-
Full CPU ✔️ :
nnls: nlls: tensor([1.9837, 2.3163, 2.3380])
Perplexity: 9.14018440246582
-
Full GPU ✔️ :
nlls: tensor([1.9834, 2.3164, 2.3379], device='cuda:0', dtype=torch.float16)
Perplexity: 9.140625
-
Split 🔴 :
nnls: tensor([24.0938, 23.8594, 24.2812], device='cuda:0', dtype=torch.float16)
Perplexity: 28641570816.0