I'm trying to calculate the softmax to get the probability of the text being real or not.
When I use the openai weights and load the checkpoint for their detector I get the following results using the pytorch softmax function.
('fake', tensor([2.3000e-01, 7.2000e-01, 5.9000e-01, 7.2000e-01, 6.8000e-01, 3.2000e-01, 1.3380e+01, 5.1000e-01, 4.3000e-01, 6.8000e-01, 9.0000e-01, 4.2000e-01,
4.3000e-01, 4.3000e-01, 4.3000e-01, 6.0000e-01, 5.0000e-01, 4.1000e-01,
4.7000e-01, 1.2000e-01, 5.1000e-01, 4.6000e-01, 4.5000e-01, 5.3000e-01,
6.1000e-01, 6.0000e-01, 5.9000e-01, 5.1000e-01, 5.1000e-01, 3.8000e-01,
4.6000e-01, 4.0000e-01, 3.9000e-01, 4.3000e-01, 4.3000e-01, 5.2000e-01,
0.0000e+00, 8.2000e-01, 9.0000e-01, 3.3000e-01, 6.8000e-01, 8.8000e-01,
7.5000e-01, 4.2000e-01, 4.2000e-01, 5.5000e-01, 5.7000e-01, 6.1000e-01,
7.9000e-01, 6.4000e-01, 5.4000e-01, 6.3000e-01, 3.1000e-01, 3.3000e-01,
5.1000e-01, 3.8000e-01, 3.7000e-01, 5.2000e-01, 4.8000e-01, 4.1000e-01,
4.0000e-01, 5.2000e-01, 5.5000e-01, 2.8000e-01, 6.1000e-01, 4.7000e-01,
5.4000e-01, 2.9000e-01, 3.4000e-01, 4.5000e-01, 3.2000e-01, 3.9000e-01,
4.1000e-01, 5.7000e-01, 7.2000e-01, 4.2000e-01, 5.4000e-01, 3.8000e-01,
3.6000e-01, 4.9000e-01, 3.6000e-01, 4.2000e-01, 5.3000e-01, 3.6000e-01,
4.4000e-01, 4.4000e-01, 6.4000e-01, 5.9000e-01, 3.8000e-01, 4.0000e-01,
4.7000e-01, 4.8000e-01, 6.5000e-01, 3.5000e-01, 2.6000e-01, 5.1000e-01,
6.0000e-01, 5.8000e-01, 4.2000e-01, 7.0000e-01, 5.8000e-01, 6.5000e-01,
9.0000e-02, 4.3000e-01, 5.1000e-01, 3.1000e-01, 4.8000e-01, 0.0000e+00,
3.3000e-01, 4.7000e-01, 2.8000e-01, 7.2000e-01, 4.1000e-01, 7.0000e-01,
5.5000e-01, 5.2000e-01, 6.5000e-01, 4.2000e-01, 6.8000e-01, 5.8000e-01,
6.9000e-01, 5.8000e-01, 5.3000e-01, 2.6000e-01, 5.2000e-01, 8.3000e-01,
4.0000e-01, 5.6000e-01, 3.2000e-01, 4.6000e-01, 6.3000e-01, 5.4000e-01,
5.0000e-01, 7.1000e-01, 4.3000e-01, 6.1000e-01, 6.8000e-01, 4.1000e-01,
6.6000e-01, 6.5000e-01, 5.3000e-01, 5.0000e-01, 3.9000e-01, 4.9000e-01,
6.6000e-01, 7.7000e-01, 4.6000e-01, 5.3000e-01, 6.0000e-01, 3.2000e-01,
3.2000e-01, 5.8000e-01, 5.2000e-01, 4.7000e-01, 4.1000e-01, 3.1000e-01,
4.8000e-01, 3.6000e-01, 5.1000e-01, 4.5000e-01, 1.1000e-01, 5.3000e-01,
4.0000e-01, 6.1000e-01, 3.1000e-01, 6.4000e-01, 5.3000e-01, 2.0000e-02,
3.2000e-01, 4.7000e-01, 4.6000e-01, 5.0000e-01, 3.6000e-01, 6.6000e-01,
2.5000e-01, 6.1000e-01, 6.1000e-01, 3.1000e-01, 7.9000e-01, 4.6000e-01,
