, "lr": 1.3312078795771307e-05, "grad_norm": 3.0, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4442, "epoch": 2, "loss": 0.1632, "lr": 1.3309252298747221e-05, "grad_norm": 4.3125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4443, "epoch": 2, "loss": 0.0013, "lr": 1.3306425504799844e-05, "grad_norm": 1.2578125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 4444, "epoch": 2, "loss": 0.0031, "lr": 1.3303598414182815e-05, "grad_norm": 2.984375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4445, "epoch": 2, "loss": 0.0019, "lr": 1.3300771027149788e-05, "grad_norm": 5.3125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4446, "epoch": 2, "loss": 0.0757, "lr": 1.3297943343954458e-05, "grad_norm": 5.46875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4447, "epoch": 2, "loss": 0.0012, "lr": 1.3295115364850535e-05, "grad_norm": 2.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4448, "epoch": 2, "loss": 0.0869, "lr": 1.329228709009176e-05, "grad_norm": 4.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4449, "epoch": 2, "loss": 0.0638, "lr": 1.3289458519931903e-05, "grad_norm": 2.453125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4450, "epoch": 2, "loss": 0.3241, "lr": 1.3286629654624756e-05, "grad_norm": 5.9375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4451, "epoch": 2, "loss": 0.0201, "lr": 1.3283800494424139e-05, "grad_norm": 5.40625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 4452, "epoch": 2, "loss": 0.0653, "lr": 1.3280971039583906e-05, "grad_norm": 4.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4453, "epoch": 2, "loss": 0.0025, "lr": 1.3278141290357918e-05, "grad_norm": 3.875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4454, "epoch": 2, "loss": 0.0007, "lr": 1.3275311247000083e-05, "grad_norm": 1.3828125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4455, "epoch": 2, "loss": 0.002, "lr": 1.3272480909764323e-05, "grad_norm": 5.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4456, "epoch": 2, "loss": 0.0046, "lr": 1.3269650278904596e-05, "grad_norm": 6.3125, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 4457, "epoch": 2, "loss": 0.0065, "lr": 1.3266819354674877e-05, "grad_norm": 3.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4458, "epoch": 2, "loss": 0.0956, "lr": 1.3263988137329173e-05, "grad_norm": 3.921875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 4459, "epoch": 2, "loss": 0.0008, "lr": 1.3261156627121512e-05, "grad_norm": 5.0625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4460, "epoch": 2, "loss": 0.0198, "lr": 1.3258324824305954e-05, "grad_norm": 1.6484375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4461, "epoch": 2, "loss": 0.071, "lr": 1.3255492729136585e-05, "grad_norm": 8.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 4462, "epoch": 2, "loss": 0.0081, "lr": 1.3252660341867514e-05, "grad_norm": 1.1015625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4463, "epoch": 2, "loss": 0.0003, "lr": 1.3249827662752876e-05, "grad_norm": 2.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 4464, "epoch": 2, "loss": 0.1066, "lr": 1.3246994692046837e-05, "grad_norm": 6.96875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4465, "epoch": 2, "loss": 0.0922, "lr": 1.3244161430003581e-05, "grad_norm": 5.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4466, "epoch": 2, "loss": 0.0012, "lr": 1.3241327876877328e-05, "grad_norm": 6.09375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4467, "epoch": 2, "loss": 0.0144, "lr": 1.3238494032922313e-05, "grad_norm": 2.921875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4468, "epoch": 2, "loss": 0.0148, "lr": 1.323565989839281e-05, "grad_norm": 1.390625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4469, "epoch": 2, "loss": 0.0063, "lr": 1.323282547354311e-05, "grad_norm": 4.6875, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 4470, "epoch": 2, "loss": 0.0006, "lr": 1.3229990758627527e-05, "grad_norm": 5.78125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 4471, "epoch": 2, "loss": 0.0099, "lr": 1.3227155753900414e-05, "grad_norm": 4.75, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4472, "epoch": 2, "loss": 0.0219, "lr": 1.3224320459616136e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 4473, "epoch": 2, "loss": 0.0005, "lr": 1.3221484876029096e-05, "grad_norm": 5.78125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 4474, "epoch": 2, "loss": 0.0021, "lr": 1.3218649003393717e-05, "grad_norm": 1.8203125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4475, "epoch": 2, "loss": 0.0639, "lr": 1.3215812841964439e-05, "grad_norm": 6.53125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4476, "epoch": 2, "loss": 0.0573, "lr": 1.3212976391995752e-05, "grad_norm": 18.5, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4477, "epoch": 2, "loss": 0.1569, "lr": 1.3210139653742142e-05, "grad_norm": 5.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4478, "epoch": 2, "loss": 0.001, "lr": 1.320730262745815e-05, "grad_norm": 2.78125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4479, "epoch": 2, "loss": 0.0033, "lr": 1.3204465313398316e-05, "grad_norm": 1.140625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 4480, "epoch": 2, "loss": 0.0016, "lr": 1.3201627711817224e-05, "grad_norm": 0.55859375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4481, "epoch": 2, "loss": 0.0005, "lr": 1.3198789822969483e-05, "grad_norm": 5.53125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 4482, "epoch": 2, "loss": 0.0005, "lr": 1.3195951647109715e-05, "grad_norm": 5.21875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 4483, "epoch": 2, "loss": 0.0102, "lr": 1.3193113184492578e-05, "grad_norm": 5.15625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4484, "epoch": 2, "loss": 0.0009, "lr": 1.319027443537276e-05, "grad_norm": 3.09375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4485, "epoch": 2, "loss": 0.0021, "lr": 1.318743540000496e-05, "grad_norm": 0.55078125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4486, "epoch": 2, "loss": 0.0154, "lr": 1.3184596078643914e-05, "grad_norm": 4.78125, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 4487, "epoch": 2, "loss": 0.0035, "lr": 1.3181756471544383e-05, "grad_norm": 0.74609375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4488, "epoch": 2, "loss": 0.0484, "lr": 1.317891657896115e-05, "grad_norm": 4.625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4489, "epoch": 2, "loss": 0.0369, "lr": 1.317607640114902e-05, "grad_norm": 3.421875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.83} {"step": 4490, "epoch": 2, "loss": 0.0001, "lr": 1.3173235938362837e-05, "grad_norm": 4.71875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4491, "epoch": 2, "loss": 0.0028, "lr": 1.3170395190857456e-05, "grad_norm": 1.859375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4492, "epoch": 2, "loss": 0.0005, "lr": 1.3167554158887765e-05, "grad_norm": 2.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4493, "epoch": 2, "loss": 0.1352, "lr": 1.3164712842708679e-05, "grad_norm": 9.8125, "samples_per_sec": 10.38, "gpu_mem_gb": 6.83} {"step": 4494, "epoch": 2, "loss": 0.021, "lr": 1.3161871242575131e-05, "grad_norm": 2.703125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4495, "epoch": 2, "loss": 0.0002, "lr": 1.3159029358742084e-05, "grad_norm": 5.4375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4496, "epoch": 2, "loss": 0.0007, "lr": 1.3156187191464534e-05, "grad_norm": 2.109375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4497, "epoch": 2, "loss": 0.1469, "lr": 1.3153344740997484e-05, "grad_norm": 6.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4498, "epoch": 2, "loss": 0.0047, "lr": 1.3150502007595982e-05, "grad_norm": 1.9765625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 4499, "epoch": 2, "loss": 0.0803, "lr": 1.314765899151509e-05, "grad_norm": 4.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4500, "epoch": 2, "loss": 0.0002, "lr": 1.31448156930099e-05, "grad_norm": 1.328125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4501, "epoch": 2, "loss": 0.0029, "lr": 1.3141972112335526e-05, "grad_norm": 4.6875, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4502, "epoch": 2, "loss": 0.1662, "lr": 1.3139128249747101e-05, "grad_norm": 5.5, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4503, "epoch": 2, "loss": 0.0576, "lr": 1.3136284105499808e-05, "grad_norm": 4.75, "samples_per_sec": 10.52, "gpu_mem_gb": 6.84} {"step": 4504, "epoch": 2, "loss": 0.0027, "lr": 1.3133439679848824e-05, "grad_norm": 5.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4505, "epoch": 2, "loss": 0.2623, "lr": 1.3130594973049374e-05, "grad_norm": 6.78125, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4506, "epoch": 2, "loss": 0.0439, "lr": 1.312774998535669e-05, "grad_norm": 3.46875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 4507, "epoch": 2, "loss": 0.0012, "lr": 1.3124904717026051e-05, "grad_norm": 6.0, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4508, "epoch": 2, "loss": 0.0613, "lr": 1.3122059168312742e-05, "grad_norm": 5.875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 4509, "epoch": 2, "loss": 0.0454, "lr": 1.3119213339472083e-05, "grad_norm": 1.9375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 4510, "epoch": 2, "loss": 0.007, "lr": 1.3116367230759415e-05, "grad_norm": 4.78125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4511, "epoch": 2, "loss": 0.0082, "lr": 1.3113520842430108e-05, "grad_norm": 1.609375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4512, "epoch": 2, "loss": 0.0261, "lr": 1.311067417473955e-05, "grad_norm": 3.546875, "samples_per_sec": 11.0, "gpu_mem_gb": 6.82} {"step": 4513, "epoch": 2, "loss": 0.0007, "lr": 1.3107827227943163e-05, "grad_norm": 2.234375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 4514, "epoch": 2, "loss": 0.0131, "lr": 1.3104980002296389e-05, "grad_norm": 3.6875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4515, "epoch": 2, "loss": 0.0096, "lr": 1.3102132498054696e-05, "grad_norm": 3.875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 4516, "epoch": 2, "loss": 0.0115, "lr": 1.3099284715473575e-05, "grad_norm": 5.1875, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4517, "epoch": 2, "loss": 0.0276, "lr": 1.3096436654808546e-05, "grad_norm": 8.25, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4518, "epoch": 2, "loss": 0.0009, "lr": 1.309358831631515e-05, "grad_norm": 5.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4519, "epoch": 2, "loss": 0.0005, "lr": 1.3090739700248958e-05, "grad_norm": 2.828125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4520, "epoch": 2, "loss": 0.0327, "lr": 1.308789080686556e-05, "grad_norm": 9.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4521, "epoch": 2, "loss": 0.0089, "lr": 1.3085041636420573e-05, "grad_norm": 6.84375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4522, "epoch": 2, "loss": 0.0002, "lr": 1.308219218916964e-05, "grad_norm": 0.11279296875, "samples_per_sec": 11.02, "gpu_mem_gb": 6.82} {"step": 4523, "epoch": 2, "loss": 0.0013, "lr": 1.307934246536843e-05, "grad_norm": 7.46875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4524, "epoch": 2, "loss": 0.0529, "lr": 1.3076492465272632e-05, "grad_norm": 4.65625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4525, "epoch": 2, "loss": 0.0157, "lr": 1.3073642189137966e-05, "grad_norm": 6.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4526, "epoch": 2, "loss": 0.0019, "lr": 1.3070791637220166e-05, "grad_norm": 1.4609375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4527, "epoch": 2, "loss": 0.0005, "lr": 1.306794080977501e-05, "grad_norm": 0.69140625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4528, "epoch": 2, "loss": 0.0001, "lr": 1.3065089707058282e-05, "grad_norm": 6.3125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 4529, "epoch": 2, "loss": 0.0019, "lr": 1.3062238329325796e-05, "grad_norm": 5.28125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4530, "epoch": 2, "loss": 0.1447, "lr": 1.3059386676833398e-05, "grad_norm": 5.71875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 4531, "epoch": 2, "loss": 0.1224, "lr": 1.305653474983695e-05, "grad_norm": 4.75, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4532, "epoch": 2, "loss": 0.0249, "lr": 1.3053682548592343e-05, "grad_norm": 1.3984375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4533, "epoch": 2, "loss": 0.0515, "lr": 1.305083007335549e-05, "grad_norm": 8.4375, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4534, "epoch": 2, "loss": 0.2091, "lr": 1.3047977324382328e-05, "grad_norm": 6.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 4535, "epoch": 2, "loss": 0.0443, "lr": 1.3045124301928826e-05, "grad_norm": 4.9375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4536, "epoch": 2, "loss": 0.1497, "lr": 1.3042271006250965e-05, "grad_norm": 8.5625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4537, "epoch": 2, "loss": 0.0005, "lr": 1.3039417437604762e-05, "grad_norm": 2.75, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 4538, "epoch": 2, "loss": 0.0006, "lr": 1.3036563596246255e-05, "grad_norm": 3.671875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4539, "epoch": 2, "loss": 0.0001, "lr": 1.3033709482431503e-05, "grad_norm": 9.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4540, "epoch": 2, "loss": 0.0023, "lr": 1.3030855096416593e-05, "grad_norm": 5.5625, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 4541, "epoch": 2, "loss": 0.1173, "lr": 1.3028000438457631e-05, "grad_norm": 5.3125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4542, "epoch": 2, "loss": 0.0482, "lr": 1.302514550881076e-05, "grad_norm": 5.0625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4543, "epoch": 2, "loss": 0.0982, "lr": 1.302229030773213e-05, "grad_norm": 5.25, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4544, "epoch": 2, "loss": 0.1389, "lr": 1.3019434835477935e-05, "grad_norm": 4.03125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4545, "epoch": 2, "loss": 0.0008, "lr": 1.3016579092304373e-05, "grad_norm": 4.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4546, "epoch": 2, "loss": 0.072, "lr": 1.3013723078467679e-05, "grad_norm": 3.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4547, "epoch": 2, "loss": 0.001, "lr": 1.3010866794224115e-05, "grad_norm": 0.41015625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 4548, "epoch": 2, "loss": 0.0037, "lr": 1.3008010239829951e-05, "grad_norm": 3.25, "samples_per_sec": 8.93, "gpu_mem_gb": 6.83} {"step": 4549, "epoch": 2, "loss": 0.0009, "lr": 1.3005153415541503e-05, "grad_norm": 1.828125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4550, "epoch": 2, "loss": 0.0019, "lr": 1.3002296321615093e-05, "grad_norm": 2.578125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4551, "epoch": 2, "loss": 0.0004, "lr": 1.2999438958307076e-05, "grad_norm": 3.734375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4552, "epoch": 2, "loss": 0.0017, "lr": 1.2996581325873832e-05, "grad_norm": 9.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4553, "epoch": 2, "loss": 0.0014, "lr": 1.2993723424571757e-05, "grad_norm": 1.7890625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4554, "epoch": 2, "loss": 0.0011, "lr": 1.2990865254657281e-05, "grad_norm": 5.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4555, "epoch": 2, "loss": 0.0016, "lr": 1.2988006816386853e-05, "grad_norm": 3.109375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4556, "epoch": 2, "loss": 0.0948, "lr": 1.2985148110016947e-05, "grad_norm": 6.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4557, "epoch": 2, "loss": 0.0018, "lr": 1.298228913580406e-05, "grad_norm": 0.419921875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.85} {"step": 4558, "epoch": 2, "loss": 0.0292, "lr": 1.2979429894004716e-05, "grad_norm": 4.125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 4559, "epoch": 2, "loss": 0.0631, "lr": 1.2976570384875462e-05, "grad_norm": 3.0, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 4560, "epoch": 2, "loss": 0.0505, "lr": 1.297371060867286e-05, "grad_norm": 2.875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 4561, "epoch": 2, "loss": 0.0239, "lr": 1.2970850565653515e-05, "grad_norm": 3.21875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4562, "epoch": 2, "loss": 0.0226, "lr": 1.296799025607404e-05, "grad_norm": 2.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4563, "epoch": 2, "loss": 0.0043, "lr": 1.296512968019107e-05, "grad_norm": 4.71875, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 4564, "epoch": 2, "loss": 0.0027, "lr": 1.2962268838261284e-05, "grad_norm": 1.703125, "samples_per_sec": 10.41, "gpu_mem_gb": 6.82} {"step": 4565, "epoch": 2, "loss": 0.0005, "lr": 1.2959407730541362e-05, "grad_norm": 0.87109375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4566, "epoch": 2, "loss": 0.0695, "lr": 1.295654635728802e-05, "grad_norm": 4.78125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4567, "epoch": 2, "loss": 0.0004, "lr": 1.2953684718757997e-05, "grad_norm": 0.41796875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4568, "epoch": 2, "loss": 0.0121, "lr": 1.2950822815208052e-05, "grad_norm": 3.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4569, "epoch": 2, "loss": 0.0847, "lr": 1.294796064689497e-05, "grad_norm": 5.59375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4570, "epoch": 2, "loss": 0.1066, "lr": 1.294509821407556e-05, "grad_norm": 9.0, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4571, "epoch": 2, "loss": 0.0555, "lr": 1.2942235517006655e-05, "grad_norm": 5.71875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4572, "epoch": 2, "loss": 0.0003, "lr": 1.2939372555945112e-05, "grad_norm": 4.03125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4573, "epoch": 2, "loss": 0.0363, "lr": 1.2936509331147804e-05, "grad_norm": 5.25, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 4574, "epoch": 2, "loss": 0.0005, "lr": 1.2933645842871643e-05, "grad_norm": 6.