[quantization] Packed sequence support for dynamic quantized RNN #27963
Description
🐛 Bug
Currently quantized LSTM supports only non-packed sequence, we should fix it.
To Reproduce
import torch
from torch.nn.utils.rnn import pack_sequence
# quantized apis require a container, not a top-level module - thus sequential
m = torch.nn.Sequential(torch.nn.LSTM(2,5,num_layers=2))
x = pack_sequence([torch.rand(4,2), torch.rand(3,2), torch.rand(2,2)])
m(x) # works
qm = torch.quantization.quantize_dynamic(m, dtype=torch.qint8)
qm(x)
Today it fails with AssertionError: assert batch_sizes is None.