expected Tensor, but got tuple

[knight-captain]

After updating Auto1111 a couple weeks ago, EM stopped working. It can analyze the tokens, but when it tries to save an embedding I get the following stack trace:

Traceback (most recent call last):
  File [file path]"\stable-diffusion-webui-embedding-merge\scripts\embedding_merge.py", line 1231, in need_save_embed
    vectors = [torch.cat([r[0] for r in pair[0]])]
TypeError: expected Tensor as element 0 in argument 0, but got tuple

[aleksusklim]

Can you remove the extension (delete stable-diffusion-webui\extensions\stable-diffusion-webui-embedding-merge\ folder) and reinstall it again with default settings?

I could not see the error with updated WebUI dev branch. If you still can trigger it with updated EM (on sdxl branch) and latest A1111, then tell me which model do you use (SD1, SD2 or SDXL)?

[victorca25]

Same issue, I just installed a fresh version of the repo for the first time an got the error on Forge

[victorca25]

In case it's useful, the problem is that the structure of pair is not correct. According to the code, it expects that r[0] is the token vector, but instead it's a tuple of (tensor, str, list), so you would need to use: vectors = [torch.cat([r[0][0] for r in pair[0]])] After this change, it works fine

[aleksusklim]

There is no way the existing code can be incorrect.

Something is probably off in inputs? Or it takes a strange codepath.

I'll try installing Forge from scratch to see this.

[zethfoxster]

going to add to this TypeError: expected Tensor as element 0 in argument 0, but got tuple it happens only when saving a "by comma" embedding

[zethfoxster]

In case it's useful, the problem is that the structure of pair is not correct. According to the code, it expects that r[0] is the token vector, but instead it's a tuple of (tensor, str, list), so you would need to use: vectors = [torch.cat([r[0][0] for r in pair[0]])] After this change, it works fine

this does indeed fix the issue completely