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4575, "epoch": 2, "loss": 0.0026, "lr": 1.2930782091373551e-05, "grad_norm": 1.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4576, "epoch": 2, "loss": 0.0825, "lr": 1.2927918076910481e-05, "grad_norm": 3.765625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.84} {"step": 4577, "epoch": 2, "loss": 0.0076, "lr": 1.2925053799739407e-05, "grad_norm": 4.0625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4578, "epoch": 2, "loss": 0.0049, "lr": 1.2922189260117324e-05, "grad_norm": 2.28125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4579, "epoch": 2, "loss": 0.0507, "lr": 1.291932445830126e-05, "grad_norm": 2.453125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4580, "epoch": 2, "loss": 0.0183, "lr": 1.291645939454825e-05, "grad_norm": 1.9296875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4581, "epoch": 2, "loss": 0.0009, "lr": 1.2913594069115371e-05, "grad_norm": 0.1748046875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4582, "epoch": 2, "loss": 0.0322, "lr": 1.2910728482259711e-05, "grad_norm": 2.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 4583, "epoch": 2, "loss": 0.1722, "lr": 1.2907862634238386e-05, "grad_norm": 9.1875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4584, "epoch": 2, "loss": 0.0027, "lr": 1.2904996525308534e-05, "grad_norm": 5.9375, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 4585, "epoch": 2, "loss": 0.0008, "lr": 1.2902130155727319e-05, "grad_norm": 6.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 4586, "epoch": 2, "loss": 0.0017, "lr": 1.2899263525751926e-05, "grad_norm": 3.234375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4587, "epoch": 2, "loss": 0.1264, "lr": 1.2896396635639565e-05, "grad_norm": 5.3125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4588, "epoch": 2, "loss": 0.0039, "lr": 1.2893529485647462e-05, "grad_norm": 6.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4589, "epoch": 2, "loss": 0.0042, "lr": 1.289066207603288e-05, "grad_norm": 6.84375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4590, "epoch": 2, "loss": 0.1449, "lr": 1.2887794407053096e-05, "grad_norm": 6.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4591, "epoch": 2, "loss": 0.0227, "lr": 1.2884926478965412e-05, "grad_norm": 6.375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4592, "epoch": 2, "loss": 0.0431, "lr": 1.2882058292027148e-05, "grad_norm": 3.765625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4593, "epoch": 2, "loss": 0.0382, "lr": 1.287918984649566e-05, "grad_norm": 4.375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4594, "epoch": 2, "loss": 0.1607, "lr": 1.2876321142628319e-05, "grad_norm": 6.1875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4595, "epoch": 2, "loss": 0.0947, "lr": 1.2873452180682516e-05, "grad_norm": 5.28125, "samples_per_sec": 11.01, "gpu_mem_gb": 6.85} {"step": 4596, "epoch": 2, "loss": 0.0023, "lr": 1.2870582960915669e-05, "grad_norm": 2.625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4597, "epoch": 2, "loss": 0.0004, "lr": 1.2867713483585224e-05, "grad_norm": 0.267578125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4598, "epoch": 2, "loss": 0.006, "lr": 1.2864843748948643e-05, "grad_norm": 6.875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4599, "epoch": 2, "loss": 0.0004, "lr": 1.2861973757263416e-05, "grad_norm": 4.5625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4600, "epoch": 2, "loss": 0.0477, "lr": 1.2859103508787045e-05, "grad_norm": 4.40625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4601, "epoch": 2, "loss": 0.0007, "lr": 1.2856233003777075e-05, "grad_norm": 5.34375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4602, "epoch": 2, "loss": 0.0007, "lr": 1.2853362242491054e-05, "grad_norm": 3.546875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4603, "epoch": 2, "loss": 0.043, "lr": 1.2850491225186567e-05, "grad_norm": 3.671875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4604, "epoch": 2, "loss": 0.3096, "lr": 1.2847619952121214e-05, "grad_norm": 6.1875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 4605, "epoch": 2, "loss": 0.003, "lr": 1.2844748423552617e-05, "grad_norm": 2.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4606, "epoch": 2, "loss": 0.0696, "lr": 1.2841876639738433e-05, "grad_norm": 5.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4607, "epoch": 2, "loss": 0.0026, "lr": 1.2839004600936326e-05, "grad_norm": 1.375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4608, "epoch": 2, "loss": 0.0004, "lr": 1.2836132307403994e-05, "grad_norm": 5.1875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 4609, "epoch": 2, "loss": 0.1011, "lr": 1.2833259759399155e-05, "grad_norm": 9.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4610, "epoch": 2, "loss": 0.0002, "lr": 1.2830386957179543e-05, "grad_norm": 3.03125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4611, "epoch": 2, "loss": 0.0208, "lr": 1.2827513901002928e-05, "grad_norm": 5.90625, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 4612, "epoch": 2, "loss": 0.0477, "lr": 1.2824640591127086e-05, "grad_norm": 4.5, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4613, "epoch": 2, "loss": 0.1222, "lr": 1.282176702780984e-05, "grad_norm": 23.625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4614, "epoch": 2, "loss": 0.0009, "lr": 1.2818893211309007e-05, "grad_norm": 4.4375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 4615, "epoch": 2, "loss": 0.1303, "lr": 1.2816019141882448e-05, "grad_norm": 7.90625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4616, "epoch": 2, "loss": 0.0037, "lr": 1.2813144819788036e-05, "grad_norm": 3.296875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4617, "epoch": 2, "loss": 0.0244, "lr": 1.2810270245283672e-05, "grad_norm": 1.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4618, "epoch": 2, "loss": 0.0066, "lr": 1.2807395418627278e-05, "grad_norm": 4.65625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 4619, "epoch": 2, "loss": 0.0006, "lr": 1.2804520340076798e-05, "grad_norm": 6.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4620, "epoch": 2, "loss": 0.041, "lr": 1.2801645009890195e-05, "grad_norm": 6.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4621, "epoch": 2, "loss": 0.001, "lr": 1.2798769428325468e-05, "grad_norm": 4.75, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4622, "epoch": 2, "loss": 0.0108, "lr": 1.279589359564062e-05, "grad_norm": 5.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4623, "epoch": 2, "loss": 0.0747, "lr": 1.2793017512093688e-05, "grad_norm": 3.84375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4624, "epoch": 2, "loss": 0.184, "lr": 1.2790141177942737e-05, "grad_norm": 8.25, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 4625, "epoch": 2, "loss": 0.0016, "lr": 1.2787264593445832e-05, "grad_norm": 3.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 4626, "epoch": 2, "loss": 0.3064, "lr": 1.2784387758861088e-05, "grad_norm": 8.875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4627, "epoch": 2, "loss": 0.0014, "lr": 1.2781510674446618e-05, "grad_norm": 5.9375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4628, "epoch": 2, "loss": 0.0371, "lr": 1.277863334046058e-05, "grad_norm": 4.09375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 4629, "epoch": 2, "loss": 0.0011, "lr": 1.277575575716114e-05, "grad_norm": 3.640625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 4630, "epoch": 2, "loss": 0.0308, "lr": 1.2772877924806486e-05, "grad_norm": 2.59375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 4631, "epoch": 2, "loss": 0.0007, "lr": 1.2769999843654834e-05, "grad_norm": 5.0625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4632, "epoch": 2, "loss": 0.0453, "lr": 1.276712151396442e-05, "grad_norm": 2.328125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4633, "epoch": 2, "loss": 0.1471, "lr": 1.2764242935993508e-05, "grad_norm": 6.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4634, "epoch": 2, "loss": 0.0003, "lr": 1.2761364110000372e-05, "grad_norm": 7.3125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4635, "epoch": 2, "loss": 0.0004, "lr": 1.2758485036243316e-05, "grad_norm": 5.71875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4636, "epoch": 2, "loss": 0.0925, "lr": 1.2755605714980667e-05, "grad_norm": 7.40625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4637, "epoch": 2, "loss": 0.0047, "lr": 1.2752726146470775e-05, "grad_norm": 4.34375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4638, "epoch": 2, "loss": 0.0915, "lr": 1.2749846330972009e-05, "grad_norm": 4.84375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 4639, "epoch": 2, "loss": 0.0334, "lr": 1.2746966268742753e-05, "grad_norm": 9.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4640, "epoch": 2, "loss": 0.0469, "lr": 1.2744085960041433e-05, "grad_norm": 5.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4641, "epoch": 2, "loss": 0.0002, "lr": 1.274120540512648e-05, "grad_norm": 5.40625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 4642, "epoch": 2, "loss": 0.0257, "lr": 1.273832460425635e-05, "grad_norm": 6.375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4643, "epoch": 2, "loss": 0.0031, "lr": 1.2735443557689526e-05, "grad_norm": 0.3046875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4644, "epoch": 2, "loss": 0.0013, "lr": 1.2732562265684512e-05, "grad_norm": 4.625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4645, "epoch": 2, "loss": 0.0007, "lr": 1.2729680728499829e-05, "grad_norm": 6.5, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 4646, "epoch": 2, "loss": 0.0013, "lr": 1.2726798946394026e-05, "grad_norm": 1.0703125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4647, "epoch": 2, "loss": 0.0444, "lr": 1.2723916919625667e-05, "grad_norm": 3.15625, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4648, "epoch": 2, "loss": 0.107, "lr": 1.2721034648453353e-05, "grad_norm": 3.453125, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 4649, "epoch": 2, "loss": 0.0022, "lr": 1.2718152133135683e-05, "grad_norm": 3.09375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4650, "epoch": 2, "loss": 0.035, "lr": 1.27152693739313e-05, "grad_norm": 7.625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4651, "epoch": 2, "loss": 0.0014, "lr": 1.2712386371098857e-05, "grad_norm": 6.90625, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 4652, "epoch": 2, "loss": 0.0019, "lr": 1.2709503124897032e-05, "grad_norm": 4.53125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 4653, "epoch": 2, "loss": 0.0017, "lr": 1.2706619635584528e-05, "grad_norm": 3.453125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4654, "epoch": 2, "loss": 0.0201, "lr": 1.2703735903420062e-05, "grad_norm": 4.625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4655, "epoch": 2, "loss": 0.0001, "lr": 1.2700851928662376e-05, "grad_norm": 0.130859375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4656, "epoch": 2, "loss": 0.1216, "lr": 1.2697967711570243e-05, "grad_norm": 3.984375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4657, "epoch": 2, "loss": 0.0602, "lr": 1.2695083252402447e-05, "grad_norm": 5.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4658, "epoch": 2, "loss": 0.0004, "lr": 1.2692198551417794e-05, "grad_norm": 3.65625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.82} {"step": 4659, "epoch": 2, "loss": 0.0005, "lr": 1.268931360887511e-05, "grad_norm": 4.46875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4660, "epoch": 2, "loss": 0.0731, "lr": 1.268642842503326e-05, "grad_norm": 2.8125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.84} {"step": 4661, "epoch": 2, "loss": 0.1336, "lr": 1.2683543000151105e-05, "grad_norm": 4.09375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 4662, "epoch": 2, "loss": 0.0261, "lr": 1.2680657334487547e-05, "grad_norm": 5.28125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 4663, "epoch": 2, "loss": 0.0002, "lr": 1.2677771428301501e-05, "grad_norm": 0.578125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 4664, "epoch": 2, "loss": 0.0919, "lr": 1.2674885281851907e-05, "grad_norm": 4.96875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4665, "epoch": 2, "loss": 0.1387, "lr": 1.2671998895397725e-05, "grad_norm": 5.65625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4666, "epoch": 2, "loss": 0.0018, "lr": 1.2669112269197935e-05, "grad_norm": 2.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4667, "epoch": 2, "loss": 0.0789, "lr": 1.2666225403511537e-05, "grad_norm": 4.90625, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4668, "epoch": 2, "loss": 0.0069, "lr": 1.2663338298597562e-05, "grad_norm": 8.25, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4669, "epoch": 2, "loss": 0.1702, "lr": 1.2660450954715053e-05, "grad_norm": 3.828125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4670, "epoch": 2, "loss": 0.0004, "lr": 1.2657563372123076e-05, "grad_norm": 3.953125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4671, "epoch": 2, "loss": 0.1037, "lr": 1.2654675551080724e-05, "grad_norm": 5.0625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4672, "epoch": 2, "loss": 0.0031, "lr": 1.2651787491847105e-05, "grad_norm": 2.1875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4673, "epoch": 2, "loss": 0.0031, "lr": 1.2648899194681347e-05, "grad_norm": 0.5625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4674, "epoch": 2, "loss": 0.1095, "lr": 1.264601065984261e-05, "grad_norm": 3.546875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4675, "epoch": 2, "loss": 0.0271, "lr": 1.2643121887590064e-05, "grad_norm": 5.25, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4676, "epoch": 2, "loss": 0.1425, "lr": 1.2640232878182905e-05, "grad_norm": 5.875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 4677, "epoch": 2, "loss": 0.0027, "lr": 1.2637343631880352e-05, "grad_norm": 1.2265625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 4678, "epoch": 2, "loss": 0.0005, "lr": 1.263445414894164e-05, "grad_norm": 1.8359375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4679, "epoch": 2, "loss": 0.0207, "lr": 1.2631564429626028e-05, "grad_norm": 2.703125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4680, "epoch": 2, "loss": 0.0407, "lr": 1.2628674474192802e-05, "grad_norm": 5.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 4681, "epoch": 2, "loss": 0.0171, "lr": 1.2625784282901259e-05, "grad_norm": 1.71875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4682, "epoch": 2, "loss": 0.1149, "lr": 1.262289385601072e-05, "grad_norm": 5.53125, "samples_per_sec": 10.78, "gpu_mem_gb": 6.85} {"step": 4683, "epoch": 2, "loss": 0.0387, "lr": 1.262000319378054e-05, "grad_norm": 6.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4684, "epoch": 2, "loss": 0.0003, "lr": 1.261711229647007e-05, "grad_norm": 3.171875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4685, "epoch": 2, "loss": 0.0178, "lr": 1.2614221164338708e-05, "grad_norm": 5.6875, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 4686, "epoch": 2, "loss": 0.0002, "lr": 1.2611329797645851e-05, "grad_norm": 3.015625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4687, "epoch": 2, "loss": 0.0134, "lr": 1.2608438196650937e-05, "grad_norm": 10.5, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4688, "epoch": 2, "loss": 0.0002, "lr": 1.2605546361613407e-05, "grad_norm": 3.3125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4689, "epoch": 2, "loss": 0.001, "lr": 1.260265429279274e-05, "grad_norm": 4.1875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4690, "epoch": 2, "loss": 0.0507, "lr": 1.259976199044842e-05, "grad_norm": 2.390625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4691, "epoch": 2, "loss": 0.0322, "lr": 1.2596869454839964e-05, "grad_norm": 4.84375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4692, "epoch": 2, "loss": 0.0007, "lr": 1.2593976686226906e-05, "grad_norm": 3.890625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 4693, "epoch": 2, "loss": 0.0004, "lr": 1.2591083684868795e-05, "grad_norm": 2.46875, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4694, "epoch": 2, "loss": 0.1118, "lr": 1.2588190451025209e-05, "grad_norm": 5.625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4695, "epoch": 2, "loss": 0.0007, "lr": 1.2585296984955746e-05, "grad_norm": 4.8125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 4696, "epoch": 2, "loss": 0.0415, "lr": 1.258240328692002e-05, "grad_norm": 4.875, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 4697, "epoch": 2, "loss": 0.2349, "lr": 1.2579509357177668e-05, "grad_norm": 10.125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4698, "epoch": 2, "loss": 0.0304, "lr": 1.2576615195988349e-05, "grad_norm": 2.203125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4699, "epoch": 2, "loss": 0.1207, "lr": 1.2573720803611746e-05, "grad_norm": 3.75, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 4700, "epoch": 2, "loss": 0.0007, "lr": 1.2570826180307556e-05, "grad_norm": 1.4296875, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 4701, "epoch": 2, "loss": 0.104, "lr": 1.2567931326335496e-05, "grad_norm": 5.125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4702, "epoch": 2, "loss": 0.0012, "lr": 1.2565036241955313e-05, "grad_norm": 0.134765625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4703, "epoch": 2, "loss": 0.1147, "lr": 1.2562140927426768e-05, "grad_norm": 2.953125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 4704, "epoch": 2, "loss": 0.0134, "lr": 1.2559245383009642e-05, "grad_norm": 5.4375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4705, "epoch": 2, "loss": 0.0003, "lr": 1.2556349608963739e-05, "grad_norm": 5.5625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 4706, "epoch": 2, "loss": 0.1649, "lr": 1.255345360554888e-05, "grad_norm": 4.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.85} {"step": 4707, "epoch": 2, "loss": 0.1798, "lr": 1.2550557373024915e-05, "grad_norm": 4.40625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4708, "epoch": 2, "loss": 0.214, "lr": 1.2547660911651702e-05, "grad_norm": 4.