Unstructured-IO / Unstructured-IO/unstructured-inference
feat: Apple Silicon support for Chipper Model
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 209
- Forks
- 73
- PR merge metrics
- No merged PRs in 30d
Description
As mentioned by @ajjimeno, the encoder is not available to MPS but the decoder is the bottleneck and can be run through a CUDA or MPS backend for GPU acceleration. This MPS backend is supported by the PyTorch framework. Pytorch backend support docs
It would just be to check if MPS is available, detach the encoder and decoder when detecting MPS instead of running model.generate, and map the computational graph of the decoder on the mps device. HugginFace example on MPS backend.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading the PyTorch backend support documentation and the MPS backend example linked in the issue, then locate the Chipper Model inference entry point and its encoder, decoder, and model.generate usage. Done means Apple Silicon can use MPS for the decoder while handling the encoder limitation, with existing inference behavior preserved.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100