fix many bugs

[kroggen]

This PR fix many bugs:

• convert_awq_to_bin.py: the tensor must be copied from GPU to CPU before converting to numpy

This is the error message:

$ python3 ../llama_cu_awq/convert_awq_to_bin.py pytorch_model.bin files
model.embed_tokens.weight <class 'torch.Tensor'>
torch.Size([32000, 4096]) torch.float16
Traceback (most recent call last):
  File "/root/models/../llama_cu_awq/convert_awq_to_bin.py", line 31, in <module>
    f.write(value.numpy().tobytes())
TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

• gets_s and open_s are Windows specific
• \\ does not work on Linux, but / works on Windows as file separator (at least it used to)

If some of these are not OK, you can further fix them