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4709, "epoch": 2, "loss": 0.0117, "lr": 1.2544764221689136e-05, "grad_norm": 4.96875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 4710, "epoch": 2, "loss": 0.0203, "lr": 1.2541867303397114e-05, "grad_norm": 1.3515625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4711, "epoch": 2, "loss": 0.06, "lr": 1.2538970157035568e-05, "grad_norm": 6.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4712, "epoch": 2, "loss": 0.002, "lr": 1.2536072782864444e-05, "grad_norm": 4.4375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 4713, "epoch": 2, "loss": 0.007, "lr": 1.2533175181143704e-05, "grad_norm": 0.6328125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4714, "epoch": 2, "loss": 0.0276, "lr": 1.2530277352133346e-05, "grad_norm": 5.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4715, "epoch": 2, "loss": 0.084, "lr": 1.252737929609337e-05, "grad_norm": 4.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4716, "epoch": 2, "loss": 0.0025, "lr": 1.2524481013283809e-05, "grad_norm": 4.53125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4717, "epoch": 2, "loss": 0.0087, "lr": 1.2521582503964708e-05, "grad_norm": 4.78125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 4718, "epoch": 2, "loss": 0.0009, "lr": 1.2518683768396141e-05, "grad_norm": 2.265625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4719, "epoch": 2, "loss": 0.0784, "lr": 1.2515784806838196e-05, "grad_norm": 6.03125, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4720, "epoch": 2, "loss": 0.0003, "lr": 1.2512885619550978e-05, "grad_norm": 5.875, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4721, "epoch": 2, "loss": 0.0008, "lr": 1.2509986206794624e-05, "grad_norm": 3.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4722, "epoch": 2, "loss": 0.1462, "lr": 1.2507086568829285e-05, "grad_norm": 6.03125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4723, "epoch": 2, "loss": 0.1077, "lr": 1.2504186705915125e-05, "grad_norm": 5.90625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.84} {"step": 4724, "epoch": 2, "loss": 0.0487, "lr": 1.2501286618312339e-05, "grad_norm": 3.296875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4725, "epoch": 2, "loss": 0.0024, "lr": 1.2498386306281132e-05, "grad_norm": 7.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4726, "epoch": 2, "loss": 0.0831, "lr": 1.2495485770081745e-05, "grad_norm": 6.5, "samples_per_sec": 10.58, "gpu_mem_gb": 6.83} {"step": 4727, "epoch": 2, "loss": 0.003, "lr": 1.2492585009974425e-05, "grad_norm": 3.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4728, "epoch": 2, "loss": 0.0325, "lr": 1.248968402621944e-05, "grad_norm": 4.5, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4729, "epoch": 2, "loss": 0.0009, "lr": 1.2486782819077083e-05, "grad_norm": 0.72265625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 4730, "epoch": 2, "loss": 0.018, "lr": 1.2483881388807666e-05, "grad_norm": 2.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4731, "epoch": 2, "loss": 0.0023, "lr": 1.2480979735671524e-05, "grad_norm": 3.90625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4732, "epoch": 2, "loss": 0.2391, "lr": 1.2478077859929e-05, "grad_norm": 4.65625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4733, "epoch": 2, "loss": 0.0157, "lr": 1.2475175761840473e-05, "grad_norm": 7.21875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4734, "epoch": 2, "loss": 0.0031, "lr": 1.2472273441666332e-05, "grad_norm": 0.69921875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4735, "epoch": 2, "loss": 0.0006, "lr": 1.2469370899666986e-05, "grad_norm": 4.53125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4736, "epoch": 2, "loss": 0.0012, "lr": 1.246646813610287e-05, "grad_norm": 4.8125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4737, "epoch": 2, "loss": 0.0349, "lr": 1.2463565151234429e-05, "grad_norm": 5.84375, "samples_per_sec": 11.04, "gpu_mem_gb": 6.85} {"step": 4738, "epoch": 2, "loss": 0.0018, "lr": 1.246066194532214e-05, "grad_norm": 0.333984375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 4739, "epoch": 2, "loss": 0.0002, "lr": 1.2457758518626493e-05, "grad_norm": 4.84375, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 4740, "epoch": 2, "loss": 0.0441, "lr": 1.2454854871407993e-05, "grad_norm": 2.828125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.85} {"step": 4741, "epoch": 2, "loss": 0.0026, "lr": 1.2451951003927176e-05, "grad_norm": 6.75, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4742, "epoch": 2, "loss": 0.1048, "lr": 1.244904691644459e-05, "grad_norm": 2.109375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 4743, "epoch": 2, "loss": 0.0137, "lr": 1.2446142609220806e-05, "grad_norm": 5.40625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 4744, "epoch": 2, "loss": 0.0019, "lr": 1.2443238082516414e-05, "grad_norm": 1.0390625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 4745, "epoch": 2, "loss": 0.0011, "lr": 1.2440333336592017e-05, "grad_norm": 3.5625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4746, "epoch": 2, "loss": 0.006, "lr": 1.2437428371708253e-05, "grad_norm": 1.5234375, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 4747, "epoch": 2, "loss": 0.0109, "lr": 1.2434523188125764e-05, "grad_norm": 4.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4748, "epoch": 2, "loss": 0.1513, "lr": 1.2431617786105223e-05, "grad_norm": 4.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4749, "epoch": 2, "loss": 0.0257, "lr": 1.2428712165907313e-05, "grad_norm": 3.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4750, "epoch": 2, "loss": 0.0533, "lr": 1.2425806327792742e-05, "grad_norm": 2.703125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4751, "epoch": 2, "loss": 0.0038, "lr": 1.2422900272022242e-05, "grad_norm": 2.375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4752, "epoch": 2, "loss": 0.0011, "lr": 1.2419993998856553e-05, "grad_norm": 6.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 4753, "epoch": 2, "loss": 0.0357, "lr": 1.2417087508556443e-05, "grad_norm": 2.515625, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 4754, "epoch": 2, "loss": 0.0007, "lr": 1.24141808013827e-05, "grad_norm": 3.984375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4755, "epoch": 2, "loss": 0.1058, "lr": 1.2411273877596126e-05, "grad_norm": 5.5625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4756, "epoch": 2, "loss": 0.0394, "lr": 1.2408366737457546e-05, "grad_norm": 3.640625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4757, "epoch": 2, "loss": 0.1837, "lr": 1.2405459381227803e-05, "grad_norm": 4.375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4758, "epoch": 2, "loss": 0.0205, "lr": 1.2402551809167762e-05, "grad_norm": 6.09375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4759, "epoch": 2, "loss": 0.0024, "lr": 1.2399644021538304e-05, "grad_norm": 1.2109375, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4760, "epoch": 2, "loss": 0.0205, "lr": 1.239673601860033e-05, "grad_norm": 6.4375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.85} {"step": 4761, "epoch": 2, "loss": 0.0002, "lr": 1.2393827800614766e-05, "grad_norm": 1.578125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4762, "epoch": 2, "loss": 0.1019, "lr": 1.2390919367842547e-05, "grad_norm": 8.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4763, "epoch": 2, "loss": 0.0037, "lr": 1.2388010720544634e-05, "grad_norm": 5.15625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4764, "epoch": 2, "loss": 0.225, "lr": 1.2385101858982004e-05, "grad_norm": 7.59375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4765, "epoch": 2, "loss": 0.2914, "lr": 1.2382192783415661e-05, "grad_norm": 6.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4766, "epoch": 2, "loss": 0.0002, "lr": 1.2379283494106618e-05, "grad_norm": 2.328125, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 4767, "epoch": 2, "loss": 0.0898, "lr": 1.2376373991315911e-05, "grad_norm": 5.3125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.85} {"step": 4768, "epoch": 2, "loss": 0.0253, "lr": 1.2373464275304596e-05, "grad_norm": 3.890625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4769, "epoch": 2, "loss": 0.1119, "lr": 1.2370554346333751e-05, "grad_norm": 5.21875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 4770, "epoch": 2, "loss": 0.1705, "lr": 1.2367644204664468e-05, "grad_norm": 4.78125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4771, "epoch": 2, "loss": 0.0306, "lr": 1.2364733850557862e-05, "grad_norm": 6.875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 4772, "epoch": 2, "loss": 0.0023, "lr": 1.2361823284275055e-05, "grad_norm": 4.03125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4773, "epoch": 2, "loss": 0.0236, "lr": 1.2358912506077214e-05, "grad_norm": 5.8125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4774, "epoch": 2, "loss": 0.0011, "lr": 1.2356001516225498e-05, "grad_norm": 3.078125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4775, "epoch": 2, "loss": 0.0003, "lr": 1.23530903149811e-05, "grad_norm": 5.09375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4776, "epoch": 2, "loss": 0.0722, "lr": 1.2350178902605228e-05, "grad_norm": 2.59375, "samples_per_sec": 8.64, "gpu_mem_gb": 6.82} {"step": 4777, "epoch": 2, "loss": 0.002, "lr": 1.2347267279359105e-05, "grad_norm": 0.9765625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4778, "epoch": 2, "loss": 0.0016, "lr": 1.2344355445503984e-05, "grad_norm": 5.5, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 4779, "epoch": 2, "loss": 0.1377, "lr": 1.2341443401301123e-05, "grad_norm": 4.9375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4780, "epoch": 2, "loss": 0.001, "lr": 1.233853114701181e-05, "grad_norm": 4.03125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 4781, "epoch": 2, "loss": 0.0095, "lr": 1.2335618682897347e-05, "grad_norm": 2.71875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4782, "epoch": 2, "loss": 0.0229, "lr": 1.2332706009219055e-05, "grad_norm": 3.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4783, "epoch": 2, "loss": 0.0829, "lr": 1.2329793126238276e-05, "grad_norm": 5.875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4784, "epoch": 2, "loss": 0.0226, "lr": 1.2326880034216363e-05, "grad_norm": 3.8125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4785, "epoch": 2, "loss": 0.0048, "lr": 1.2323966733414698e-05, "grad_norm": 4.71875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4786, "epoch": 2, "loss": 0.009, "lr": 1.2321053224094678e-05, "grad_norm": 3.5625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.82} {"step": 4787, "epoch": 2, "loss": 0.0081, "lr": 1.2318139506517718e-05, "grad_norm": 4.28125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4788, "epoch": 2, "loss": 0.0004, "lr": 1.2315225580945252e-05, "grad_norm": 0.95703125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4789, "epoch": 2, "loss": 0.0003, "lr": 1.2312311447638731e-05, "grad_norm": 1.640625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4790, "epoch": 2, "loss": 0.0765, "lr": 1.2309397106859628e-05, "grad_norm": 6.28125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 4791, "epoch": 2, "loss": 0.0462, "lr": 1.230648255886943e-05, "grad_norm": 2.53125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4792, "epoch": 2, "loss": 0.0013, "lr": 1.2303567803929646e-05, "grad_norm": 2.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4793, "epoch": 2, "loss": 0.0006, "lr": 1.2300652842301808e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.55, "gpu_mem_gb": 6.85} {"step": 4794, "epoch": 2, "loss": 0.0003, "lr": 1.2297737674247454e-05, "grad_norm": 2.25, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4795, "epoch": 2, "loss": 0.003, "lr": 1.2294822300028153e-05, "grad_norm": 4.5, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4796, "epoch": 2, "loss": 0.1198, "lr": 1.2291906719905487e-05, "grad_norm": 4.90625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4797, "epoch": 2, "loss": 0.0034, "lr": 1.2288990934141051e-05, "grad_norm": 0.251953125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4798, "epoch": 2, "loss": 0.0003, "lr": 1.2286074942996476e-05, "grad_norm": 1.1171875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.83} {"step": 4799, "epoch": 2, "loss": 0.003, "lr": 1.228315874673339e-05, "grad_norm": 11.25, "samples_per_sec": 10.83, "gpu_mem_gb": 6.85} {"step": 4800, "epoch": 2, "loss": 0.0057, "lr": 1.228024234561345e-05, "grad_norm": 4.4375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 4801, "epoch": 2, "loss": 0.0005, "lr": 1.2277325739898337e-05, "grad_norm": 5.375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4802, "epoch": 2, "loss": 0.139, "lr": 1.2274408929849741e-05, "grad_norm": 3.703125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.84} {"step": 4803, "epoch": 2, "loss": 0.0063, "lr": 1.227149191572937e-05, "grad_norm": 3.640625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 4804, "epoch": 2, "loss": 0.0102, "lr": 1.2268574697798956e-05, "grad_norm": 8.25, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4805, "epoch": 2, "loss": 0.0115, "lr": 1.226565727632025e-05, "grad_norm": 5.78125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4806, "epoch": 2, "loss": 0.1407, "lr": 1.2262739651555011e-05, "grad_norm": 4.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4807, "epoch": 2, "loss": 0.0331, "lr": 1.225982182376503e-05, "grad_norm": 1.6875, "samples_per_sec": 10.29, "gpu_mem_gb": 6.83} {"step": 4808, "epoch": 2, "loss": 0.0004, "lr": 1.2256903793212107e-05, "grad_norm": 4.28125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4809, "epoch": 2, "loss": 0.0251, "lr": 1.2253985560158064e-05, "grad_norm": 1.3671875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4810, "epoch": 2, "loss": 0.0007, "lr": 1.225106712486474e-05, "grad_norm": 1.0859375, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 4811, "epoch": 2, "loss": 0.0518, "lr": 1.2248148487593988e-05, "grad_norm": 5.5, "samples_per_sec": 11.06, "gpu_mem_gb": 6.83} {"step": 4812, "epoch": 2, "loss": 0.0019, "lr": 1.224522964860769e-05, "grad_norm": 5.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4813, "epoch": 2, "loss": 0.0565, "lr": 1.2242310608167735e-05, "grad_norm": 6.875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 4814, "epoch": 2, "loss": 0.0636, "lr": 1.2239391366536036e-05, "grad_norm": 7.34375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 4815, "epoch": 2, "loss": 0.0071, "lr": 1.223647192397452e-05, "grad_norm": 6.0625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4816, "epoch": 2, "loss": 0.0688, "lr": 1.2233552280745138e-05, "grad_norm": 3.984375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4817, "epoch": 2, "loss": 0.308, "lr": 1.2230632437109855e-05, "grad_norm": 7.53125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4818, "epoch": 2, "loss": 0.1881, "lr": 1.2227712393330653e-05, "grad_norm": 7.28125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 4819, "epoch": 2, "loss": 0.0124, "lr": 1.2224792149669533e-05, "grad_norm": 2.125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4820, "epoch": 2, "loss": 0.0035, "lr": 1.2221871706388517e-05, "grad_norm": 4.78125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4821, "epoch": 2, "loss": 0.1173, "lr": 1.221895106374964e-05, "grad_norm": 2.921875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4822, "epoch": 2, "loss": 0.0679, "lr": 1.2216030222014959e-05, "grad_norm": 7.375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4823, "epoch": 2, "loss": 0.0158, "lr": 1.2213109181446546e-05, "grad_norm": 2.59375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4824, "epoch": 2, "loss": 0.039, "lr": 1.2210187942306491e-05, "grad_norm": 4.40625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 4825, "epoch": 2, "loss": 0.0068, "lr": 1.2207266504856905e-05, "grad_norm": 0.87109375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4826, "epoch": 2, "loss": 0.027, "lr": 1.2204344869359912e-05, "grad_norm": 2.953125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4827, "epoch": 2, "loss": 0.0182, "lr": 1.2201423036077657e-05, "grad_norm": 6.59375, "samples_per_sec": 10.38, "gpu_mem_gb": 6.83} {"step": 4828, "epoch": 2, "loss": 0.1477, "lr": 1.2198501005272304e-05, "grad_norm": 4.40625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4829, "epoch": 2, "loss": 0.0348, "lr": 1.2195578777206032e-05, "grad_norm": 5.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4830, "epoch": 2, "loss": 0.0002, "lr": 1.219265635214104e-05, "grad_norm": 5.75, "samples_per_sec": 10.27, "gpu_mem_gb": 6.85} {"step": 4831, "epoch": 2, "loss": 0.0017, "lr": 1.2189733730339535e-05, "grad_norm": 0.30078125, "samples_per_sec": 11.04, "gpu_mem_gb": 6.82} {"step": 4832, "epoch": 2, "loss": 0.0007, "lr": 1.218681091206376e-05, "grad_norm": 3.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4833, "epoch": 2, "loss": 0.083, "lr": 1.2183887897575959e-05, "grad_norm": 6.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4834, "epoch": 2, "loss": 0.0473, "lr": 1.2180964687138406e-05, "grad_norm": 2.109375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 4835, "epoch": 2, "loss": 0.0013, "lr": 1.217804128101338e-05, "grad_norm": 5.40625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4836, "epoch": 2, "loss": 0.1407, "lr": 1.2175117679463187e-05, "grad_norm": 4.125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4837, "epoch": 2, "loss": 0.0014, "lr": 1.2172193882750149e-05, "grad_norm": 4.96875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 4838, "epoch": 2, "loss": 0.1208, "lr": 1.2169269891136603e-05, "grad_norm": 3.109375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 4839, "epoch": 2, "loss": 0.0939, "lr": 1.2166345704884903e-05, "grad_norm": 5.