4.8000e-01, 5.3000e-01, 4.7000e-01, 4.3000e-01, 5.1000e-01, 6.5000e-01,
5.7000e-01, 4.5000e-01, 3.5000e-01, 3.7000e-01, 5.6000e-01, 5.4000e-01,
6.2000e-01, 5.0000e-01, 5.6000e-01, 2.0000e-01, 6.0000e-01, 5.1000e-01,
3.7000e-01, 3.7000e-01, 3.5000e-01, 4.6000e-01, 6.4000e-01, 4.1000e-01,
5.1000e-01, 4.4000e-01, 3.3000e-01, 4.8000e-01, 7.3000e-01, 6.1000e-01,
5.7000e-01, 7.2000e-01, 4.6000e-01, 4.7000e-01, 7.9000e-01, 5.2000e-01,
3.6000e-01, 3.5000e-01, 6.9000e-01, 5.0000e-01, 4.5000e-01, 4.9000e-01,
8.8000e-01, 4.4000e-01, 6.8000e-01, 5.4000e-01, 4.4000e-01, 5.3000e-01,
7.1000e-01, 3.5000e-01, 3.4000e-01, 4.7000e-01, 3.8600e+00, 5.2000e-01,
4.0000e-01, 5.4000e-01, 4.1000e-01, 3.7000e-01, 4.0000e-01, 6.6000e-01,
5.7000e-01, 5.5000e-01, 5.9000e-01, 7.5000e-01, 6.6000e-01, 6.2000e-01,
4.8000e-01, 8.2000e-01, 4.1000e-01, 4.8000e-01, 7.1000e-01, 4.6000e-01,
5.1000e-01, 5.4000e-01, 3.1000e-01, 5.3000e-01, 4.7000e-01, 7.7000e-01,
4.2000e-01, 7.3000e-01, 6.2000e-01, 3.3000e-01, 3.9000e-01, 7.0000e-01,
4.5000e-01, 4.1000e-01, 6.0000e-01, 5.6000e-01, 5.0000e-01, 7.5000e-01,
2.8000e-01, 5.6000e-01, 4.4000e-01, 3.6000e-01, 4.2000e-01, 4.0000e-01,
2.9000e-01, 3.6000e-01, 5.0000e-01, 8.2000e-01, 5.8000e-01, 5.0000e-01,
6.3000e-01, 5.2000e-01, 3.1000e-01, 4.5000e-01, 3.6000e-01, 6.4000e-01,
2.2000e-01, 2.5000e-01, 5.6000e-01, 5.6000e-01, 6.9000e-01, 5.4000e-01,
4.6000e-01, 5.1000e-01, 5.0000e-01, 4.3000e-01, 6.3000e-01, 4.7000e-01,
7.2000e-01, 4.0000e-01, 5.8000e-01, 4.9000e-01, 5.4000e-01, 7.7000e-01,
5.7000e-01, 3.3000e-01, 4.3000e-01, 6.2000e-01, 5.1000e-01, 6.2000e-01,
4.7000e-01, 2.9000e-01, 0.0000e+00, 3.0000e-01, 3.4000e-01, 5.6000e-01,
6.4000e-01, 3.9000e-01, 7.2000e-01, 4.0000e-01, 5.0000e-01, 3.5000e-01,
4.0000e-01, 5.3000e-01, 3.7000e-01, 5.7000e-01, 5.2000e-01, 4.8000e-01,
5.1000e-01, 4.5000e-01, 3.6000e-01, 4.3000e-01, 4.8000e-01, 6.7000e-01,
6.5000e-01, 4.8000e-01, 4.6000e-01, 8.5000e-01, 5.7000e-01, 3.4000e-01,
4.1000e-01, 2.7000e-01, 4.6000e-01, 5.0000e-01, 5.4000e-01, 2.9000e-01,
5.8000e-01, 5.7000e-01, 4.1000e-01, 5.0000e-01, 4.4000e-01, 4.4000e-01,
4.6000e-01, 4.4000e-01, 2.9000e-01, 6.1000e-01, 7.0000e-01, 5.2000e-01,
4.1000e-01, 3.8000e-01, 3.2000e-01, 4.4000e-01, 4.7000e-01, 4.0000e-02,
4.1000e-01, 4.6000e-01, 4.7000e-01, 6.0000e-01, 6.4000e-01, 4.5000e-01,
5.1000e-01, 8.7000e-01, 0.0000e+00, 4.7000e-01, 5.5000e-01, 8.3000e-01,
3.3000e-01, 5.1000e-01, 3.4000e-01, 5.3000e-01, 6.4000e-01, 3.7000e-01,
