人工智能大模型探索路径-培训第 11 部分：大语言模型转换器库-模型组件实践

BaseModelOutputWithPoolingAndCrossAttentions(last_hidden_state=tensor([[[ 0.3299, 0.8761, 1.1550, ..., -0.2296, 0.3674, 0.1555], [ 0.6773, -0.5668, 0.0701, ..., -0.3799, -0.2055, -0.2795], [ 0.0841, -0.0825, 0.5001, ..., -0.3421, -0.8017, 0.3085], [ 0.0224, 0.4439, -0.1954, ..., -0.0618, -0.2570, -0.1142], [ 0.1476, 0.7324, -0.2727, ..., -0.1874, 0.1372, -0.3034], [ 0.3260, 0.8858, 1.1529, ..., -0.2277, 0.3656, 0.1587]]], grad_fn=<NativeLayerNormBackward0>), pooler_output=tensor([[ 2.3760e-02, -9.9773e-01, -9.9995e-01, -7.9692e-01, 9.9645e-01, 1.8497e-01, -2.9150e-01, -1.9350e-02, 9.9650e-01, 9.9989e-01, 1.6279e-01, -1.0000e+00, -3.5968e-02, 9.9875e-01, -9.9999e-01, 9.9879e-01, 9.9744e-01, 8.8014e-01, -9.8801e-01, 8.8588e-03, -9.3252e-01, -7.1405e-01, 2.3119e-01, 9.7772e-01, 9.9352e-01, -9.9826e-01, -9.9987e-01, -7.6323e-02, -8.6268e-01, -9.9992e-01, 。。。 -1.0000e+00, -1.0385e-02, 3.3378e-01, -9.7509e-01, -3.8623e-01, -9.2922e-01, 1.8362e-01, 9.9848e-01, -6.6866e-01, 9.1038e-01, -9.6579e-01, -9.9962e-01, 1.7735e-01, -9.9997e-01, 9.8384e-01, 1.0000e+00, -1.6213e-01, 9.7850e-01, -9.0667e-01, 1.2217e-01, -9.9999e-01, -9.9999e-01, 8.7128e-01, 9.9946e-01, 1.0102e-01, -9.9855e-01, 1.5214e-01, -9.9987e-01, -2.8880e-01, 5.7587e-01, 9.9336e-01, -9.9998e-01, 9.9947e-01, -6.4215e-01, 1.2852e-01, 9.8215e-01, -1.0000e+00, 8.4377e-01, -9.9904e-01, 9.9924e-01, -1.0000e+00, 9.9885e-01, 9.1444e-02, 2.1949e-01, 3.0374e-01, 9.7917e-01, -9.9957e-01, -1.9862e-01, 9.8820e-01, 9.9878e-01, -4.6083e-01, 9.8808e-01, 2.5509e-02]], grad_fn=<TanhBackward0>), hidden_states=None, past_key_values=None, attentions=(tensor([[[[5.6623e-01, 8.2719e-04, 7.1828e-04, 3.0170e-04, 3.7589e-04, 4.3154e-01], [1.0803e-02, 1.0227e-01, 1.1626e-01, 2.4532e-01, 5.2110e-01, 4.2390e-03], [4.1117e-02, 1.4175e-01, 2.8284e-01, 2.3116e-01, 2.9142e-01, 1.1724e-02], [3.2763e-02, 1.7774e-01, 8.3885e-02, 1.2232e-01, 5.7365e-01, 9.6361e-03], [9.1095e-02, 8.2052e-02, 7.2211e-02, 6.6084e-02, 6.5778e-01, 3.0778e-02], [6.0973e-01, 1.3369e-03, 1.1642e-03, 7.6050e-04, 1.2266e-03, 3.8578e-01]], ... [[4.1663e-01, 2.5743e-02, 3.3069e-02, 2.7573e-02, 4.8997e-02, 4.4799e-01], [9.3001e-01, 2.1486e-02, 2.9548e-02, 5.0943e-03, 5.1063e-03, 