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4840, "epoch": 2, "loss": 0.2519, "lr": 1.2163421324257428e-05, "grad_norm": 5.96875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 4841, "epoch": 2, "loss": 0.0029, "lr": 1.2160496749516559e-05, "grad_norm": 4.90625, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4842, "epoch": 2, "loss": 0.0229, "lr": 1.2157571980924713e-05, "grad_norm": 4.1875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4843, "epoch": 2, "loss": 0.0863, "lr": 1.215464701874431e-05, "grad_norm": 4.625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4844, "epoch": 2, "loss": 0.0029, "lr": 1.2151721863237795e-05, "grad_norm": 3.453125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 4845, "epoch": 2, "loss": 0.0685, "lr": 1.2148796514667629e-05, "grad_norm": 5.84375, "samples_per_sec": 11.01, "gpu_mem_gb": 6.83} {"step": 4846, "epoch": 2, "loss": 0.0708, "lr": 1.2145870973296288e-05, "grad_norm": 3.71875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4847, "epoch": 2, "loss": 0.0012, "lr": 1.2142945239386266e-05, "grad_norm": 0.8203125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 4848, "epoch": 2, "loss": 0.0306, "lr": 1.2140019313200078e-05, "grad_norm": 5.28125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4849, "epoch": 2, "loss": 0.1162, "lr": 1.2137093195000251e-05, "grad_norm": 6.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4850, "epoch": 2, "loss": 0.0289, "lr": 1.213416688504933e-05, "grad_norm": 2.53125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4851, "epoch": 2, "loss": 0.1029, "lr": 1.2131240383609881e-05, "grad_norm": 4.03125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4852, "epoch": 2, "loss": 0.002, "lr": 1.2128313690944487e-05, "grad_norm": 4.71875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.85} {"step": 4853, "epoch": 2, "loss": 0.0004, "lr": 1.212538680731574e-05, "grad_norm": 4.75, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4854, "epoch": 2, "loss": 0.0771, "lr": 1.212245973298626e-05, "grad_norm": 4.84375, "samples_per_sec": 10.44, "gpu_mem_gb": 6.83} {"step": 4855, "epoch": 2, "loss": 0.0036, "lr": 1.2119532468218677e-05, "grad_norm": 5.84375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4856, "epoch": 2, "loss": 0.0446, "lr": 1.2116605013275646e-05, "grad_norm": 3.6875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4857, "epoch": 2, "loss": 0.025, "lr": 1.2113677368419827e-05, "grad_norm": 3.328125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4858, "epoch": 2, "loss": 0.0254, "lr": 1.21107495339139e-05, "grad_norm": 2.046875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 4859, "epoch": 2, "loss": 0.0994, "lr": 1.2107821510020576e-05, "grad_norm": 7.40625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4860, "epoch": 2, "loss": 0.0112, "lr": 1.2104893297002566e-05, "grad_norm": 2.703125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 4861, "epoch": 2, "loss": 0.0003, "lr": 1.2101964895122606e-05, "grad_norm": 6.84375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4862, "epoch": 2, "loss": 0.0005, "lr": 1.2099036304643447e-05, "grad_norm": 4.09375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 4863, "epoch": 2, "loss": 0.0095, "lr": 1.2096107525827859e-05, "grad_norm": 0.78125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4864, "epoch": 2, "loss": 0.0724, "lr": 1.2093178558938625e-05, "grad_norm": 3.046875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4865, "epoch": 2, "loss": 0.1005, "lr": 1.2090249404238548e-05, "grad_norm": 3.734375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4866, "epoch": 2, "loss": 0.0018, "lr": 1.2087320061990443e-05, "grad_norm": 5.65625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4867, "epoch": 2, "loss": 0.0085, "lr": 1.2084390532457155e-05, "grad_norm": 2.90625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4868, "epoch": 2, "loss": 0.0005, "lr": 1.2081460815901529e-05, "grad_norm": 2.625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.85} {"step": 4869, "epoch": 2, "loss": 0.0039, "lr": 1.207853091258644e-05, "grad_norm": 3.4375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4870, "epoch": 2, "loss": 0.104, "lr": 1.2075600822774764e-05, "grad_norm": 4.09375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 4871, "epoch": 2, "loss": 0.0092, "lr": 1.2072670546729415e-05, "grad_norm": 4.59375, "samples_per_sec": 10.58, "gpu_mem_gb": 6.85} {"step": 4872, "epoch": 2, "loss": 0.2589, "lr": 1.2069740084713307e-05, "grad_norm": 8.625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4873, "epoch": 2, "loss": 0.0107, "lr": 1.206680943698938e-05, "grad_norm": 1.59375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4874, "epoch": 2, "loss": 0.0005, "lr": 1.206387860382058e-05, "grad_norm": 4.125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.82} {"step": 4875, "epoch": 2, "loss": 0.0504, "lr": 1.2060947585469885e-05, "grad_norm": 2.921875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4876, "epoch": 2, "loss": 0.0387, "lr": 1.2058016382200274e-05, "grad_norm": 2.453125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.84} {"step": 4877, "epoch": 2, "loss": 0.0018, "lr": 1.2055084994274755e-05, "grad_norm": 3.828125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4878, "epoch": 2, "loss": 0.0333, "lr": 1.2052153421956343e-05, "grad_norm": 3.703125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4879, "epoch": 2, "loss": 0.0007, "lr": 1.204922166550808e-05, "grad_norm": 1.2421875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 4880, "epoch": 2, "loss": 0.0156, "lr": 1.2046289725193013e-05, "grad_norm": 5.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4881, "epoch": 2, "loss": 0.0495, "lr": 1.204335760127421e-05, "grad_norm": 5.34375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 4882, "epoch": 2, "loss": 0.0021, "lr": 1.2040425294014759e-05, "grad_norm": 2.0625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 4883, "epoch": 2, "loss": 0.0002, "lr": 1.2037492803677763e-05, "grad_norm": 4.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4884, "epoch": 2, "loss": 0.0076, "lr": 1.2034560130526341e-05, "grad_norm": 4.15625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4885, "epoch": 2, "loss": 0.341, "lr": 1.2031627274823623e-05, "grad_norm": 7.40625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4886, "epoch": 2, "loss": 0.0004, "lr": 1.202869423683276e-05, "grad_norm": 5.9375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4887, "epoch": 2, "loss": 0.0001, "lr": 1.2025761016816925e-05, "grad_norm": 5.03125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4888, "epoch": 2, "loss": 0.0452, "lr": 1.20228276150393e-05, "grad_norm": 5.34375, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 4889, "epoch": 2, "loss": 0.0748, "lr": 1.2019894031763082e-05, "grad_norm": 2.234375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4890, "epoch": 2, "loss": 0.0059, "lr": 1.2016960267251488e-05, "grad_norm": 2.828125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 4891, "epoch": 2, "loss": 0.0036, "lr": 1.2014026321767754e-05, "grad_norm": 4.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4892, "epoch": 2, "loss": 0.0007, "lr": 1.2011092195575124e-05, "grad_norm": 4.0, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4893, "epoch": 2, "loss": 0.2593, "lr": 1.2008157888936867e-05, "grad_norm": 8.0, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4894, "epoch": 2, "loss": 0.1122, "lr": 1.2005223402116262e-05, "grad_norm": 9.1875, "samples_per_sec": 10.74, "gpu_mem_gb": 6.84} {"step": 4895, "epoch": 2, "loss": 0.002, "lr": 1.2002288735376606e-05, "grad_norm": 2.84375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 4896, "epoch": 2, "loss": 0.0336, "lr": 1.1999353888981218e-05, "grad_norm": 3.796875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4897, "epoch": 2, "loss": 0.0161, "lr": 1.1996418863193416e-05, "grad_norm": 3.6875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 4898, "epoch": 2, "loss": 0.0012, "lr": 1.1993483658276559e-05, "grad_norm": 3.796875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4899, "epoch": 2, "loss": 0.0059, "lr": 1.1990548274494e-05, "grad_norm": 4.71875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4900, "epoch": 2, "loss": 0.0019, "lr": 1.1987612712109123e-05, "grad_norm": 4.15625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4901, "epoch": 2, "loss": 0.0025, "lr": 1.1984676971385314e-05, "grad_norm": 0.7890625, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 4902, "epoch": 2, "loss": 0.1213, "lr": 1.1981741052585987e-05, "grad_norm": 5.90625, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 4903, "epoch": 2, "loss": 0.0009, "lr": 1.1978804955974573e-05, "grad_norm": 12.625, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 4904, "epoch": 2, "loss": 0.0004, "lr": 1.1975868681814506e-05, "grad_norm": 0.5390625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4905, "epoch": 2, "loss": 0.0981, "lr": 1.1972932230369247e-05, "grad_norm": 3.4375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4906, "epoch": 2, "loss": 0.0092, "lr": 1.1969995601902272e-05, "grad_norm": 2.03125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 4907, "epoch": 2, "loss": 0.0001, "lr": 1.1967058796677065e-05, "grad_norm": 3.5, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 4908, "epoch": 2, "loss": 0.0146, "lr": 1.1964121814957136e-05, "grad_norm": 5.34375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 4909, "epoch": 2, "loss": 0.0317, "lr": 1.1961184657006004e-05, "grad_norm": 1.9296875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4910, "epoch": 2, "loss": 0.0094, "lr": 1.1958247323087206e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4911, "epoch": 2, "loss": 0.2767, "lr": 1.19553098134643e-05, "grad_norm": 5.28125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 4912, "epoch": 2, "loss": 0.0065, "lr": 1.1952372128400845e-05, "grad_norm": 2.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 4913, "epoch": 2, "loss": 0.0115, "lr": 1.1949434268160432e-05, "grad_norm": 2.859375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4914, "epoch": 2, "loss": 0.0069, "lr": 1.1946496233006659e-05, "grad_norm": 3.390625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 4915, "epoch": 2, "loss": 0.0006, "lr": 1.1943558023203145e-05, "grad_norm": 2.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4916, "epoch": 2, "loss": 0.0001, "lr": 1.1940619639013518e-05, "grad_norm": 1.8828125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4917, "epoch": 2, "loss": 0.0041, "lr": 1.1937681080701424e-05, "grad_norm": 3.5625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 4918, "epoch": 2, "loss": 0.0216, "lr": 1.1934742348530533e-05, "grad_norm": 7.34375, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83} {"step": 4919, "epoch": 2, "loss": 0.009, "lr": 1.1931803442764511e-05, "grad_norm": 3.75, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 4920, "epoch": 2, "loss": 0.0485, "lr": 1.1928864363667064e-05, "grad_norm": 4.53125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.84} {"step": 4921, "epoch": 2, "loss": 0.1995, "lr": 1.1925925111501893e-05, "grad_norm": 6.3125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 4922, "epoch": 2, "loss": 0.019, "lr": 1.1922985686532726e-05, "grad_norm": 5.34375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 4923, "epoch": 2, "loss": 0.0382, "lr": 1.1920046089023307e-05, "grad_norm": 3.71875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 4924, "epoch": 2, "loss": 0.0011, "lr": 1.1917106319237386e-05, "grad_norm": 6.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 4925, "epoch": 2, "loss": 0.0029, "lr": 1.1914166377438736e-05, "grad_norm": 5.9375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 4926, "epoch": 2, "loss": 0.0408, "lr": 1.191122626389115e-05, "grad_norm": 4.0, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 4927, "epoch": 2, "loss": 0.0188, "lr": 1.1908285978858421e-05, "grad_norm": 1.8046875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 4928, "epoch": 2, "loss": 0.0898, "lr": 1.190534552260437e-05, "grad_norm": 2.5, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 4929, "epoch": 2, "loss": 0.0274, "lr": 1.1902404895392829e-05, "grad_norm": 2.890625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4930, "epoch": 2, "loss": 0.0002, "lr": 1.1899464097487654e-05, "grad_norm": 6.09375, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 4931, "epoch": 2, "loss": 0.0012, "lr": 1.1896523129152697e-05, "grad_norm": 1.5078125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 4932, "epoch": 2, "loss": 0.0029, "lr": 1.1893581990651848e-05, "grad_norm": 2.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4933, "epoch": 2, "loss": 0.1338, "lr": 1.1890640682248993e-05, "grad_norm": 4.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 4934, "epoch": 2, "loss": 0.0004, "lr": 1.1887699204208043e-05, "grad_norm": 3.0625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 4935, "epoch": 2, "loss": 0.2013, "lr": 1.1884757556792928e-05, "grad_norm": 7.90625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 4936, "epoch": 2, "loss": 0.0024, "lr": 1.188181574026758e-05, "grad_norm": 2.59375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4937, "epoch": 2, "loss": 0.0348, "lr": 1.187887375489596e-05, "grad_norm": 3.890625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 4938, "epoch": 2, "loss": 0.0068, "lr": 1.1875931600942035e-05, "grad_norm": 6.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 4939, "epoch": 2, "loss": 0.0243, "lr": 1.187298927866979e-05, "grad_norm": 2.6875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4940, "epoch": 2, "loss": 0.003, "lr": 1.1870046788343228e-05, "grad_norm": 3.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 4941, "epoch": 2, "loss": 0.2042, "lr": 1.1867104130226363e-05, "grad_norm": 8.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4942, "epoch": 2, "loss": 0.0962, "lr": 1.1864161304583228e-05, "grad_norm": 6.40625, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 4943, "epoch": 2, "loss": 0.0003, "lr": 1.1861218311677867e-05, "grad_norm": 2.796875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 4944, "epoch": 2, "loss": 0.0084, "lr": 1.1858275151774334e-05, "grad_norm": 0.96875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4945, "epoch": 2, "loss": 0.0521, "lr": 1.1855331825136718e-05, "grad_norm": 4.0625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 4946, "epoch": 2, "loss": 0.1116, "lr": 1.1852388332029096e-05, "grad_norm": 2.984375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4947, "epoch": 2, "loss": 0.1711, "lr": 1.1849444672715587e-05, "grad_norm": 8.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4948, "epoch": 2, "loss": 0.0275, "lr": 1.1846500847460298e-05, "grad_norm": 2.953125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4949, "epoch": 2, "loss": 0.0016, "lr": 1.1843556856527371e-05, "grad_norm": 4.8125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4950, "epoch": 2, "loss": 0.0015, "lr": 1.1840612700180957e-05, "grad_norm": 3.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 4951, "epoch": 2, "loss": 0.2682, "lr": 1.1837668378685221e-05, "grad_norm": 7.84375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4952, "epoch": 2, "loss": 0.1164, "lr": 1.183472389230434e-05, "grad_norm": 4.375, "samples_per_sec": 10.55, "gpu_mem_gb": 6.85} {"step": 4953, "epoch": 2, "loss": 0.14, "lr": 1.1831779241302507e-05, "grad_norm": 4.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 4954, "epoch": 2, "loss": 0.0012, "lr": 1.1828834425943935e-05, "grad_norm": 2.609375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4955, "epoch": 2, "loss": 0.0577, "lr": 1.1825889446492854e-05, "grad_norm": 3.4375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 4956, "epoch": 2, "loss": 0.0191, "lr": 1.1822944303213486e-05, "grad_norm": 5.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4957, "epoch": 2, "loss": 0.0132, "lr": 1.18199989963701e-05, "grad_norm": 2.765625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 4958, "epoch": 2, "loss": 0.0009, "lr": 1.1817053526226961e-05, "grad_norm": 0.8359375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 4959, "epoch": 2, "loss": 0.0749, "lr": 1.1814107893048348e-05, "grad_norm": 9.3125, "samples_per_sec": 10.66, "gpu_mem_gb": 6.84} {"step": 4960, "epoch": 2, "loss": 0.0725, "lr": 1.1811162097098559e-05, "grad_norm": 8.4375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.84} {"step": 4961, "epoch": 2, "loss": 0.0096, "lr": 1.1808216138641908e-05, "grad_norm": 11.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4962, "epoch": 2, "loss": 0.0007, "lr": 1.1805270017942723e-05, "grad_norm": 3.0, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 4963, "epoch": 2, "loss": 0.0247, "lr": 1.1802323735265339e-05, "grad_norm": 4.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 4964, "epoch": 2, "loss": 0.0937, "lr": 1.1799377290874119e-05, "grad_norm": 6.