5.3000e-01, 5.8000e-01, 9.1000e-01, 4.3000e-01, 6.0000e-01, 4.5000e-01,
7.7000e-01, 6.5000e-01, 6.7000e-01, 3.2000e-01, 5.6000e-01, 5.6000e-01,
6.5000e-01, 5.7000e-01, 6.2000e-01, 3.8000e-01, 5.2000e-01, 3.3000e-01,
3.2000e-01, 6.4000e-01, 5.1000e-01, 5.0000e-01, 3.0000e-01, 4.6000e-01,
6.9000e-01, 4.5000e-01, 2.9000e-01, 5.2000e-01, 3.1000e-01, 3.6000e-01,
5.8000e-01, 3.5000e-01, 5.9000e-01, 5.5000e-01, 5.6000e-01, 5.9000e-01,
3.6000e-01, 5.3000e-01, 8.4000e-01, 4.2000e-01, 4.7000e-01, 6.8000e-01,
5.6000e-01, 6.7000e-01, 5.4000e-01, 6.2000e-01, 0.0000e+00, 3.5000e-01,
6.5000e-01, 5.2000e-01, 4.3000e-01, 6.4000e-01, 5.1000e-01, 4.6000e-01,
5.5000e-01, 4.3000e-01, 6.0000e-01, 3.4000e-01, 0.0000e+00, 5.1000e-01,
5.0000e-01, 5.5000e-01, 3.4000e-01, 1.1500e+00, 4.7000e-01, 4.7000e-01,
4.0000e-01, 3.5000e-01, 5.6000e-01, 1.3000e-01, 6.5000e-01, 6.0000e-02,
3.6000e-01, 2.7000e-01, 5.7000e-01, 2.7000e-01, 5.7000e-01, 5.4000e-01,
5.6000e-01, 4.9000e-01, 5.6000e-01, 4.5000e-01, 4.7000e-01, 6.4000e-01,
4.2000e-01, 4.6000e-01, 4.5000e-01, 5.0000e-01, 7.6000e-01, 5.3000e-01,
5.7000e-01, 3.3000e-01, 5.5000e-01, 5.2000e-01, 5.2000e-01, 1.5000e-01,
5.9000e-01, 3.6000e-01, 3.3000e-01, 4.3000e-01, 5.0000e-01, 4.6000e-01,
3.3000e-01, 5.1000e-01, 4.4000e-01, 5.4000e-01, 5.9000e-01, 4.5000e-01,
5.4000e-01, 5.7000e-01, 2.9000e-01, 5.7000e-01, 0.0000e+00, 3.4000e-01,
5.9000e-01, 7.1000e-01, 4.6000e-01, 6.9000e-01, 6.4000e-01, 2.0000e-02,
3.6000e-01, 4.0000e-01, 5.4000e-01, 5.8000e-01, 5.1000e-01, 6.8000e-01,
5.0000e-01, 3.0000e-01, 2.8000e-01, 3.0000e-01, 4.4000e-01, 4.2000e-01,
6.2000e-01, 5.1000e-01, 4.6000e-01, 3.5000e-01, 4.0000e-01, 5.7000e-01,
6.1000e-01, 4.8000e-01, 5.4000e-01, 4.3000e-01, 3.0000e-01, 3.7000e-01,
2.2000e-01, 4.7000e-01, 4.7000e-01, 5.6000e-01, 4.0000e-01, 4.3000e-01,
4.8000e-01, 6.9000e-01, 6.7000e-01, 5.4000e-01, 4.6000e-01, 6.1000e-01,
6.5000e-01, 3.6000e-01, 4.1000e-01, 8.8000e-01, 4.2000e-01, 6.0000e-01,
4.4000e-01, 4.5000e-01, 5.2000e-01, 6.0000e-01, 5.5000e-01, 4.1000e-01,
5.3000e-01, 5.9000e-01, 5.0000e-01, 5.1000e-01, 6.0000e-01, 4.3000e-01,
3.5000e-01, 4.7000e-01, 3.4000e-01, 8.3000e-01, 4.8000e-01, 4.6000e-01,
6.5000e-01, 5.1000e-01, 5.2000e-01, 5.6000e-01, 4.4000e-01, 4.3000e-01,
5.3000e-01, 5.5000e-01, 4.7000e-01, 5.5000e-01, 5.0000e-01, 4.8000e-01,
5.1000e-01, 4.2000e-01, 2.3000e-01, 4.2000e-01, 4.0000e-01, 4.9000e-01,