8.7589e-03], [5.5021e-02, 8.8170e-01, 3.0817e-02, 7.5048e-03, 2.2834e-02, 2.1212e-03], [3.3265e-02, 1.9535e-02, 9.0744e-01, 1.3685e-02, 1.4819e-02, 1.1257e-02], [1.7344e-01, 1.0500e-01, 1.1638e-01, 5.0705e-01, 7.0892e-02, 2.7225e-02], [7.2889e-02, 9.4428e-03, 1.2270e-02, 2.8002e-02, 4.1791e-01, 4.5948e-01]]]], grad_fn=<SoftmaxBackward0>), tensor([[[[4.5796e-01, 9.9558e-03, 1.0020e-02, 2.2092e-02, 6.5916e-02, 4.3405e-01], [4.3907e-01, 8.2675e-03, 1.1408e-01, 9.8204e-03, 6.1844e-03, 4.2258e-01], [2.1006e-01, 5.2901e-01, 3.7502e-03, 1.8742e-03, 3.6851e-02, 2.1846e-01], [3.0209e-01, 1.3334e-04, 3.6354e-01, 1.5319e-03, 4.1311e-02, 2.9139e-01], [8.7918e-02, 2.8945e-02, 3.2605e-03, 7.6759e-01, 2.3085e-02, 8.9203e-02], [4.5857e-01, 9.8717e-03, 9.9597e-03, 2.0971e-02, 6.6149e-02, 4.3448e-01]], ... [[5.0059e-01, 1.7413e-03, 9.1523e-04, 5.8339e-03, 6.8815e-03, 4.8404e-01], [4.3327e-01, 3.0781e-02, 1.8540e-02, 4.1750e-02, 4.3144e-02, 4.3252e-01], [4.6648e-01, 1.6748e-02, 2.3486e-03, 4.6985e-02, 7.2329e-03, 4.6021e-01], [4.6306e-01, 8.3569e-03, 5.8932e-03, 5.5269e-02, 1.3125e-02, 4.5429e-01], [4.5754e-01, 3.4543e-02, 7.7011e-03, 2.6113e-02, 2.0748e-02, 4.5336e-01], [5.0072e-01, 1.7318e-03, 9.1770e-04, 5.7174e-03, 6.6844e-03, 4.8423e-01]]]], grad_fn=<SoftmaxBackward0>), tensor([[[[3.1533e-01, 6.1664e-02, 3.2057e-02, 7.3921e-02, 2.0709e-01, 3.0993e-01], [3.5292e-01, 9.7867e-02, 7.6420e-02, 5.6472e-02, 6.7811e-02, 3.4851e-01], [3.7263e-01, 1.6694e-01, 4.0402e-02, 3.4887e-02, 1.7787e-02, 3.6736e-01], [4.6658e-01, 3.5042e-02, 9.0006e-03, 1.9572e-02, 1.0726e-02, 4.5908e-01], [3.6866e-01, 7.5453e-02, 4.1273e-02, 4.2048e-02, 1.1026e-01, 3.6231e-01], [3.1706e-01, 6.0658e-02, 3.1613e-02, 7.3170e-02, 2.0586e-01, 3.1164e-01]], ... [[2.7216e-01, 1.2583e-01, 5.1468e-02, 8.0823e-02, 2.0379e-01, 2.6593e-01], [1.3690e-01, 3.7407e-02, 4.5398e-01, 2.0540e-01, 3.1681e-02, 1.3463e-01], [2.4628e-01, 1.2131e-02, 9.3940e-03, 3.9019e-01, 9.8574e-02, 2.4343e-01], [4.4038e-01, 3.6646e-03, 3.7501e-03, 2.5508e-02, 8.7129e-02, 4.3957e-01], [3.6957e-01, 3.0530e-02, 6.2554e-02, 1.2588e-01, 4.2998e-02, 3.6847e-01], [2.7344e-01, 1.2471e-01, 5.0839e-02, 8.0282e-02, 2.0355e-01, 2.6718e-01]]]], grad_fn=<SoftmaxBackward0>)), cross_attentions=None)