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4965, "epoch": 2, "loss": 0.0391, "lr": 1.1796430685033434e-05, "grad_norm": 5.8125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4966, "epoch": 2, "loss": 0.0571, "lr": 1.179348391800766e-05, "grad_norm": 4.375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 4967, "epoch": 2, "loss": 0.0004, "lr": 1.1790536990061203e-05, "grad_norm": 4.9375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 4968, "epoch": 2, "loss": 0.0048, "lr": 1.1787589901458474e-05, "grad_norm": 1.6015625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 4969, "epoch": 2, "loss": 0.0412, "lr": 1.1784642652463903e-05, "grad_norm": 4.84375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 4970, "epoch": 2, "loss": 0.0983, "lr": 1.1781695243341933e-05, "grad_norm": 4.6875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 4971, "epoch": 2, "loss": 0.0407, "lr": 1.1778747674357019e-05, "grad_norm": 4.21875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4972, "epoch": 2, "loss": 0.0176, "lr": 1.177579994577363e-05, "grad_norm": 3.828125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 4973, "epoch": 2, "loss": 0.0024, "lr": 1.1772852057856254e-05, "grad_norm": 0.7421875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 4974, "epoch": 2, "loss": 0.1328, "lr": 1.1769904010869396e-05, "grad_norm": 4.0625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 4975, "epoch": 2, "loss": 0.0126, "lr": 1.176695580507756e-05, "grad_norm": 5.71875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 4976, "epoch": 2, "loss": 0.0504, "lr": 1.1764007440745282e-05, "grad_norm": 4.09375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 4977, "epoch": 2, "loss": 0.0261, "lr": 1.1761058918137103e-05, "grad_norm": 5.84375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 4978, "epoch": 2, "loss": 0.0719, "lr": 1.1758110237517575e-05, "grad_norm": 5.15625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 4979, "epoch": 2, "loss": 0.0177, "lr": 1.1755161399151277e-05, "grad_norm": 3.75, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 4980, "epoch": 2, "loss": 0.064, "lr": 1.1752212403302785e-05, "grad_norm": 3.21875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 4981, "epoch": 2, "loss": 0.0272, "lr": 1.1749263250236704e-05, "grad_norm": 2.90625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4982, "epoch": 2, "loss": 0.0788, "lr": 1.174631394021765e-05, "grad_norm": 4.28125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.85} {"step": 4983, "epoch": 2, "loss": 0.0013, "lr": 1.1743364473510239e-05, "grad_norm": 3.15625, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 4984, "epoch": 2, "loss": 0.1551, "lr": 1.1740414850379127e-05, "grad_norm": 10.0625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 4985, "epoch": 2, "loss": 0.0009, "lr": 1.1737465071088961e-05, "grad_norm": 4.34375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 4986, "epoch": 2, "loss": 0.1396, "lr": 1.1734515135904412e-05, "grad_norm": 4.84375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 4987, "epoch": 2, "loss": 0.0192, "lr": 1.1731565045090161e-05, "grad_norm": 2.78125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 4988, "epoch": 2, "loss": 0.1979, "lr": 1.1728614798910911e-05, "grad_norm": 11.25, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 4989, "epoch": 2, "loss": 0.0505, "lr": 1.1725664397631369e-05, "grad_norm": 5.875, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 4990, "epoch": 2, "loss": 0.0585, "lr": 1.1722713841516262e-05, "grad_norm": 2.484375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 4991, "epoch": 2, "loss": 0.0247, "lr": 1.1719763130830328e-05, "grad_norm": 3.28125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4992, "epoch": 2, "loss": 0.007, "lr": 1.1716812265838325e-05, "grad_norm": 2.1875, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 4993, "epoch": 2, "loss": 0.0201, "lr": 1.1713861246805013e-05, "grad_norm": 6.65625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 4994, "epoch": 2, "loss": 0.0535, "lr": 1.1710910073995174e-05, "grad_norm": 3.046875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 4995, "epoch": 2, "loss": 0.0016, "lr": 1.1707958747673605e-05, "grad_norm": 3.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 4996, "epoch": 2, "loss": 0.0322, "lr": 1.1705007268105115e-05, "grad_norm": 3.71875, "samples_per_sec": 10.35, "gpu_mem_gb": 6.82} {"step": 4997, "epoch": 2, "loss": 0.0036, "lr": 1.1702055635554526e-05, "grad_norm": 4.15625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 4998, "epoch": 2, "loss": 0.0017, "lr": 1.1699103850286668e-05, "grad_norm": 3.96875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 4999, "epoch": 2, "loss": 0.2073, "lr": 1.1696151912566398e-05, "grad_norm": 6.15625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 5000, "epoch": 2, "loss": 0.0635, "lr": 1.1693199822658574e-05, "grad_norm": 3.109375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 5001, "epoch": 2, "loss": 0.1737, "lr": 1.1690247580828075e-05, "grad_norm": 8.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.84} {"step": 5002, "epoch": 2, "loss": 0.0649, "lr": 1.1687295187339794e-05, "grad_norm": 9.25, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5003, "epoch": 2, "loss": 0.1715, "lr": 1.168434264245863e-05, "grad_norm": 6.6875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 5004, "epoch": 2, "loss": 0.0252, "lr": 1.1681389946449504e-05, "grad_norm": 3.75, "samples_per_sec": 8.9, "gpu_mem_gb": 6.83} {"step": 5005, "epoch": 2, "loss": 0.0012, "lr": 1.1678437099577347e-05, "grad_norm": 4.59375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 5006, "epoch": 2, "loss": 0.001, "lr": 1.1675484102107103e-05, "grad_norm": 4.96875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5007, "epoch": 2, "loss": 0.0008, "lr": 1.1672530954303726e-05, "grad_norm": 4.78125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 5008, "epoch": 2, "loss": 0.034, "lr": 1.1669577656432195e-05, "grad_norm": 4.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5009, "epoch": 2, "loss": 0.0046, "lr": 1.1666624208757496e-05, "grad_norm": 2.671875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.85} {"step": 5010, "epoch": 2, "loss": 0.0264, "lr": 1.166367061154462e-05, "grad_norm": 3.953125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5011, "epoch": 2, "loss": 0.1194, "lr": 1.1660716865058581e-05, "grad_norm": 8.4375, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 5012, "epoch": 2, "loss": 0.0023, "lr": 1.1657762969564408e-05, "grad_norm": 1.1015625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 5013, "epoch": 2, "loss": 0.0027, "lr": 1.1654808925327141e-05, "grad_norm": 3.1875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5014, "epoch": 2, "loss": 0.0242, "lr": 1.1651854732611829e-05, "grad_norm": 1.78125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5015, "epoch": 2, "loss": 0.0208, "lr": 1.1648900391683533e-05, "grad_norm": 2.53125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5016, "epoch": 2, "loss": 0.0006, "lr": 1.164594590280734e-05, "grad_norm": 2.265625, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 5017, "epoch": 2, "loss": 0.0121, "lr": 1.1642991266248338e-05, "grad_norm": 3.59375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.83} {"step": 5018, "epoch": 2, "loss": 0.1368, "lr": 1.1640036482271635e-05, "grad_norm": 6.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 5019, "epoch": 2, "loss": 0.1178, "lr": 1.1637081551142347e-05, "grad_norm": 6.28125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5020, "epoch": 2, "loss": 0.0004, "lr": 1.1634126473125605e-05, "grad_norm": 6.75, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 5021, "epoch": 2, "loss": 0.0144, "lr": 1.1631171248486556e-05, "grad_norm": 1.8125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 5022, "epoch": 2, "loss": 0.0002, "lr": 1.1628215877490359e-05, "grad_norm": 3.40625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 5023, "epoch": 2, "loss": 0.0623, "lr": 1.162526036040218e-05, "grad_norm": 4.125, "samples_per_sec": 10.5, "gpu_mem_gb": 6.84} {"step": 5024, "epoch": 2, "loss": 0.108, "lr": 1.1622304697487212e-05, "grad_norm": 6.65625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5025, "epoch": 2, "loss": 0.0615, "lr": 1.1619348889010643e-05, "grad_norm": 4.78125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5026, "epoch": 2, "loss": 0.0134, "lr": 1.161639293523769e-05, "grad_norm": 3.75, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5027, "epoch": 2, "loss": 0.0006, "lr": 1.1613436836433571e-05, "grad_norm": 6.5, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5028, "epoch": 2, "loss": 0.0004, "lr": 1.161048059286353e-05, "grad_norm": 6.8125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 5029, "epoch": 2, "loss": 0.1946, "lr": 1.1607524204792813e-05, "grad_norm": 4.375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.85} {"step": 5030, "epoch": 2, "loss": 0.1273, "lr": 1.160456767248668e-05, "grad_norm": 3.9375, "samples_per_sec": 10.74, "gpu_mem_gb": 6.82} {"step": 5031, "epoch": 2, "loss": 0.0002, "lr": 1.1601610996210408e-05, "grad_norm": 5.0625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 5032, "epoch": 2, "loss": 0.006, "lr": 1.1598654176229288e-05, "grad_norm": 4.09375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5033, "epoch": 2, "loss": 0.017, "lr": 1.1595697212808617e-05, "grad_norm": 2.421875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 5034, "epoch": 2, "loss": 0.0084, "lr": 1.1592740106213711e-05, "grad_norm": 2.671875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 5035, "epoch": 2, "loss": 0.1728, "lr": 1.1589782856709895e-05, "grad_norm": 9.1875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5036, "epoch": 2, "loss": 0.082, "lr": 1.1586825464562515e-05, "grad_norm": 2.859375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 5037, "epoch": 2, "loss": 0.2583, "lr": 1.1583867930036917e-05, "grad_norm": 7.40625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5038, "epoch": 2, "loss": 0.1943, "lr": 1.1580910253398468e-05, "grad_norm": 5.28125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5039, "epoch": 2, "loss": 0.1168, "lr": 1.157795243491255e-05, "grad_norm": 6.0, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5040, "epoch": 2, "loss": 0.1192, "lr": 1.1574994474844545e-05, "grad_norm": 5.125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 5041, "epoch": 2, "loss": 0.049, "lr": 1.1572036373459868e-05, "grad_norm": 3.40625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5042, "epoch": 2, "loss": 0.0607, "lr": 1.1569078131023924e-05, "grad_norm": 3.859375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5043, "epoch": 2, "loss": 0.0174, "lr": 1.1566119747802151e-05, "grad_norm": 2.953125, "samples_per_sec": 10.58, "gpu_mem_gb": 6.82} {"step": 5044, "epoch": 2, "loss": 0.0006, "lr": 1.1563161224059984e-05, "grad_norm": 5.5625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5045, "epoch": 2, "loss": 0.0118, "lr": 1.1560202560062883e-05, "grad_norm": 2.203125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 5046, "epoch": 2, "loss": 0.0002, "lr": 1.1557243756076311e-05, "grad_norm": 0.431640625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 5047, "epoch": 2, "loss": 0.0007, "lr": 1.1554284812365744e-05, "grad_norm": 3.53125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 5048, "epoch": 2, "loss": 0.0001, "lr": 1.1551325729196682e-05, "grad_norm": 0.2001953125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5049, "epoch": 2, "loss": 0.0213, "lr": 1.1548366506834624e-05, "grad_norm": 4.96875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5050, "epoch": 2, "loss": 0.0294, "lr": 1.1545407145545084e-05, "grad_norm": 3.140625, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 5051, "epoch": 2, "loss": 0.1871, "lr": 1.1542447645593598e-05, "grad_norm": 6.28125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.84} {"step": 5052, "epoch": 2, "loss": 0.1026, "lr": 1.1539488007245704e-05, "grad_norm": 3.4375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 5053, "epoch": 2, "loss": 0.3861, "lr": 1.1536528230766958e-05, "grad_norm": 10.6875, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 5054, "epoch": 2, "loss": 0.0626, "lr": 1.153356831642292e-05, "grad_norm": 1.5625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 5055, "epoch": 2, "loss": 0.0194, "lr": 1.153060826447918e-05, "grad_norm": 1.4921875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 5056, "epoch": 2, "loss": 0.0293, "lr": 1.152764807520132e-05, "grad_norm": 4.5625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 5057, "epoch": 2, "loss": 0.042, "lr": 1.1524687748854948e-05, "grad_norm": 3.5625, "samples_per_sec": 10.46, "gpu_mem_gb": 6.82} {"step": 5058, "epoch": 2, "loss": 0.1634, "lr": 1.1521727285705677e-05, "grad_norm": 7.40625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5059, "epoch": 2, "loss": 0.0001, "lr": 1.1518766686019139e-05, "grad_norm": 4.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5060, "epoch": 2, "loss": 0.0037, "lr": 1.1515805950060969e-05, "grad_norm": 1.765625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 5061, "epoch": 2, "loss": 0.0012, "lr": 1.1512845078096826e-05, "grad_norm": 0.408203125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 5062, "epoch": 2, "loss": 0.0597, "lr": 1.1509884070392369e-05, "grad_norm": 5.1875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5063, "epoch": 2, "loss": 0.2363, "lr": 1.150692292721328e-05, "grad_norm": 6.65625, "samples_per_sec": 10.51, "gpu_mem_gb": 6.84} {"step": 5064, "epoch": 2, "loss": 0.0141, "lr": 1.1503961648825244e-05, "grad_norm": 1.5625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 5065, "epoch": 2, "loss": 0.272, "lr": 1.1501000235493966e-05, "grad_norm": 16.375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5066, "epoch": 2, "loss": 0.0013, "lr": 1.1498038687485155e-05, "grad_norm": 8.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5067, "epoch": 2, "loss": 0.0731, "lr": 1.1495077005064539e-05, "grad_norm": 6.03125, "samples_per_sec": 10.45, "gpu_mem_gb": 6.83} {"step": 5068, "epoch": 2, "loss": 0.0272, "lr": 1.1492115188497857e-05, "grad_norm": 5.40625, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 5069, "epoch": 2, "loss": 0.0444, "lr": 1.1489153238050854e-05, "grad_norm": 3.84375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 5070, "epoch": 2, "loss": 0.1379, "lr": 1.1486191153989294e-05, "grad_norm": 4.65625, "samples_per_sec": 10.41, "gpu_mem_gb": 6.83} {"step": 5071, "epoch": 2, "loss": 0.0013, "lr": 1.1483228936578953e-05, "grad_norm": 0.74609375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5072, "epoch": 2, "loss": 0.1149, "lr": 1.1480266586085616e-05, "grad_norm": 4.71875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5073, "epoch": 2, "loss": 0.0529, "lr": 1.1477304102775075e-05, "grad_norm": 3.75, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5074, "epoch": 2, "loss": 0.1433, "lr": 1.1474341486913146e-05, "grad_norm": 5.65625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 5075, "epoch": 2, "loss": 0.0001, "lr": 1.1471378738765648e-05, "grad_norm": 4.21875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5076, "epoch": 2, "loss": 0.2067, "lr": 1.1468415858598413e-05, "grad_norm": 7.5625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5077, "epoch": 2, "loss": 0.0409, "lr": 1.1465452846677282e-05, "grad_norm": 6.3125, "samples_per_sec": 10.47, "gpu_mem_gb": 6.82} {"step": 5078, "epoch": 2, "loss": 0.0002, "lr": 1.1462489703268123e-05, "grad_norm": 9.75, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 5079, "epoch": 2, "loss": 0.0219, "lr": 1.1459526428636795e-05, "grad_norm": 2.484375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 5080, "epoch": 2, "loss": 0.0318, "lr": 1.1456563023049183e-05, "grad_norm": 2.0625, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5081, "epoch": 2, "loss": 0.0028, "lr": 1.1453599486771173e-05, "grad_norm": 2.046875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 5082, "epoch": 2, "loss": 0.0778, "lr": 1.145063582006868e-05, "grad_norm": 5.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5083, "epoch": 2, "loss": 0.0027, "lr": 1.1447672023207612e-05, "grad_norm": 6.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5084, "epoch": 2, "loss": 0.0215, "lr": 1.1444708096453894e-05, "grad_norm": 3.8125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 5085, "epoch": 2, "loss": 0.0014, "lr": 1.1441744040073469e-05, "grad_norm": 3.234375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5086, "epoch": 2, "loss": 0.0029, "lr": 1.1438779854332286e-05, "grad_norm": 3.96875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5087, "epoch": 2, "loss": 0.3415, "lr": 1.1435815539496312e-05, "grad_norm": 6.25, "samples_per_sec": 10.73, "gpu_mem_gb": 6.84} {"step": 5088, "epoch": 2, "loss": 0.1799, "lr": 1.1432851095831511e-05, "grad_norm": 7.5625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5089, "epoch": 2, "loss": 0.1687, "lr": 1.1429886523603874e-05, "grad_norm": 6.25, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5090, "epoch": 2, "loss": 0.0288, "lr": 1.1426921823079402e-05, "grad_norm": 4.9375, "samples_per_sec": 10.4, "gpu_mem_gb": 6.82} {"step": 5091, "epoch": 2, "loss": 0.0002, "lr": 1.1423956994524095e-05, "grad_norm": 3.3125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 5092, "epoch": 2, "loss": 0.1075, "lr": 1.1420992038203979e-05, "grad_norm": 3.734375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5093, "epoch": 2, "loss": 0.0706, "lr": 1.1418026954385082e-05, "grad_norm": 3.734375, "samples_per_sec": 10.64, "gpu_mem_gb": 6.83} {"step": 5094, "epoch": 2, "loss": 0.1184, "lr": 1.1415061743333446e-05, "grad_norm": 6.34375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.84} {"step": 5095, "epoch": 2, "loss": 0.0016, "lr": 1.1412096405315131e-05, "grad_norm": 2.984375, "samples_per_sec": 10.86, "gpu_mem_gb": 6.82} {"step": 5096, "epoch": 2, "loss": 0.0008, "lr": 1.1409130940596196e-05, "grad_norm": 3.203125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5097, "epoch": 2, "loss": 0.0017, "lr": 1.140616534944272e-05, "grad_norm": 0.1455078125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 5098, "epoch": 2, "loss": 0.0954, "lr": 1.140319963212079e-05, "grad_norm": 7.5625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5099, "epoch": 2, "loss": 0.0069, "lr": 1.1400233788896509e-05, "grad_norm": 1.5859375, "samples_per_sec": 10.99, "gpu_mem_gb": 6.85} {"step": 5100, "epoch": 2, "loss": 0.0011, "lr": 1.1397267820035986e-05, "grad_norm": 2.171875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5101, "epoch": 2, "loss": 0.0742, "lr": 1.1394301725805339e-05, "grad_norm": 5.875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 5102, "epoch": 2, "loss": 0.0267, "lr": 1.139133550647071e-05, "grad_norm": 8.25, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5103, "epoch": 2, "loss": 0.0034, "lr": 1.1388369162298236e-05, "grad_norm": 4.09375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 5104, "epoch": 2, "loss": 0.0739, "lr": 1.1385402693554077e-05, "grad_norm": 4.15625, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 5105, "epoch": 2, "loss": 0.0005, "lr": 1.13824361005044e-05, "grad_norm": 0.328125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5106, "epoch": 2, "loss": 0.0167, "lr": 1.1379469383415378e-05, "grad_norm": 5.4375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.85} {"step": 5107, "epoch": 2, "loss": 0.001, "lr": 1.1376502542553208e-05, "grad_norm": 0.384765625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5108, "epoch": 2, "loss": 0.0498, "lr": 1.1373535578184083e-05, "grad_norm": 5.0625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.82} {"step": 5109, "epoch": 2, "loss": 0.2263, "lr": 1.1370568490574219e-05, "grad_norm": 5.875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5110, "epoch": 2, "loss": 0.0354, "lr": 1.136760127998984e-05, "grad_norm": 4.0, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 5111, "epoch": 2, "loss": 0.1036, "lr": 1.1364633946697173e-05, "grad_norm": 6.8125, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 5112, "epoch": 2, "loss": 0.0393, "lr": 1.1361666490962468e-05, "grad_norm": 1.65625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5113, "epoch": 2, "loss": 0.0073, "lr": 1.1358698913051977e-05, "grad_norm": 2.703125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5114, "epoch": 2, "loss": 0.0782, "lr": 1.1355731213231971e-05, "grad_norm": 3.90625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 5115, "epoch": 2, "loss": 0.0503, "lr": 1.1352763391768725e-05, "grad_norm": 2.640625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5116, "epoch": 2, "loss": 0.0002, "lr": 1.1349795448928525e-05, "grad_norm": 0.1982421875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.85} {"step": 5117, "epoch": 2, "loss": 0.0006, "lr": 1.1346827384977674e-05, "grad_norm": 6.34375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.85} {"step": 5118, "epoch": 2, "loss": 0.0009, "lr": 1.1343859200182479e-05, "grad_norm": 1.0234375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 5119, "epoch": 2, "loss": 0.0003, "lr": 1.1340890894809264e-05, "grad_norm": 2.828125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 5120, "epoch": 2, "loss": 0.1609, "lr": 1.1337922469124358e-05, "grad_norm": 7.625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5121, "epoch": 2, "loss": 0.0023, "lr": 1.1334953923394105e-05, "grad_norm": 7.625, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 5122, "epoch": 2, "loss": 0.0001, "lr": 1.1331985257884861e-05, "grad_norm": 3.421875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 5123, "epoch": 2, "loss": 0.0188, "lr": 1.1329016472862983e-05, "grad_norm": 5.