Port CLIP tokenizer which leverages byte-level BPE. This tokenizer enables scenarios like StableDiffusion May be dependent on https://github.com/dotnet/machinelearning/issues/6992. Reference: https://huggingface.co/docs/transformers/main/en/model_doc/clip https://github.com/huggingface/transformers/blob/0549000c5bf6c7249f411917f2a6f0b6d0f06da1/src/transformers/models/codegen/tokenization_codegen.py#L98 https://onnxruntime.ai/docs/tutorials/csharp/stable-diffusion-csharp.html#tokenization-with-onnx-runtime-extensions Paper: https://arxiv.org/abs/2103.00020 https://arxiv.org/pdf/2103.00020.pdf
Activity
ericstj commentedon Mar 18, 2024
Note - ONNX sample doesn't require separate tokenizer.
@LittleLittleCloud might need this for a solution that works with torchsharp.
rosebyte commentedon Sep 23, 2026
Triage: we may consider adding this if we will have time.