5.2000e-01, 7.2000e-01, 3.9000e-01, 4.6000e-01, 3.8000e-01, 5.6000e-01,
5.4000e-01, 6.7000e-01, 4.0000e-01, 4.0000e-01, 6.6000e-01, 4.6000e-01,
5.5000e-01, 5.4000e-01, 3.9000e-01, 5.5000e-01, 5.2000e-01, 3.2000e-01,
4.2000e-01, 4.9000e-01, 5.6000e-01, 5.5000e-01, 4.3000e-01, 7.6000e-01,
3.4000e-01, 4.7000e-01, 3.7000e-01, 4.7000e-01, 5.2000e-01, 5.7000e-01,
5.2000e-01, 4.9000e-01, 3.9000e-01, 5.2000e-01, 3.4000e-01, 4.8000e-01,
6.3000e-01, 5.4000e-01, 5.1000e-01, 3.8000e-01, 6.0000e-01, 2.8000e-01,
5.3000e-01, 5.7000e-01, 4.6000e-01, 2.0000e-02, 2.9000e-01, 3.1000e-01,
4.6000e-01, 3.4000e-01, 5.5000e-01, 5.7000e-01, 3.9000e-01, 1.6000e-01,
5.1000e-01, 6.0000e-01, 3.5000e-01, 5.9000e-01, 6.0000e-01, 4.6000e-01,
3.7000e-01, 4.6000e-01, 5.4000e-01, 4.1000e-01, 4.2000e-01, 3.5000e-01,
6.7000e-01, 4.7000e-01, 6.4000e-01, 4.2000e-01, 5.3000e-01, 4.9000e-01,
3.5000e-01, 3.8000e-01, 5.7000e-01, 3.6000e-01, 4.0000e-01, 3.9000e-01,
3.8000e-01, 4.7000e-01, 4.1000e-01, 4.6000e-01, 6.0000e-01, 3.8000e-01,
5.5000e-01, 4.9000e-01, 3.6000e-01, 5.7000e-01, 6.0000e-01, 4.8000e-01,
6.5000e-01, 6.1000e-01, 1.8000e-01, 4.8000e-01, 3.2000e-01, 5.2000e-01,
4.6000e-01, 2.8000e-01, 4.5000e-01, 4.1000e-01, 3.7000e-01, 4.1000e-01,
5.1000e-01, 7.0000e-01, 4.3000e-01, 5.4000e-01, 5.2000e-01, 4.3000e-01,
6.0000e-01, 6.1000e-01, 3.2000e-01, 4.4000e-01, 6.3000e-01, 4.7000e-01,
4.2000e-01, 3.5000e-01, 5.5000e-01, 3.5000e-01, 3.9000e-01, 3.1000e-01,
4.8000e-01, 1.0000e-02, 8.4000e-01, 5.4000e-01, 3.0000e-01, 3.4000e-01,
4.8000e-01, 3.7000e-01, 3.0000e-01, 5.0000e-01, 4.8000e-01, 5.0000e-01,
5.4000e-01, 4.0000e-01, 3.7000e-01, 3.9000e-01, 4.4000e-01, 4.7000e-01,
7.1000e-01, 8.4000e-01, 4.5000e-01, 4.0000e-01, 2.7000e-01, 5.6000e-01,
4.7000e-01, 4.7000e-01, 5.5000e-01, 4.8000e-01, 7.4000e-01, 4.2000e-01,
5.3000e-01, 6.8000e-01, 4.5000e-01, 5.8000e-01, 6.1000e-01, 7.0000e-01,
3.8000e-01, 5.1000e-01, 5.6000e-01, 5.8000e-01, 5.0000e-01, 6.1000e-01,
3.7000e-01, 2.6000e-01, 5.3000e-01, 5.1000e-01, 5.7000e-01, 3.9000e-01,
4.8000e-01, 5.3000e-01, 4.7000e-01, 4.6000e-01, 4.7000e-01, 5.3000e-01,
3.4000e-01, 6.8000e-01, 6.1000e-01, 4.6000e-01, 4.0000e-01, 4.6000e-01,
1.3000e-01, 4.1000e-01, 4.9000e-01, 3.9000e-01, 5.3000e-01, 4.5000e-01])
When I use my custom weights and parameters of my fine-tuned gpt2 model I get similar output:
('fake', tensor([0.2400, 0.3700, 0.2300, 0.6100, 0.4700, 0.3400, 0.0000, 0.3500, 0.3900,
0.3800, 0.6100, 0.4200, 0.3100, 0.5600, 0.7200, 0.6100, 0.5900, 0.3200,
