optimum
optimum copied to clipboard
🚀 Accelerate training and inference of 🤗 Transformers and 🤗 Diffusers with easy to use hardware optimization tools
# What does this PR do? Fixed Key Error "last_hidden_state" bug when generating the embeddings using optimized models. Fixes # 1673
# What does this PR do? Adds the --fuse_lora command which can optionally fuse a specified LoRA with the ONNX model for Stable Diffusion. This allows easily exporting fine tuned...
### Feature request I have been trying to optimize a mixtral model trained for text-classification, into its ONNX counterpart. I have tried it using torch level extraction - torch.onnx.export( model,...
### System Info ```shell optimum version: 405199457d9f6b8e060c043216eb717bc7a4c4c1 Platform Linux: Python version: 3.11 ``` ### Who can help? @SunMarc @younesbelkada @fxmarty ### Information - [ ] The official example scripts -...
### Feature request Hi all, I am trying to convert Table-transformer model from transformers(pretrained) to ONNX. Error reads something like " 'table-transformer' is not a supported format. Is there any...
### Feature request Hi, I went through optimum's code base and could not find support for object detection models. Is there plan to add ORTModelForObjectDetection just like ORTModelForImageClassification exists? Would...
### Feature request Musicgen was recently added to 🤗 Transformers ([model doc](https://huggingface.co/docs/transformers/model_doc/musicgen)) and it would be great to be able to export those models to ONNX with Optimum. ### Motivation...
# What does this PR do? Fixes # (issue) @fxmarty I couldn't find a tiny model to test with - once we have that, this should be ready to merge...
# What does this PR do? Not sure if this is actually needed, since DiT uses the same architecture as BeiT, but it adds it to the list 👍 Fixes...
# What does this PR do? This PR adds support for exporting SigLIP models to ONNX. WIP for [this issue](https://github.com/xenova/transformers.js/issues/430). ## Before submitting - [ ] This PR fixes a...