rocca commited on
Commit
f79fe3a
1 Parent(s): fd15049

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -2
README.md CHANGED
@@ -7,10 +7,11 @@ license: mit
7
 
8
  ### Notes:
9
  * Working tflite conversion: https://colab.research.google.com/gist/josephrocca/f427377f76c574f1c1e8e4d6d62c34b6/tflite-dalle-mini-conversion-separated-encoder-and-decoder.ipynb
10
- * Note that the encoder and decoder need to be converted separately for some reason. More info: https://github.com/tensorflow/tensorflow/issues/56629#issuecomment-1172928973
11
  * But these models currently require TF Select operators due to bitwise operations that aren't supported in the default tflite opset. [This issue](https://github.com/tensorflow/tfjs/issues/5844) tracks the potential inclusion of TF Select ops in tfjs-tflite, so models like this can be used on the web.
12
- * The encoder requires the TF Select `Erf` op. The decoder requires these TF Select ops: Bitcast, BitwiseOr, BitwiseXor, ConcatV2, Erf, LeftShift, RightShift, ScatterNd, SelectV2, StridedSlice
13
  * Here are the two models:
14
  * https://huggingface.co/rocca/dalle-mini-js/blob/main/debug/dalle_bart_encoder_flax_model.tflite
15
  * https://huggingface.co/rocca/dalle-mini-js/blob/main/debug/dalle_bart_decoder_flax_model.tflite
16
  * ONNX conversion is blocked on support for `Bitcast`, `BitwiseOr` and `BitwiseXor`: https://github.com/onnx/tensorflow-onnx/issues/1985 The need for the `BitwiseAnd` op can be avoided by pulling `jax.random.PRNGKey` outside of the encode-decode function, but I'm not sure where the other ops are coming from - netron doesn't seem to "expand" certain types of nodes, so I think they're hidden within some other nodes.
 
 
7
 
8
  ### Notes:
9
  * Working tflite conversion: https://colab.research.google.com/gist/josephrocca/f427377f76c574f1c1e8e4d6d62c34b6/tflite-dalle-mini-conversion-separated-encoder-and-decoder.ipynb
10
+ * Note that the encoder and decoder need to be converted separately for some reason. More info on this bug: https://github.com/tensorflow/tensorflow/issues/56629#issuecomment-1172928973
11
  * But these models currently require TF Select operators due to bitwise operations that aren't supported in the default tflite opset. [This issue](https://github.com/tensorflow/tfjs/issues/5844) tracks the potential inclusion of TF Select ops in tfjs-tflite, so models like this can be used on the web.
12
+ * The encoder requires the TF Select `Erf` op. The decoder requires these TF Select ops: `Bitcast`, `BitwiseOr`, `BitwiseXor`, `ConcatV2`, `Erf`, `LeftShift`, `RightShift`, `ScatterNd`, `SelectV2`, `StridedSlice`
13
  * Here are the two models:
14
  * https://huggingface.co/rocca/dalle-mini-js/blob/main/debug/dalle_bart_encoder_flax_model.tflite
15
  * https://huggingface.co/rocca/dalle-mini-js/blob/main/debug/dalle_bart_decoder_flax_model.tflite
16
  * ONNX conversion is blocked on support for `Bitcast`, `BitwiseOr` and `BitwiseXor`: https://github.com/onnx/tensorflow-onnx/issues/1985 The need for the `BitwiseAnd` op can be avoided by pulling `jax.random.PRNGKey` outside of the encode-decode function, but I'm not sure where the other ops are coming from - netron doesn't seem to "expand" certain types of nodes, so I think they're hidden within some other nodes.
17
+ * TF.js conversion is blocked on lack of support for these ops: `BitwiseXor`, `Bitcast`, `LeftShift`, `BitwiseOr`, `RightShift`