0.2700, 0.3700, 0.4700, 0.3400, 0.4100, 0.6300, 0.5900, 0.5700, 0.5700,
0.3600, 0.6400, 0.4000, 0.5900, 0.4100, 0.3400, 0.4700, 0.4100, 0.3400,
0.0000, 0.5700, 0.5700, 0.2300, 0.5800, 0.4800, 0.5900, 0.4800, 0.4600,
0.5500, 0.4500, 0.3900, 0.6300, 0.2700, 0.4200, 0.5000, 0.4100, 0.4000,
0.5000, 0.2900, 0.4500, 0.3600, 0.4500, 0.5100, 0.4000, 0.5700, 0.4000,
0.4000, 0.6000, 0.4800, 0.4800, 0.2200, 0.5000, 0.4000, 0.3200, 0.4500,
0.3600, 0.5300, 0.6600, 0.4600, 0.4200, 0.0200, 0.3100, 0.5100, 0.2800,
0.3700, 0.6600, 0.4100, 0.5700, 0.5600, 0.1400, 0.5900, 0.4900, 0.3200,
0.3200, 0.3600, 0.2900, 0.3200, 0.5500, 0.4800, 0.3200, 0.7400, 0.3900,
0.5900, 0.5600, 0.6800, 0.4100, 0.3400, 0.4900, 0.3700, 0.3900, 0.0000,
0.3900, 0.4400, 0.4200, 0.8900, 0.2900, 0.5200, 0.3100, 0.3500, 0.7400,
0.1400, 0.3100, 0.9700, 0.5300, 0.2700, 0.3000, 0.4500, 0.6500, 0.3900,
0.2700, 0.5100, 0.3000, 0.4900, 0.5700, 0.7500, 0.5400, 0.6100, 0.7100,
0.7000, 0.4600, 0.4300, 0.6000, 0.2600, 0.4100, 0.5500, 0.0100, 0.6900,
0.6600, 0.0100, 0.3800, 0.5300, 0.3900, 0.3300, 0.4900, 0.5200, 0.4300,
0.3300, 0.2800, 0.3400, 0.4900, 0.4200, 0.3700, 0.5000, 0.8500, 0.5800,
0.5100, 0.7100, 0.3900, 0.7700, 0.5600, 0.9300, 0.3400, 0.5500, 0.4500,
0.5400, 0.4600, 0.5400, 0.4000, 0.4300, 0.4700, 0.3400, 0.5300, 0.6200,
0.3700, 0.4400, 0.5100, 0.4300, 0.5500, 0.4100, 0.5900, 0.3600, 0.5200,
0.5700, 0.6200, 0.8100, 0.4600, 0.4000, 0.3900, 0.6000, 0.6400, 0.6100,
0.3200, 0.4600, 0.3400, 0.3600, 0.6900, 0.5200, 0.3900, 0.4000, 0.3900,
0.4500, 0.5800, 0.5400, 0.5800, 0.5900, 0.4900, 0.3100, 0.7600, 0.5700,
0.5200, 0.5000, 0.5500, 0.3600, 0.4100, 0.3900, 0.4600, 0.3100, 0.2100,
0.4500, 0.5300, 0.4900, 0.7200, 0.3900, 0.2800, 0.5800, 0.0000, 0.4800,
0.4400, 0.2900, 0.4000, 0.5000, 0.3600, 0.4400, 0.4400, 0.2700, 0.5000,
0.6000, 0.7800, 0.5400, 0.4200, 0.4700, 0.6200, 0.7000, 0.3900, 0.4000,
0.5700, 0.5600, 0.3200, 0.7000, 0.6100, 0.5600, 0.4200, 0.4000, 0.3700,
0.3100, 0.4300, 0.8100, 0.4200, 0.3100, 0.5900, 0.4100, 0.5400, 0.2500,
0.5000, 0.0200, 0.3400, 0.3900, 0.3700, 0.4300, 0.3500, 0.3200, 0.4500,
0.6000, 0.4100, 0.3600, 0.4500, 0.4100, 0.6400, 0.3600, 0.4200, 0.4600,
0.2100, 0.1900, 0.8700, 0.5300, 0.4800, 0.4100, 0.6100, 0.3200, 0.4300,
0.5400, 0.5200, 0.2800, 0.4400, 0.4700, 0.3400, 0.4000, 0.6100, 0.7400,
0.5500, 0.4000, 0.2700, 0.3700, 0.3700, 0.4400, 0.4200, 0.5200, 0.0000,