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 5124, "epoch": 2, "loss": 0.0002, "lr": 1.1326047568594852e-05, "grad_norm": 2.234375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 5125, "epoch": 2, "loss": 0.0002, "lr": 1.132307854534685e-05, "grad_norm": 3.421875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5126, "epoch": 2, "loss": 0.0637, "lr": 1.1320109403385377e-05, "grad_norm": 2.078125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5127, "epoch": 2, "loss": 0.0764, "lr": 1.1317140142976839e-05, "grad_norm": 2.765625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5128, "epoch": 2, "loss": 0.0477, "lr": 1.1314170764387647e-05, "grad_norm": 3.828125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5129, "epoch": 2, "loss": 0.0363, "lr": 1.1311201267884238e-05, "grad_norm": 5.03125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 5130, "epoch": 2, "loss": 0.0008, "lr": 1.1308231653733043e-05, "grad_norm": 1.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 5131, "epoch": 2, "loss": 0.0366, "lr": 1.130526192220052e-05, "grad_norm": 2.484375, "samples_per_sec": 10.31, "gpu_mem_gb": 6.83} {"step": 5132, "epoch": 2, "loss": 0.0149, "lr": 1.1302292073553119e-05, "grad_norm": 0.384765625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5133, "epoch": 2, "loss": 0.0015, "lr": 1.1299322108057313e-05, "grad_norm": 1.890625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5134, "epoch": 2, "loss": 0.0164, "lr": 1.129635202597959e-05, "grad_norm": 3.3125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 5135, "epoch": 2, "loss": 0.0002, "lr": 1.1293381827586432e-05, "grad_norm": 2.03125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5136, "epoch": 2, "loss": 0.0001, "lr": 1.1290411513144343e-05, "grad_norm": 4.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5137, "epoch": 2, "loss": 0.0016, "lr": 1.1287441082919842e-05, "grad_norm": 4.3125, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 5138, "epoch": 2, "loss": 0.0007, "lr": 1.1284470537179442e-05, "grad_norm": 4.375, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 5139, "epoch": 2, "loss": 0.0209, "lr": 1.128149987618968e-05, "grad_norm": 3.046875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5140, "epoch": 2, "loss": 0.0249, "lr": 1.1278529100217098e-05, "grad_norm": 1.4140625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 5141, "epoch": 2, "loss": 0.0496, "lr": 1.127555820952825e-05, "grad_norm": 4.15625, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5142, "epoch": 2, "loss": 0.0557, "lr": 1.12725872043897e-05, "grad_norm": 1.5390625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 5143, "epoch": 2, "loss": 0.0023, "lr": 1.1269616085068026e-05, "grad_norm": 5.03125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 5144, "epoch": 2, "loss": 0.1219, "lr": 1.1266644851829802e-05, "grad_norm": 4.90625, "samples_per_sec": 10.54, "gpu_mem_gb": 6.82} {"step": 5145, "epoch": 2, "loss": 0.0022, "lr": 1.126367350494163e-05, "grad_norm": 8.5, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 5146, "epoch": 2, "loss": 0.0973, "lr": 1.1260702044670115e-05, "grad_norm": 5.5625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5147, "epoch": 2, "loss": 0.001, "lr": 1.1257730471281869e-05, "grad_norm": 1.90625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5148, "epoch": 2, "loss": 0.0006, "lr": 1.1254758785043516e-05, "grad_norm": 3.234375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 5149, "epoch": 2, "loss": 0.0002, "lr": 1.1251786986221697e-05, "grad_norm": 5.90625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5150, "epoch": 2, "loss": 0.0868, "lr": 1.1248815075083051e-05, "grad_norm": 5.5, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5151, "epoch": 2, "loss": 0.0006, "lr": 1.1245843051894237e-05, "grad_norm": 2.84375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.82} {"step": 5152, "epoch": 2, "loss": 0.0677, "lr": 1.1242870916921917e-05, "grad_norm": 9.875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.85} {"step": 5153, "epoch": 2, "loss": 0.0766, "lr": 1.123989867043277e-05, "grad_norm": 6.09375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5154, "epoch": 2, "loss": 0.0006, "lr": 1.123692631269348e-05, "grad_norm": 3.5625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5155, "epoch": 2, "loss": 0.0155, "lr": 1.123395384397074e-05, "grad_norm": 4.09375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5156, "epoch": 2, "loss": 0.0435, "lr": 1.1230981264531255e-05, "grad_norm": 2.046875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5157, "epoch": 2, "loss": 0.0007, "lr": 1.1228008574641746e-05, "grad_norm": 3.296875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5158, "epoch": 2, "loss": 0.0358, "lr": 1.1225035774568937e-05, "grad_norm": 1.765625, "samples_per_sec": 10.46, "gpu_mem_gb": 6.83} {"step": 5159, "epoch": 2, "loss": 0.0082, "lr": 1.1222062864579557e-05, "grad_norm": 3.265625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 5160, "epoch": 2, "loss": 0.1984, "lr": 1.1219089844940356e-05, "grad_norm": 6.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5161, "epoch": 2, "loss": 0.0145, "lr": 1.1216116715918091e-05, "grad_norm": 5.6875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 5162, "epoch": 2, "loss": 0.0952, "lr": 1.1213143477779517e-05, "grad_norm": 3.15625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5163, "epoch": 2, "loss": 0.0037, "lr": 1.1210170130791417e-05, "grad_norm": 4.71875, "samples_per_sec": 10.99, "gpu_mem_gb": 6.82} {"step": 5164, "epoch": 2, "loss": 0.0011, "lr": 1.1207196675220575e-05, "grad_norm": 3.84375, "samples_per_sec": 10.67, "gpu_mem_gb": 6.82} {"step": 5165, "epoch": 2, "loss": 0.1019, "lr": 1.1204223111333778e-05, "grad_norm": 6.40625, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 5166, "epoch": 2, "loss": 0.0152, "lr": 1.120124943939784e-05, "grad_norm": 2.71875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 5167, "epoch": 2, "loss": 0.0872, "lr": 1.119827565967956e-05, "grad_norm": 5.96875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 5168, "epoch": 2, "loss": 0.0377, "lr": 1.1195301772445778e-05, "grad_norm": 3.40625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 5169, "epoch": 2, "loss": 0.0956, "lr": 1.1192327777963313e-05, "grad_norm": 3.078125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.84} {"step": 5170, "epoch": 2, "loss": 0.0105, "lr": 1.1189353676499013e-05, "grad_norm": 0.7421875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 5171, "epoch": 2, "loss": 0.1217, "lr": 1.1186379468319728e-05, "grad_norm": 4.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5172, "epoch": 2, "loss": 0.0445, "lr": 1.118340515369232e-05, "grad_norm": 6.9375, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 5173, "epoch": 2, "loss": 0.0006, "lr": 1.1180430732883663e-05, "grad_norm": 1.6640625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5174, "epoch": 2, "loss": 0.0015, "lr": 1.117745620616063e-05, "grad_norm": 2.546875, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 5175, "epoch": 2, "loss": 0.1965, "lr": 1.117448157379012e-05, "grad_norm": 7.3125, "samples_per_sec": 10.68, "gpu_mem_gb": 6.84} {"step": 5176, "epoch": 2, "loss": 0.0515, "lr": 1.1171506836039025e-05, "grad_norm": 6.25, "samples_per_sec": 10.95, "gpu_mem_gb": 6.84} {"step": 5177, "epoch": 2, "loss": 0.0784, "lr": 1.1168531993174259e-05, "grad_norm": 3.359375, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 5178, "epoch": 2, "loss": 0.0031, "lr": 1.1165557045462738e-05, "grad_norm": 6.96875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.82} {"step": 5179, "epoch": 2, "loss": 0.0003, "lr": 1.1162581993171391e-05, "grad_norm": 3.453125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5180, "epoch": 2, "loss": 0.0265, "lr": 1.1159606836567151e-05, "grad_norm": 1.609375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 5181, "epoch": 2, "loss": 0.0014, "lr": 1.1156631575916972e-05, "grad_norm": 3.59375, "samples_per_sec": 10.69, "gpu_mem_gb": 6.83} {"step": 5182, "epoch": 2, "loss": 0.0, "lr": 1.1153656211487802e-05, "grad_norm": 5.0, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5183, "epoch": 2, "loss": 0.0013, "lr": 1.115068074354661e-05, "grad_norm": 1.9375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5184, "epoch": 2, "loss": 0.2105, "lr": 1.114770517236037e-05, "grad_norm": 6.96875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5185, "epoch": 2, "loss": 0.002, "lr": 1.114472949819607e-05, "grad_norm": 5.34375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 5186, "epoch": 2, "loss": 0.0487, "lr": 1.1141753721320697e-05, "grad_norm": 5.875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5187, "epoch": 2, "loss": 0.044, "lr": 1.113877784200125e-05, "grad_norm": 2.734375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 5188, "epoch": 2, "loss": 0.0018, "lr": 1.113580186050475e-05, "grad_norm": 5.375, "samples_per_sec": 10.57, "gpu_mem_gb": 6.83} {"step": 5189, "epoch": 2, "loss": 0.1701, "lr": 1.113282577709821e-05, "grad_norm": 6.9375, "samples_per_sec": 10.68, "gpu_mem_gb": 6.83} {"step": 5190, "epoch": 2, "loss": 0.0006, "lr": 1.1129849592048665e-05, "grad_norm": 4.375, "samples_per_sec": 10.89, "gpu_mem_gb": 6.85} {"step": 5191, "epoch": 2, "loss": 0.0194, "lr": 1.112687330562315e-05, "grad_norm": 6.25, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5192, "epoch": 2, "loss": 0.0013, "lr": 1.1123896918088713e-05, "grad_norm": 4.71875, "samples_per_sec": 10.21, "gpu_mem_gb": 6.83} {"step": 5193, "epoch": 2, "loss": 0.0033, "lr": 1.1120920429712415e-05, "grad_norm": 2.046875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5194, "epoch": 2, "loss": 0.1959, "lr": 1.1117943840761315e-05, "grad_norm": 4.96875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5195, "epoch": 2, "loss": 0.0844, "lr": 1.1114967151502494e-05, "grad_norm": 3.3125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 5196, "epoch": 2, "loss": 0.001, "lr": 1.1111990362203034e-05, "grad_norm": 4.03125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5197, "epoch": 2, "loss": 0.0004, "lr": 1.1109013473130026e-05, "grad_norm": 2.1875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5198, "epoch": 2, "loss": 0.0425, "lr": 1.1106036484550578e-05, "grad_norm": 3.5, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 5199, "epoch": 2, "loss": 0.0019, "lr": 1.1103059396731791e-05, "grad_norm": 4.40625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 5200, "epoch": 2, "loss": 0.008, "lr": 1.1100082209940795e-05, "grad_norm": 4.8125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.85} {"step": 5201, "epoch": 2, "loss": 0.026, "lr": 1.109710492444471e-05, "grad_norm": 3.96875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 5202, "epoch": 2, "loss": 0.0482, "lr": 1.1094127540510684e-05, "grad_norm": 4.3125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 5203, "epoch": 2, "loss": 0.0018, "lr": 1.109115005840585e-05, "grad_norm": 1.53125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 5204, "epoch": 2, "loss": 0.0133, "lr": 1.1088172478397375e-05, "grad_norm": 3.75, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5205, "epoch": 2, "loss": 0.0003, "lr": 1.1085194800752418e-05, "grad_norm": 1.203125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.82} {"step": 5206, "epoch": 2, "loss": 0.0002, "lr": 1.1082217025738148e-05, "grad_norm": 1.6015625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5207, "epoch": 2, "loss": 0.0007, "lr": 1.1079239153621753e-05, "grad_norm": 4.28125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5208, "epoch": 2, "loss": 0.0005, "lr": 1.107626118467042e-05, "grad_norm": 2.546875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5209, "epoch": 2, "loss": 0.003, "lr": 1.1073283119151345e-05, "grad_norm": 2.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5210, "epoch": 2, "loss": 0.0006, "lr": 1.1070304957331742e-05, "grad_norm": 3.859375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5211, "epoch": 2, "loss": 0.005, "lr": 1.1067326699478821e-05, "grad_norm": 6.875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 5212, "epoch": 2, "loss": 0.0017, "lr": 1.1064348345859815e-05, "grad_norm": 0.380859375, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 5213, "epoch": 2, "loss": 0.0008, "lr": 1.1061369896741949e-05, "grad_norm": 0.2734375, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5214, "epoch": 2, "loss": 0.158, "lr": 1.1058391352392465e-05, "grad_norm": 5.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5215, "epoch": 2, "loss": 0.0002, "lr": 1.1055412713078619e-05, "grad_norm": 5.84375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 5216, "epoch": 2, "loss": 0.1557, "lr": 1.1052433979067667e-05, "grad_norm": 8.125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5217, "epoch": 2, "loss": 0.0413, "lr": 1.1049455150626878e-05, "grad_norm": 5.96875, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5218, "epoch": 2, "loss": 0.4131, "lr": 1.1046476228023524e-05, "grad_norm": 6.46875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.84} {"step": 5219, "epoch": 2, "loss": 0.0005, "lr": 1.1043497211524891e-05, "grad_norm": 4.90625, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 5220, "epoch": 2, "loss": 0.0755, "lr": 1.1040518101398277e-05, "grad_norm": 4.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5221, "epoch": 2, "loss": 0.026, "lr": 1.1037538897910975e-05, "grad_norm": 2.265625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5222, "epoch": 2, "loss": 0.0001, "lr": 1.1034559601330301e-05, "grad_norm": 4.4375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 5223, "epoch": 2, "loss": 0.0007, "lr": 1.103158021192357e-05, "grad_norm": 1.203125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5224, "epoch": 2, "loss": 0.2091, "lr": 1.102860072995811e-05, "grad_norm": 11.