0

The example code of DDPM works well in python. But after I convert it into tensorflowjs model, and run it in web browser, this line failed with error.

tf.einsum("bhwHW,bHWc->bhwc", attn_score, v)

Error: Found duplicate axes in input component bhwhw. Support for duplicate axes in input is not implemented yet.
    at Object.K_ [as decodeEinsumEquation] (einsum_util.js:95:13)
    at Object.kernelFunc (Einsum.ts:36:20)
    at n (engine.js:647:22)
    at engine.js:713:23
    at t.scopedRun (engine.js:480:19)
    at t.runKernelFunc (engine.js:709:10)
    at t.runKernel (engine.js:554:17)
    at einsum_ (einsum.js:107:17)
    at Object.einsum__op [as einsum] (operation.js:51:22)
    at matrices_executor.js:43:23

I doubt it is because incompability of tensorflowjs. Tried to rewrite this line without using einstein notation but always fails.

UPDATE:

This is a bug in tfjs, I have reported here: https://github.com/tensorflow/tfjs/issues/7404

Mr.Wang from Next Door
  • 13,670
  • 12
  • 64
  • 97

0 Answers0