0.4600, 0.2700, 0.5500, 0.6200, 0.5800, 0.5400, 0.3300, 0.3200, 0.4600,
0.3900, 0.4100, 0.2900, 0.5900, 0.5600, 0.4200, 0.4200, 0.6600, 0.4900,
0.4200, 0.5000, 0.5000, 1.0200, 0.4200, 0.5000, 0.6300, 0.6700, 0.4400,
0.3700, 0.3600, 0.3000, 0.4200, 0.5600, 0.2900, 0.5100, 0.5700, 0.4500,
0.4300, 0.5300, 0.4300, 0.3000, 0.6300, 0.3000, 0.5100, 0.5100, 0.5100,
0.5200, 0.2000, 0.0500, 0.3400, 0.4200, 0.1200, 0.6500, 0.5200, 0.4100,
0.7100, 0.4900, 0.3900, 0.4100, 0.7700, 0.0000, 0.5700, 0.4800, 0.1800,
0.2100, 0.5400, 0.3600, 0.4500, 0.7600, 0.2400, 0.3800, 0.5600, 0.5800,
0.6400, 0.7000, 0.0300, 0.5700, 0.5700, 0.6100, 0.3800, 0.5100, 0.6200,
0.5100, 0.5700, 0.8100, 0.4400, 0.5000, 0.3400, 0.2100, 0.4100, 0.4500,
0.5700, 0.3600, 0.3700, 0.0300, 0.3200, 0.3000, 0.5100, 0.3800, 0.4400,
0.3700, 0.2500, 0.3900, 0.7200, 0.3600, 0.5500, 0.4900, 0.6000, 0.8900,
0.4400, 0.4400, 0.4900, 0.5900, 0.4400, 0.6900, 0.3400, 0.0000, 0.3500,
0.6300, 0.5400, 0.5300, 0.3500, 0.5400, 0.3200, 0.4300, 0.4500, 0.4100,
0.3100, 0.0000, 0.5400, 0.5400, 0.5000, 0.3600, 1.2000, 0.4500, 0.3200,
0.3500, 0.4200, 0.5600, 0.2400, 0.4300, 0.2400, 0.5800, 0.3900, 0.4900,
0.1300, 0.4300, 0.5300, 0.4200, 0.3600, 0.6700, 0.3300, 0.3700, 0.4600,
0.3300, 0.2400, 0.5000, 0.5100, 0.4300, 0.4300, 0.4900, 0.3700, 0.6100,
0.3600, 0.5100, 0.1600, 0.5900, 0.4100, 0.3300, 0.3500, 0.5200, 0.4600,
0.3800, 0.4500, 0.3400, 0.4400, 0.4900, 0.5100, 0.4900, 0.3700, 0.3900,
0.4300, 0.0000, 0.3100, 0.6400, 0.5300, 0.3300, 0.5300, 0.4400, 0.0500,
0.3500, 0.4500, 0.5100, 0.3900, 0.4300, 0.5700, 0.6500, 0.5200, 0.3700,
0.3000, 0.2800, 0.3900, 0.4800, 0.2000, 0.4900, 0.3100, 0.6400, 0.4500,
0.5100, 0.5400, 0.3200, 0.1400, 0.6300, 0.2900, 0.2600, 0.3300, 0.4100,
0.3900, 0.3900, 0.6400, 0.4500, 0.6400, 0.4500, 0.3500, 0.3900, 0.5500,
0.4700, 0.4700, 0.3800, 0.4100, 0.4000, 0.4000, 0.5800, 0.4000, 0.4600,
0.4400, 0.8700, 0.0000, 0.5500, 0.5200, 0.2200, 0.5300, 0.5500, 0.4900,
0.5100, 0.4700, 0.5000, 0.6400, 0.5300, 0.5800, 0.6000, 0.2700, 0.3000,
0.5800, 0.5000, 0.4400, 0.2400, 0.4700, 0.4300, 0.4000, 0.5600, 0.4200,
0.5100, 0.5000, 0.4400, 0.3200, 0.5100, 0.4900, 0.4200, 0.6900, 0.3800,
0.4700, 0.4100, 0.5400, 0.5000, 0.6300, 0.6300, 0.4500, 0.5400, 0.5000,
0.5000, 0.6400, 0.3800, 0.5800, 0.4900, 0.5100, 0.5300, 0.6000, 0.6100,
0.4600, 0.3700, 0.6100, 0.3900, 0.5500, 0.2400, 0.6400, 0.7300, 0.3600,
0.6700, 0.3600, 0.3700, 0.3700, 0.3100, 0.3200, 0.6500, 0.7200, 0.4600,