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5225, "epoch": 2, "loss": 0.1713, "lr": 1.1025621155701253e-05, "grad_norm": 5.0625, "samples_per_sec": 10.66, "gpu_mem_gb": 6.84} {"step": 5226, "epoch": 2, "loss": 0.0619, "lr": 1.1022641489420342e-05, "grad_norm": 2.65625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 5227, "epoch": 2, "loss": 0.2145, "lr": 1.1019661731382733e-05, "grad_norm": 5.375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 5228, "epoch": 2, "loss": 0.0741, "lr": 1.101668188185578e-05, "grad_norm": 6.5, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5229, "epoch": 2, "loss": 0.0005, "lr": 1.1013701941106854e-05, "grad_norm": 2.59375, "samples_per_sec": 8.74, "gpu_mem_gb": 6.82} {"step": 5230, "epoch": 2, "loss": 0.0199, "lr": 1.1010721909403326e-05, "grad_norm": 4.84375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5231, "epoch": 2, "loss": 0.0288, "lr": 1.1007741787012584e-05, "grad_norm": 5.84375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.84} {"step": 5232, "epoch": 2, "loss": 0.0007, "lr": 1.100476157420202e-05, "grad_norm": 3.828125, "samples_per_sec": 10.54, "gpu_mem_gb": 6.83} {"step": 5233, "epoch": 2, "loss": 0.0007, "lr": 1.1001781271239033e-05, "grad_norm": 0.51171875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 5234, "epoch": 2, "loss": 0.171, "lr": 1.0998800878391032e-05, "grad_norm": 9.375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.85} {"step": 5235, "epoch": 2, "loss": 0.0354, "lr": 1.0995820395925431e-05, "grad_norm": 2.6875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5236, "epoch": 2, "loss": 0.0201, "lr": 1.0992839824109658e-05, "grad_norm": 3.875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 5237, "epoch": 2, "loss": 0.0009, "lr": 1.0989859163211146e-05, "grad_norm": 4.1875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5238, "epoch": 2, "loss": 0.1769, "lr": 1.0986878413497325e-05, "grad_norm": 4.875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 5239, "epoch": 2, "loss": 0.0031, "lr": 1.0983897575235658e-05, "grad_norm": 0.345703125, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5240, "epoch": 2, "loss": 0.0015, "lr": 1.0980916648693593e-05, "grad_norm": 8.375, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 5241, "epoch": 2, "loss": 0.0741, "lr": 1.0977935634138595e-05, "grad_norm": 6.46875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 5242, "epoch": 2, "loss": 0.0205, "lr": 1.0974954531838135e-05, "grad_norm": 1.4453125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.83} {"step": 5243, "epoch": 2, "loss": 0.0079, "lr": 1.0971973342059699e-05, "grad_norm": 6.09375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5244, "epoch": 2, "loss": 0.0016, "lr": 1.096899206507077e-05, "grad_norm": 5.28125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5245, "epoch": 2, "loss": 0.0173, "lr": 1.0966010701138841e-05, "grad_norm": 0.953125, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 5246, "epoch": 2, "loss": 0.0868, "lr": 1.0963029250531418e-05, "grad_norm": 3.34375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 5247, "epoch": 2, "loss": 0.0354, "lr": 1.0960047713516018e-05, "grad_norm": 3.546875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5248, "epoch": 2, "loss": 0.0985, "lr": 1.0957066090360151e-05, "grad_norm": 6.09375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 5249, "epoch": 2, "loss": 0.0264, "lr": 1.095408438133135e-05, "grad_norm": 3.390625, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 5250, "epoch": 2, "loss": 0.0002, "lr": 1.0951102586697144e-05, "grad_norm": 5.8125, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 5251, "epoch": 2, "loss": 0.0351, "lr": 1.0948120706725081e-05, "grad_norm": 2.3125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 5252, "epoch": 2, "loss": 0.0999, "lr": 1.0945138741682706e-05, "grad_norm": 5.59375, "samples_per_sec": 10.45, "gpu_mem_gb": 6.84} {"step": 5253, "epoch": 2, "loss": 0.0356, "lr": 1.0942156691837575e-05, "grad_norm": 7.65625, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 5254, "epoch": 2, "loss": 0.2225, "lr": 1.0939174557457262e-05, "grad_norm": 5.5, "samples_per_sec": 11.02, "gpu_mem_gb": 6.83} {"step": 5255, "epoch": 2, "loss": 0.0435, "lr": 1.093619233880933e-05, "grad_norm": 4.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5256, "epoch": 2, "loss": 0.2584, "lr": 1.0933210036161368e-05, "grad_norm": 6.15625, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 5257, "epoch": 2, "loss": 0.0202, "lr": 1.0930227649780954e-05, "grad_norm": 3.109375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5258, "epoch": 2, "loss": 0.0014, "lr": 1.0927245179935687e-05, "grad_norm": 2.390625, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5259, "epoch": 2, "loss": 0.002, "lr": 1.0924262626893175e-05, "grad_norm": 2.046875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 5260, "epoch": 2, "loss": 0.0005, "lr": 1.0921279990921018e-05, "grad_norm": 1.484375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5261, "epoch": 2, "loss": 0.3563, "lr": 1.0918297272286842e-05, "grad_norm": 6.59375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 5262, "epoch": 2, "loss": 0.156, "lr": 1.0915314471258271e-05, "grad_norm": 7.0, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5263, "epoch": 2, "loss": 0.4683, "lr": 1.0912331588102933e-05, "grad_norm": 8.0, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5264, "epoch": 2, "loss": 0.001, "lr": 1.0909348623088472e-05, "grad_norm": 4.3125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.85} {"step": 5265, "epoch": 2, "loss": 0.1194, "lr": 1.0906365576482532e-05, "grad_norm": 3.328125, "samples_per_sec": 11.0, "gpu_mem_gb": 6.83} {"step": 5266, "epoch": 2, "loss": 0.0016, "lr": 1.0903382448552769e-05, "grad_norm": 3.578125, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 5267, "epoch": 2, "loss": 0.2152, "lr": 1.0900399239566846e-05, "grad_norm": 5.25, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5268, "epoch": 2, "loss": 0.001, "lr": 1.0897415949792427e-05, "grad_norm": 3.421875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 5269, "epoch": 2, "loss": 0.0008, "lr": 1.0894432579497192e-05, "grad_norm": 0.431640625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 5270, "epoch": 2, "loss": 0.0987, "lr": 1.0891449128948826e-05, "grad_norm": 3.015625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5271, "epoch": 2, "loss": 0.0013, "lr": 1.0888465598415018e-05, "grad_norm": 5.53125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 5272, "epoch": 2, "loss": 0.1259, "lr": 1.0885481988163467e-05, "grad_norm": 3.59375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.84} {"step": 5273, "epoch": 2, "loss": 0.0699, "lr": 1.0882498298461875e-05, "grad_norm": 3.703125, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5274, "epoch": 2, "loss": 0.0029, "lr": 1.0879514529577956e-05, "grad_norm": 2.171875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5275, "epoch": 2, "loss": 0.001, "lr": 1.0876530681779428e-05, "grad_norm": 7.34375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5276, "epoch": 2, "loss": 0.0045, "lr": 1.0873546755334022e-05, "grad_norm": 1.15625, "samples_per_sec": 10.75, "gpu_mem_gb": 6.82} {"step": 5277, "epoch": 2, "loss": 0.1253, "lr": 1.0870562750509464e-05, "grad_norm": 5.03125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5278, "epoch": 2, "loss": 0.0049, "lr": 1.08675786675735e-05, "grad_norm": 5.21875, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5279, "epoch": 2, "loss": 0.2008, "lr": 1.0864594506793876e-05, "grad_norm": 4.34375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.83} {"step": 5280, "epoch": 2, "loss": 0.0267, "lr": 1.0861610268438343e-05, "grad_norm": 3.671875, "samples_per_sec": 10.68, "gpu_mem_gb": 6.82} {"step": 5281, "epoch": 2, "loss": 0.0018, "lr": 1.0858625952774666e-05, "grad_norm": 1.578125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.82} {"step": 5282, "epoch": 2, "loss": 0.0995, "lr": 1.0855641560070616e-05, "grad_norm": 4.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5283, "epoch": 2, "loss": 0.0014, "lr": 1.0852657090593961e-05, "grad_norm": 4.15625, "samples_per_sec": 10.6, "gpu_mem_gb": 6.82} {"step": 5284, "epoch": 2, "loss": 0.034, "lr": 1.084967254461249e-05, "grad_norm": 3.578125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 5285, "epoch": 2, "loss": 0.001, "lr": 1.0846687922393985e-05, "grad_norm": 9.375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5286, "epoch": 2, "loss": 0.0935, "lr": 1.084370322420625e-05, "grad_norm": 4.375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5287, "epoch": 2, "loss": 0.1744, "lr": 1.0840718450317081e-05, "grad_norm": 4.09375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5288, "epoch": 2, "loss": 0.0549, "lr": 1.0837733600994289e-05, "grad_norm": 6.3125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 5289, "epoch": 2, "loss": 0.001, "lr": 1.083474867650569e-05, "grad_norm": 1.1796875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 5290, "epoch": 2, "loss": 0.0227, "lr": 1.083176367711911e-05, "grad_norm": 4.5625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5291, "epoch": 2, "loss": 0.0943, "lr": 1.0828778603102377e-05, "grad_norm": 4.3125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 5292, "epoch": 2, "loss": 0.0006, "lr": 1.0825793454723325e-05, "grad_norm": 2.921875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.82} {"step": 5293, "epoch": 2, "loss": 0.1296, "lr": 1.0822808232249798e-05, "grad_norm": 8.625, "samples_per_sec": 10.39, "gpu_mem_gb": 6.83} {"step": 5294, "epoch": 2, "loss": 0.1645, "lr": 1.081982293594965e-05, "grad_norm": 6.5625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.84} {"step": 5295, "epoch": 2, "loss": 0.0823, "lr": 1.081683756609073e-05, "grad_norm": 2.59375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 5296, "epoch": 2, "loss": 0.0537, "lr": 1.0813852122940907e-05, "grad_norm": 4.4375, "samples_per_sec": 10.63, "gpu_mem_gb": 6.83} {"step": 5297, "epoch": 2, "loss": 0.0033, "lr": 1.0810866606768042e-05, "grad_norm": 2.828125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5298, "epoch": 2, "loss": 0.0261, "lr": 1.0807881017840023e-05, "grad_norm": 6.0, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 5299, "epoch": 2, "loss": 0.3152, "lr": 1.0804895356424727e-05, "grad_norm": 6.40625, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 5300, "epoch": 2, "loss": 0.0008, "lr": 1.0801909622790036e-05, "grad_norm": 5.6875, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 5301, "epoch": 2, "loss": 0.0002, "lr": 1.079892381720386e-05, "grad_norm": 0.421875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 5302, "epoch": 2, "loss": 0.0001, "lr": 1.0795937939934088e-05, "grad_norm": 2.9375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 5303, "epoch": 2, "loss": 0.0028, "lr": 1.0792951991248637e-05, "grad_norm": 5.875, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 5304, "epoch": 2, "loss": 0.0003, "lr": 1.0789965971415415e-05, "grad_norm": 1.9296875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.85} {"step": 5305, "epoch": 2, "loss": 0.2235, "lr": 1.0786979880702345e-05, "grad_norm": 6.34375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.84} {"step": 5306, "epoch": 2, "loss": 0.0026, "lr": 1.0783993719377362e-05, "grad_norm": 2.75, "samples_per_sec": 10.94, "gpu_mem_gb": 6.82} {"step": 5307, "epoch": 2, "loss": 0.0315, "lr": 1.0781007487708388e-05, "grad_norm": 5.75, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5308, "epoch": 2, "loss": 0.0011, "lr": 1.0778021185963371e-05, "grad_norm": 1.734375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 5309, "epoch": 2, "loss": 0.0016, "lr": 1.0775034814410258e-05, "grad_norm": 0.2412109375, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 5310, "epoch": 2, "loss": 0.0005, "lr": 1.0772048373316997e-05, "grad_norm": 1.1328125, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 5311, "epoch": 2, "loss": 0.0229, "lr": 1.076906186295155e-05, "grad_norm": 1.0078125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.84} {"step": 5312, "epoch": 2, "loss": 0.0538, "lr": 1.076607528358188e-05, "grad_norm": 2.8125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5313, "epoch": 2, "loss": 0.0585, "lr": 1.0763088635475963e-05, "grad_norm": 2.390625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.84} {"step": 5314, "epoch": 2, "loss": 0.1837, "lr": 1.076010191890177e-05, "grad_norm": 6.125, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5315, "epoch": 2, "loss": 0.0594, "lr": 1.0757115134127291e-05, "grad_norm": 6.0, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5316, "epoch": 2, "loss": 0.4318, "lr": 1.0754128281420511e-05, "grad_norm": 8.0, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 5317, "epoch": 2, "loss": 0.0005, "lr": 1.0751141361049428e-05, "grad_norm": 8.4375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5318, "epoch": 2, "loss": 0.0136, "lr": 1.0748154373282047e-05, "grad_norm": 5.1875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5319, "epoch": 2, "loss": 0.0234, "lr": 1.074516731838637e-05, "grad_norm": 4.75, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 5320, "epoch": 2, "loss": 0.0007, "lr": 1.0742180196630414e-05, "grad_norm": 1.6328125, "samples_per_sec": 10.51, "gpu_mem_gb": 6.85} {"step": 5321, "epoch": 2, "loss": 0.086, "lr": 1.0739193008282203e-05, "grad_norm": 3.265625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5322, "epoch": 2, "loss": 0.0002, "lr": 1.0736205753609754e-05, "grad_norm": 1.9765625, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5323, "epoch": 2, "loss": 0.0224, "lr": 1.0733218432881107e-05, "grad_norm": 4.4375, "samples_per_sec": 10.6, "gpu_mem_gb": 6.85} {"step": 5324, "epoch": 2, "loss": 0.0654, "lr": 1.0730231046364294e-05, "grad_norm": 3.375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5325, "epoch": 2, "loss": 0.2153, "lr": 1.0727243594327366e-05, "grad_norm": 6.125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5326, "epoch": 2, "loss": 0.0054, "lr": 1.0724256077038368e-05, "grad_norm": 2.828125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5327, "epoch": 2, "loss": 0.0006, "lr": 1.0721268494765356e-05, "grad_norm": 8.4375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.82} {"step": 5328, "epoch": 2, "loss": 0.0848, "lr": 1.0718280847776387e-05, "grad_norm": 3.515625, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5329, "epoch": 2, "loss": 0.002, "lr": 1.0715293136339538e-05, "grad_norm": 3.25, "samples_per_sec": 10.98, "gpu_mem_gb": 6.82} {"step": 5330, "epoch": 2, "loss": 0.0856, "lr": 1.0712305360722876e-05, "grad_norm": 5.78125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 5331, "epoch": 2, "loss": 0.0031, "lr": 1.0709317521194478e-05, "grad_norm": 0.5, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5332, "epoch": 2, "loss": 0.0624, "lr": 1.0706329618022434e-05, "grad_norm": 5.5625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5333, "epoch": 2, "loss": 0.0785, "lr": 1.070334165147483e-05, "grad_norm": 3.46875, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 5334, "epoch": 2, "loss": 0.0018, "lr": 1.0700353621819761e-05, "grad_norm": 5.28125, "samples_per_sec": 10.44, "gpu_mem_gb": 6.82} {"step": 5335, "epoch": 2, "loss": 0.0001, "lr": 1.0697365529325334e-05, "grad_norm": 0.55078125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.82} {"step": 5336, "epoch": 2, "loss": 0.0202, "lr": 1.069437737425965e-05, "grad_norm": 2.40625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5337, "epoch": 2, "loss": 0.016, "lr": 1.0691389156890823e-05, "grad_norm": 3.6875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 5338, "epoch": 2, "loss": 0.0012, "lr": 1.0688400877486978e-05, "grad_norm": 4.4375, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5339, "epoch": 2, "loss": 0.2912, "lr": 1.0685412536316228e-05, "grad_norm": 5.75, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5340, "epoch": 2, "loss": 0.0019, "lr": 1.0682424133646712e-05, "grad_norm": 3.28125, "samples_per_sec": 10.63, "gpu_mem_gb": 6.82} {"step": 5341, "epoch": 2, "loss": 0.008, "lr": 1.0679435669746558e-05, "grad_norm": 0.482421875, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5342, "epoch": 2, "loss": 0.0201, "lr": 1.0676447144883912e-05, "grad_norm": 3.046875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5343, "epoch": 2, "loss": 0.0238, "lr": 1.067345855932692e-05, "grad_norm": 8.0625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.82} {"step": 5344, "epoch": 2, "loss": 0.0251, "lr": 1.0670469913343727e-05, "grad_norm": 9.0625, "samples_per_sec": 10.67, "gpu_mem_gb": 6.83} {"step": 5345, "epoch": 2, "loss": 0.1509, "lr": 1.0667481207202494e-05, "grad_norm": 7.59375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5346, "epoch": 2, "loss": 0.0059, "lr": 1.0664492441171385e-05, "grad_norm": 1.40625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5347, "epoch": 2, "loss": 0.0007, "lr": 1.0661503615518564e-05, "grad_norm": 0.322265625, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 5348, "epoch": 2, "loss": 0.2895, "lr": 1.0658514730512208e-05, "grad_norm": 6.53125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5349, "epoch": 2, "loss": 0.0885, "lr": 1.0655525786420493e-05, "grad_norm": 4.09375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5350, "epoch": 2, "loss": 0.0179, "lr": 1.0652536783511602e-05, "grad_norm": 2.59375, "samples_per_sec": 10.7, "gpu_mem_gb": 6.83} {"step": 5351, "epoch": 2, "loss": 0.0975, "lr": 1.0649547722053721e-05, "grad_norm": 3.203125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5352, "epoch": 2, "loss": 0.0958, "lr": 1.0646558602315052e-05, "grad_norm": 4.125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 5353, "epoch": 2, "loss": 0.0321, "lr": 1.0643569424563789e-05, "grad_norm": 3.390625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 5354, "epoch": 2, "loss": 0.1637, "lr": 1.0640580189068135e-05, "grad_norm": 9.3125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.83} {"step": 5355, "epoch": 2, "loss": 0.1518, "lr": 1.0637590896096303e-05, "grad_norm": 5.46875, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5356, "epoch": 2, "loss": 0.0015, "lr": 1.0634601545916508e-05, "grad_norm": 3.203125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 5357, "epoch": 2, "loss": 0.0073, "lr": 1.063161213879697e-05, "grad_norm": 2.390625, "samples_per_sec": 10.72, "gpu_mem_gb": 6.83} {"step": 5358, "epoch": 2, "loss": 0.0087, "lr": 1.062862267500591e-05, "grad_norm": 5.15625, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 5359, "epoch": 2, "loss": 0.0009, "lr": 1.062563315481156e-05, "grad_norm": 3.03125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5360, "epoch": 2, "loss": 0.0014, "lr": 1.0622643578482161e-05, "grad_norm": 7.3125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 5361, "epoch": 2, "loss": 0.0281, "lr": 1.0619653946285948e-05, "grad_norm": 1.5078125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 5362, "epoch": 2, "loss": 0.135, "lr": 1.0616664258491167e-05, "grad_norm": 5.