CLIPTorchEncoder "expected scalar type Float but found Half" error

[delgermurun]

I am trying to use CLIP executors for Dalle-Flow.
CLIPTorchEncoder doesn't work.

  - name: clip_encoder
    uses: jinahub+docker://CLIPTorchEncoder/latest-gpu
    uses_with:
      name: ViT-L-14-336::openai

status {
  code: ERROR
  description: "RuntimeError(\'expected scalar type Float but found Half\')"
  exception {
    name: "RuntimeError"
    args: "expected scalar type Float but found Half"
    stacks: "Traceback (most recent call last):\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/jina/serve/runtimes/worker/__init__.py\", line 164, in process_data\n    return await self._data_request_handler.handle(requests=requests)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/jina/serve/runtimes/request_handlers/data_request_handler.py\", line 155, in handle\n    return_data = await self._executor.__acall__(\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/jina/serve/executors/__init__.py\", line 291, in __acall__\n    return await self.__acall_endpoint__(__default_endpoint__, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/jina/serve/executors/__init__.py\", line 310, in __acall_endpoint__\n    return await func(self, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/jina/serve/executors/decorators.py\", line 207, in arg_wrapper\n    return await fn(executor_instance, *args, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/executors/clip_torch.py\", line 141, in encode\n    self._model.encode_text(**batch_data)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/model/openclip_model.py\", line 48, in encode_text\n    return self._model.encode_text(input_ids)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/model/model.py\", line 573, in encode_text\n    x = self.transformer(x, attn_mask=self.attn_mask)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py\", line 1130, in _call_impl\n    return forward_call(*input, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/model/model.py\", line 328, in forward\n    x = r(x, attn_mask=attn_mask)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py\", line 1130, in _call_impl\n    return forward_call(*input, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/model/model.py\", line 297, in forward\n    x = x + self.attention(self.ln_1(x), attn_mask=attn_mask)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/clip_server/model/model.py\", line 294, in attention\n    return self.attn(x, x, x, need_weights=False, attn_mask=attn_mask)[0]\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py\", line 1130, in _call_impl\n    return forward_call(*input, **kwargs)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/modules/activation.py\", line 1153, in forward\n    attn_output, attn_output_weights = F.multi_head_attention_forward(\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/functional.py\", line 5179, in multi_head_attention_forward\n    attn_output, attn_output_weights = _scaled_dot_product_attention(q, k, v, attn_mask, dropout_p)\n"
    stacks: "  File \"/usr/local/lib/python3.8/dist-packages/torch/nn/functional.py\", line 4852, in _scaled_dot_product_attention\n    attn = torch.baddbmm(attn_mask, q, k.transpose(-2, -1))\n"
    stacks: "RuntimeError: expected scalar type Float but found Half\n"
    executor: "CLIPEncoder"
  }
}

---

CLIPOnnxEncoder works fine

  - name: clip_encoder
    uses: jinahub+docker://CLIPOnnxEncoder/latest-gpu
    uses_with:
      name: ViT-L-14-336::openai