Pengchong Jin

mask r-cnn model under official/vision/detection is undergoing some major refactor. We will release the new implementation which supports exporting to saved model and tflite model later.

Hi there, unfortunately, this repo doesn't provide finetuning features. Here are a few alternatives that might fit your needs: 1. On [Gemma](https://console.cloud.google.com/vertex-ai/publishers/google/model-garden/335) model card in Vertex Model Garden, there are...

@Khajaamee455 Thanks for the contribution, config.py only contains dataclass so I feel this is a little un-necessary. Closing it for now.

Hi, which model variant did you use? On which platform? Can you please share the command to reproduce this?

Could you please try the instruction-tuned model instead? It should give you better results.

It's kind of expected that the pre-trained models only try to complete text. Maybe one way you could try is to tune the sampling parameters to see if you can...

Hi @markusheimerl , it seems you are using v4-16 TPU which 2 host VMs. This multi-host setup is currently not supported. To test it on TPU, I suggest you try...

Unfortunately, we haven't convert pytorch checkpoint to onnx or tflite ourselves. Maybe you can follow the standard guides `torch.onnx`: https://pytorch.org/docs/stable/onnx.html and `tf-lite` to try the conversion.

Hi @Cguanqin, have you downloaded the checkpoint from Kaggle or HuggingFace Hub before running the docker? You can follow the instruction here to get the checkpoint. https://www.kaggle.com/models/google/gemma/frameworks/pyTorch

Thanks for the contribution! To clarify, have you included changes for RoPE embedding dtype issue that is mentioned above? I only see GeLU fix.