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 5363, "epoch": 2, "loss": 0.0259, "lr": 1.0613674515366069e-05, "grad_norm": 4.65625, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 5364, "epoch": 2, "loss": 0.0002, "lr": 1.0610684717178905e-05, "grad_norm": 4.25, "samples_per_sec": 10.69, "gpu_mem_gb": 6.82} {"step": 5365, "epoch": 2, "loss": 0.035, "lr": 1.0607694864197943e-05, "grad_norm": 3.921875, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5366, "epoch": 2, "loss": 0.0179, "lr": 1.0604704956691443e-05, "grad_norm": 6.46875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5367, "epoch": 2, "loss": 0.0017, "lr": 1.0601714994927673e-05, "grad_norm": 3.515625, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 5368, "epoch": 2, "loss": 0.0138, "lr": 1.0598724979174915e-05, "grad_norm": 2.4375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.85} {"step": 5369, "epoch": 2, "loss": 0.1859, "lr": 1.0595734909701444e-05, "grad_norm": 5.71875, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5370, "epoch": 2, "loss": 0.3023, "lr": 1.0592744786775547e-05, "grad_norm": 7.25, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 5371, "epoch": 2, "loss": 0.0009, "lr": 1.0589754610665505e-05, "grad_norm": 6.71875, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 5372, "epoch": 2, "loss": 0.0608, "lr": 1.0586764381639625e-05, "grad_norm": 5.25, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5373, "epoch": 2, "loss": 0.173, "lr": 1.0583774099966197e-05, "grad_norm": 7.6875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5374, "epoch": 2, "loss": 0.073, "lr": 1.0580783765913529e-05, "grad_norm": 3.984375, "samples_per_sec": 10.43, "gpu_mem_gb": 6.83} {"step": 5375, "epoch": 2, "loss": 0.0523, "lr": 1.0577793379749925e-05, "grad_norm": 5.09375, "samples_per_sec": 10.94, "gpu_mem_gb": 6.85} {"step": 5376, "epoch": 2, "loss": 0.0304, "lr": 1.0574802941743702e-05, "grad_norm": 5.9375, "samples_per_sec": 10.97, "gpu_mem_gb": 6.85} {"step": 5377, "epoch": 2, "loss": 0.0468, "lr": 1.0571812452163178e-05, "grad_norm": 4.6875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5378, "epoch": 2, "loss": 0.2273, "lr": 1.056882191127667e-05, "grad_norm": 5.25, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 5379, "epoch": 2, "loss": 0.0063, "lr": 1.0565831319352508e-05, "grad_norm": 0.3984375, "samples_per_sec": 10.87, "gpu_mem_gb": 6.85} {"step": 5380, "epoch": 2, "loss": 0.0004, "lr": 1.0562840676659025e-05, "grad_norm": 9.8125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5381, "epoch": 2, "loss": 0.0478, "lr": 1.0559849983464557e-05, "grad_norm": 2.5, "samples_per_sec": 10.42, "gpu_mem_gb": 6.83} {"step": 5382, "epoch": 2, "loss": 0.4107, "lr": 1.0556859240037443e-05, "grad_norm": 8.375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5383, "epoch": 2, "loss": 0.0357, "lr": 1.0553868446646023e-05, "grad_norm": 4.625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5384, "epoch": 2, "loss": 0.0164, "lr": 1.0550877603558656e-05, "grad_norm": 1.1875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.83} {"step": 5385, "epoch": 2, "loss": 0.1331, "lr": 1.0547886711043689e-05, "grad_norm": 5.5, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5386, "epoch": 2, "loss": 0.0007, "lr": 1.0544895769369488e-05, "grad_norm": 2.34375, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5387, "epoch": 2, "loss": 0.0009, "lr": 1.0541904778804405e-05, "grad_norm": 0.162109375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.83} {"step": 5388, "epoch": 2, "loss": 0.0001, "lr": 1.0538913739616817e-05, "grad_norm": 5.34375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.83} {"step": 5389, "epoch": 2, "loss": 0.0017, "lr": 1.0535922652075091e-05, "grad_norm": 1.15625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5390, "epoch": 2, "loss": 0.0014, "lr": 1.05329315164476e-05, "grad_norm": 5.4375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 5391, "epoch": 2, "loss": 0.0017, "lr": 1.0529940333002731e-05, "grad_norm": 1.7890625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5392, "epoch": 2, "loss": 0.0776, "lr": 1.0526949102008865e-05, "grad_norm": 4.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.84} {"step": 5393, "epoch": 2, "loss": 0.0673, "lr": 1.052395782373439e-05, "grad_norm": 3.046875, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5394, "epoch": 2, "loss": 0.0771, "lr": 1.0520966498447698e-05, "grad_norm": 9.125, "samples_per_sec": 10.77, "gpu_mem_gb": 6.82} {"step": 5395, "epoch": 2, "loss": 0.014, "lr": 1.0517975126417192e-05, "grad_norm": 4.9375, "samples_per_sec": 10.78, "gpu_mem_gb": 6.83} {"step": 5396, "epoch": 2, "loss": 0.0013, "lr": 1.0514983707911269e-05, "grad_norm": 5.59375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5397, "epoch": 2, "loss": 0.0488, "lr": 1.0511992243198335e-05, "grad_norm": 3.953125, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5398, "epoch": 2, "loss": 0.0007, "lr": 1.0509000732546794e-05, "grad_norm": 5.53125, "samples_per_sec": 10.72, "gpu_mem_gb": 6.82} {"step": 5399, "epoch": 2, "loss": 0.2564, "lr": 1.0506009176225071e-05, "grad_norm": 7.3125, "samples_per_sec": 10.89, "gpu_mem_gb": 6.84} {"step": 5400, "epoch": 2, "loss": 0.0011, "lr": 1.050301757450158e-05, "grad_norm": 1.5546875, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5401, "epoch": 2, "loss": 0.1425, "lr": 1.0500025927644742e-05, "grad_norm": 8.1875, "samples_per_sec": 10.52, "gpu_mem_gb": 6.83} {"step": 5402, "epoch": 2, "loss": 0.0821, "lr": 1.0497034235922978e-05, "grad_norm": 4.46875, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5403, "epoch": 2, "loss": 0.0083, "lr": 1.0494042499604726e-05, "grad_norm": 4.15625, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5404, "epoch": 2, "loss": 0.0317, "lr": 1.0491050718958418e-05, "grad_norm": 5.1875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5405, "epoch": 2, "loss": 0.0147, "lr": 1.0488058894252488e-05, "grad_norm": 2.578125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.85} {"step": 5406, "epoch": 2, "loss": 0.0057, "lr": 1.0485067025755384e-05, "grad_norm": 5.4375, "samples_per_sec": 11.03, "gpu_mem_gb": 6.83} {"step": 5407, "epoch": 2, "loss": 0.0591, "lr": 1.048207511373555e-05, "grad_norm": 2.734375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 5408, "epoch": 2, "loss": 0.0211, "lr": 1.0479083158461432e-05, "grad_norm": 4.25, "samples_per_sec": 10.48, "gpu_mem_gb": 6.82} {"step": 5409, "epoch": 2, "loss": 0.0156, "lr": 1.047609116020149e-05, "grad_norm": 0.99609375, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5410, "epoch": 2, "loss": 0.0025, "lr": 1.0473099119224172e-05, "grad_norm": 0.9296875, "samples_per_sec": 10.88, "gpu_mem_gb": 6.82} {"step": 5411, "epoch": 2, "loss": 0.0345, "lr": 1.0470107035797951e-05, "grad_norm": 6.65625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.84} {"step": 5412, "epoch": 2, "loss": 0.0356, "lr": 1.046711491019129e-05, "grad_norm": 6.59375, "samples_per_sec": 10.96, "gpu_mem_gb": 6.82} {"step": 5413, "epoch": 2, "loss": 0.0087, "lr": 1.0464122742672651e-05, "grad_norm": 6.125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.85} {"step": 5414, "epoch": 2, "loss": 0.0084, "lr": 1.0461130533510511e-05, "grad_norm": 4.9375, "samples_per_sec": 10.61, "gpu_mem_gb": 6.85} {"step": 5415, "epoch": 2, "loss": 0.0014, "lr": 1.0458138282973345e-05, "grad_norm": 6.0625, "samples_per_sec": 10.77, "gpu_mem_gb": 6.83} {"step": 5416, "epoch": 2, "loss": 0.1499, "lr": 1.0455145991329639e-05, "grad_norm": 8.8125, "samples_per_sec": 10.94, "gpu_mem_gb": 6.84} {"step": 5417, "epoch": 2, "loss": 0.0007, "lr": 1.0452153658847867e-05, "grad_norm": 6.40625, "samples_per_sec": 10.94, "gpu_mem_gb": 6.83} {"step": 5418, "epoch": 2, "loss": 0.0001, "lr": 1.0449161285796526e-05, "grad_norm": 6.53125, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 5419, "epoch": 2, "loss": 0.0786, "lr": 1.0446168872444104e-05, "grad_norm": 6.21875, "samples_per_sec": 10.92, "gpu_mem_gb": 6.82} {"step": 5420, "epoch": 2, "loss": 0.0883, "lr": 1.0443176419059093e-05, "grad_norm": 6.375, "samples_per_sec": 10.93, "gpu_mem_gb": 6.84} {"step": 5421, "epoch": 2, "loss": 0.0023, "lr": 1.0440183925909998e-05, "grad_norm": 1.9609375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.82} {"step": 5422, "epoch": 2, "loss": 0.0427, "lr": 1.0437191393265311e-05, "grad_norm": 1.8125, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5423, "epoch": 2, "loss": 0.3109, "lr": 1.0434198821393547e-05, "grad_norm": 5.625, "samples_per_sec": 10.79, "gpu_mem_gb": 6.83} {"step": 5424, "epoch": 2, "loss": 0.4, "lr": 1.0431206210563211e-05, "grad_norm": 5.875, "samples_per_sec": 10.96, "gpu_mem_gb": 6.83} {"step": 5425, "epoch": 2, "loss": 0.0314, "lr": 1.0428213561042815e-05, "grad_norm": 1.4609375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.83} {"step": 5426, "epoch": 2, "loss": 0.001, "lr": 1.0425220873100877e-05, "grad_norm": 4.15625, "samples_per_sec": 10.81, "gpu_mem_gb": 6.82} {"step": 5427, "epoch": 2, "loss": 0.0164, "lr": 1.0422228147005914e-05, "grad_norm": 1.6171875, "samples_per_sec": 10.85, "gpu_mem_gb": 6.85} {"step": 5428, "epoch": 2, "loss": 0.0317, "lr": 1.0419235383026451e-05, "grad_norm": 2.734375, "samples_per_sec": 10.56, "gpu_mem_gb": 6.83} {"step": 5429, "epoch": 2, "loss": 0.0469, "lr": 1.0416242581431018e-05, "grad_norm": 1.828125, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5430, "epoch": 2, "loss": 0.0025, "lr": 1.0413249742488132e-05, "grad_norm": 3.921875, "samples_per_sec": 10.8, "gpu_mem_gb": 6.82} {"step": 5431, "epoch": 2, "loss": 0.0276, "lr": 1.0410256866466342e-05, "grad_norm": 4.53125, "samples_per_sec": 10.64, "gpu_mem_gb": 6.82} {"step": 5432, "epoch": 2, "loss": 0.0587, "lr": 1.0407263953634173e-05, "grad_norm": 4.5625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.83} {"step": 5433, "epoch": 2, "loss": 0.0229, "lr": 1.040427100426017e-05, "grad_norm": 5.3125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5434, "epoch": 2, "loss": 0.0003, "lr": 1.0401278018612874e-05, "grad_norm": 4.71875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.82} {"step": 5435, "epoch": 2, "loss": 0.1256, "lr": 1.039828499696083e-05, "grad_norm": 4.84375, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 5436, "epoch": 2, "loss": 0.112, "lr": 1.0395291939572593e-05, "grad_norm": 5.46875, "samples_per_sec": 11.02, "gpu_mem_gb": 6.84} {"step": 5437, "epoch": 2, "loss": 0.011, "lr": 1.0392298846716706e-05, "grad_norm": 4.53125, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 5438, "epoch": 2, "loss": 0.0345, "lr": 1.0389305718661737e-05, "grad_norm": 2.78125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5439, "epoch": 2, "loss": 0.1245, "lr": 1.0386312555676236e-05, "grad_norm": 5.3125, "samples_per_sec": 10.82, "gpu_mem_gb": 6.84} {"step": 5440, "epoch": 2, "loss": 0.0075, "lr": 1.0383319358028766e-05, "grad_norm": 4.15625, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5441, "epoch": 2, "loss": 0.2729, "lr": 1.0380326125987896e-05, "grad_norm": 3.703125, "samples_per_sec": 10.49, "gpu_mem_gb": 6.83} {"step": 5442, "epoch": 2, "loss": 0.0019, "lr": 1.0377332859822192e-05, "grad_norm": 1.34375, "samples_per_sec": 10.59, "gpu_mem_gb": 6.85} {"step": 5443, "epoch": 2, "loss": 0.0266, "lr": 1.0374339559800226e-05, "grad_norm": 3.8125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5444, "epoch": 2, "loss": 0.0341, "lr": 1.0371346226190571e-05, "grad_norm": 2.578125, "samples_per_sec": 10.95, "gpu_mem_gb": 6.83} {"step": 5445, "epoch": 2, "loss": 0.0007, "lr": 1.0368352859261805e-05, "grad_norm": 3.796875, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 5446, "epoch": 2, "loss": 0.0024, "lr": 1.0365359459282511e-05, "grad_norm": 0.259765625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5447, "epoch": 2, "loss": 0.0246, "lr": 1.0362366026521269e-05, "grad_norm": 4.53125, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5448, "epoch": 2, "loss": 0.0612, "lr": 1.0359372561246668e-05, "grad_norm": 5.1875, "samples_per_sec": 10.51, "gpu_mem_gb": 6.83} {"step": 5449, "epoch": 2, "loss": 0.0293, "lr": 1.0356379063727293e-05, "grad_norm": 3.6875, "samples_per_sec": 10.75, "gpu_mem_gb": 6.83} {"step": 5450, "epoch": 2, "loss": 0.2905, "lr": 1.0353385534231739e-05, "grad_norm": 6.15625, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5451, "epoch": 2, "loss": 0.0003, "lr": 1.0350391973028604e-05, "grad_norm": 1.859375, "samples_per_sec": 10.71, "gpu_mem_gb": 6.82} {"step": 5452, "epoch": 2, "loss": 0.0345, "lr": 1.0347398380386479e-05, "grad_norm": 4.6875, "samples_per_sec": 10.61, "gpu_mem_gb": 6.82} {"step": 5453, "epoch": 2, "loss": 0.0007, "lr": 1.0344404756573972e-05, "grad_norm": 6.84375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.85} {"step": 5454, "epoch": 2, "loss": 0.0779, "lr": 1.034141110185968e-05, "grad_norm": 6.21875, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5455, "epoch": 2, "loss": 0.0821, "lr": 1.0338417416512215e-05, "grad_norm": 3.375, "samples_per_sec": 10.53, "gpu_mem_gb": 6.83} {"step": 5456, "epoch": 2, "loss": 0.0168, "lr": 1.0335423700800182e-05, "grad_norm": 10.625, "samples_per_sec": 10.82, "gpu_mem_gb": 6.83} {"step": 5457, "epoch": 2, "loss": 0.0005, "lr": 1.0332429954992191e-05, "grad_norm": 1.3515625, "samples_per_sec": 8.78, "gpu_mem_gb": 6.82} {"step": 5458, "epoch": 2, "loss": 0.0001, "lr": 1.0329436179356862e-05, "grad_norm": 0.1943359375, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5459, "epoch": 2, "loss": 0.0115, "lr": 1.0326442374162808e-05, "grad_norm": 4.5625, "samples_per_sec": 10.97, "gpu_mem_gb": 6.82} {"step": 5460, "epoch": 2, "loss": 0.0003, "lr": 1.0323448539678653e-05, "grad_norm": 3.296875, "samples_per_sec": 10.97, "gpu_mem_gb": 6.83} {"step": 5461, "epoch": 2, "loss": 0.2953, "lr": 1.0320454676173015e-05, "grad_norm": 6.03125, "samples_per_sec": 10.87, "gpu_mem_gb": 6.83} {"step": 5462, "epoch": 2, "loss": 0.0004, "lr": 1.031746078391452e-05, "grad_norm": 0.51953125, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5463, "epoch": 2, "loss": 0.0015, "lr": 1.0314466863171794e-05, "grad_norm": 5.90625, "samples_per_sec": 10.9, "gpu_mem_gb": 6.82} {"step": 5464, "epoch": 2, "loss": 0.1026, "lr": 1.0311472914213471e-05, "grad_norm": 4.59375, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5465, "epoch": 2, "loss": 0.0464, "lr": 1.0308478937308182e-05, "grad_norm": 6.9375, "samples_per_sec": 10.73, "gpu_mem_gb": 6.82} {"step": 5466, "epoch": 2, "loss": 0.0064, "lr": 1.0305484932724563e-05, "grad_norm": 3.09375, "samples_per_sec": 10.92, "gpu_mem_gb": 6.83} {"step": 5467, "epoch": 2, "loss": 0.0255, "lr": 1.0302490900731247e-05, "grad_norm": 1.296875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.83} {"step": 5468, "epoch": 2, "loss": 0.0007, "lr": 1.0299496841596884e-05, "grad_norm": 0.10498046875, "samples_per_sec": 10.22, "gpu_mem_gb": 6.82} {"step": 5469, "epoch": 2, "loss": 0.0014, "lr": 1.02965027555901e-05, "grad_norm": 3.421875, "samples_per_sec": 10.98, "gpu_mem_gb": 6.85} {"step": 5470, "epoch": 2, "loss": 0.0003, "lr": 1.0293508642979557e-05, "grad_norm": 3.765625, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5471, "epoch": 2, "loss": 0.0002, "lr": 1.0290514504033895e-05, "grad_norm": 1.109375, "samples_per_sec": 10.84, "gpu_mem_gb": 6.82} {"step": 5472, "epoch": 2, "loss": 0.1401, "lr": 1.0287520339021763e-05, "grad_norm": 19.5, "samples_per_sec": 10.65, "gpu_mem_gb": 6.83} {"step": 5473, "epoch": 2, "loss": 0.0065, "lr": 1.0284526148211815e-05, "grad_norm": 2.359375, "samples_per_sec": 10.88, "gpu_mem_gb": 6.83} {"step": 5474, "epoch": 2, "loss": 0.022, "lr": 1.0281531931872704e-05, "grad_norm": 4.1875, "samples_per_sec": 10.78, "gpu_mem_gb": 6.82} {"step": 5475, "epoch": 2, "loss": 0.0006, "lr": 1.0278537690273088e-05, "grad_norm": 3.453125, "samples_per_sec": 10.76, "gpu_mem_gb": 6.82} {"step": 5476, "epoch": 2, "loss": 0.1605, "lr": 1.0275543423681622e-05, "grad_norm": 6.03125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 5477, "epoch": 2, "loss": 0.0007, "lr": 1.0272549132366971e-05, "grad_norm": 2.40625, "samples_per_sec": 10.89, "gpu_mem_gb": 6.82} {"step": 5478, "epoch": 2, "loss": 0.2235, "lr": 1.02695548165978e-05, "grad_norm": 7.15625, "samples_per_sec": 10.62, "gpu_mem_gb": 6.83} {"step": 5479, "epoch": 2, "loss": 0.0854, "lr": 1.0266560476642768e-05, "grad_norm": 17.125, "samples_per_sec": 10.74, "gpu_mem_gb": 6.85} {"step": 5480, "epoch": 2, "loss": 0.0249, "lr": 1.026356611277055e-05, "grad_norm": 1.546875, "samples_per_sec": 10.83, "gpu_mem_gb": 6.83} {"step": 5481, "epoch": 2, "loss": 0.0002, "lr": 1.0260571725249808e-05, "grad_norm": 5.6875, "samples_per_sec": 10.84, "gpu_mem_gb": 6.83} {"step": 5482, "epoch": 2, "loss": 0.1237, "lr": 1.0257577314349223e-05, "grad_norm": 6.46875, "samples_per_sec": 10.57, "gpu_mem_gb": 6.82} {"step": 5483, "epoch": 2, "loss": 0.0296, "lr": 1.0254582880337465e-05, "grad_norm": 3.703125, "samples_per_sec": 10.93, "gpu_mem_gb": 6.83} {"step": 5484, "epoch": 2, "loss": 0.1278, "lr": 1.0251588423483205e-05, "grad_norm": 4.3125, "samples_per_sec": 10.9, "gpu_mem_gb": 6.83} {"step": 5485, "epoch": 2, "loss": 0.0104, "lr": 1.0248593944055131e-05, "grad_norm": 2.953125, "samples_per_sec": 10.7, "gpu_mem_gb": 6.82} {"step": 5486, "epoch": 2, "loss": 0.0002, "lr": 1.0245599442321916e-05, "grad_norm": 2.09375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 5487, "epoch": 2, "loss": 0.0936, "lr": 1.0242604918552245e-05, "grad_norm": 4.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.82} {"step": 5488, "epoch": 2, "loss": 0.2814, "lr": 1.0239610373014804e-05, "grad_norm": 4.65625, "samples_per_sec": 10.5, "gpu_mem_gb": 6.83} {"step": 5489, "epoch": 2, "loss": 0.1236, "lr": 1.0236615805978272e-05, "grad_norm": 4.4375, "samples_per_sec": 10.66, "gpu_mem_gb": 6.82} {"step": 5490, "epoch": 2, "loss": 0.0016, "lr": 1.0233621217711348e-05, "grad_norm": 3.53125, "samples_per_sec": 10.91, "gpu_mem_gb": 6.83} {"step": 5491, "epoch": 2, "loss": 0.0003, "lr": 1.023062660848271e-05, "grad_norm": 5.25, "samples_per_sec": 10.86, "gpu_mem_gb": 6.83} {"step": 5492, "epoch": 2, "loss": 0.0007, "lr": 1.0227631978561057e-05, "grad_norm": 6.53125, "samples_per_sec": 10.8, "gpu_mem_gb": 6.83} {"step": 5493, "epoch": 2, "loss": 0.0043, "lr": 1.0224637328215086e-05, "grad_norm": 5.09375, "samples_per_sec": 10.85, "gpu_mem_gb": 6.83} {"step": 5494, "epoch": 2, "loss": 0.0005, "lr": 1.0221642657713483e-05, "grad_norm": 8.625, "samples_per_sec": 10.76, "gpu_mem_gb": 6.83} {"step": 5495, "epoch": 2, "loss": 0.2153, "lr": 1.0218647967324952e-05, "grad_norm": 3.921875, "samples_per_sec": 10.48, "gpu_mem_gb": 6.83}