0.3500, 0.6100, 0.3500, 0.4500, 0.5900, 0.4100, 0.0400, 0.3100, 0.3200,
0.4500, 0.4000, 0.5600, 0.5400, 0.5500, 1.3600, 0.4900, 0.5700, 0.5100,
0.6400, 0.6300, 0.4400, 0.4100, 0.4000, 0.5900, 0.6600, 0.4200, 0.4200,
0.4700, 0.4500, 0.5600, 0.1800, 0.3200, 0.5200, 0.5200, 0.3800, 0.6000,
0.5000, 0.4600, 0.0200, 0.5300, 0.4000, 0.5600, 0.6100, 0.6000, 0.3200,
0.3700, 0.4900, 0.2600, 0.5800, 0.4200, 0.6100, 0.5100, 0.5400, 0.2100,
0.8000, 0.3200, 0.5700, 0.3900, 0.0000, 0.4900, 0.6600, 0.3600, 0.5600,
0.7400, 0.5000, 0.3600, 0.4400, 0.3500, 0.4100, 0.3900, 0.4800, 0.3500,
0.3400, 0.3900, 0.5800, 0.4100, 0.3900, 0.4700, 0.3000, 0.4100, 0.2300,
0.3700, 0.0000, 0.5000, 0.4500, 0.1900, 0.3700, 0.4100, 0.6400, 0.5000,
0.4200, 0.4800, 0.6500, 0.3600, 0.3400, 0.4100, 0.4100, 0.0100, 0.3500,
0.2700, 0.8200, 0.7100, 0.3400, 0.2300, 0.3800, 0.2600, 0.6200, 0.6500,
0.5300, 0.5200, 0.5700, 0.3100, 0.5400, 0.5700, 0.4600, 0.5200, 0.5300,
0.4400, 0.3300, 0.4400, 0.4900, 0.3500, 0.4800, 0.4300, 0.0600, 0.6200,
0.4400, 0.4800, 0.5800, 0.5000, 0.3100, 0.2800, 0.5100, 0.5400, 0.5200,
0.3100, 0.3400, 0.6700, 0.4000, 0.3400, 0.4600, 0.0100, 0.5700, 0.4700,
0.3900, 0.5400, 0.4100]
I don't understand what is causing this.
I expected outputs to be fake % and real % that equals to 1.
The code is:
def gpt_detector(query, onnx_path=ONNX_path):
start = time.time()
ort_session = onnxruntime.InferenceSession(onnx_path)
tokens_onnx = tokenizer.encode(query)
all_tokens = len(tokens_onnx)
tokens_onnx = tokens_onnx[:tokenizer.model_max_length - 2]
used_tokens = len(tokens_onnx)
input_ids_o = torch.tensor([tokenizer.bos_token_id] + tokens_onnx + [tokenizer.eos_token_id]).unsqueeze(0)
mask = torch.ones_like(input_ids_o)
with torch.no_grad():
ort_inputs = {ort_session.get_inputs()[0].name: to_numpy(input_ids_o)}
logits_onnx = ort_session.run(None, ort_inputs)
#ipdb.set_trace()
# logits = model(tokens.to(device), attention_mask=mask.to(device))[0]
logits_onnx_t = torch.tensor(logits_onnx[0])
probs_onnx = torch.softmax(logits_onnx_t, 1)
#ipdb.set_trace()
end = time.time()
total_time = np.round(end - start, decimals=2)
fake_onnx, real_onnx = probs_onnx[0][0], probs_onnx[0][1]
fake_onnx = np.round(fake_onnx*100, decimals=2)
real_onnx = np.round(real_onnx*100, decimals=2)
return "fake onnx ", fake_onnx, "real onnx ", real_onnx, "used tokens ", used_tokens, "total time ", total_time
What could be causing this?