pytorch
[RFC] Improve example discoverability
🚀 The feature
Improve discoverability of our exampels.
Motivation, pitch
We've heard from several sides now that our list of examples are confusing and hide valuable topics and pointers. So lets try to sort them in a different way to improve discoverability to users started easier without overwhelming them when browsing our examples.
I've grouped our current examples by category and introduced a new category for tutorials which span techniques that can be implemented with TorchServe like microbatching and DALI preprocessing. Each category would be its own folder and the README would show the hierarchy similar to the list shown below. We can either keep the tutorials under examples or create a new folder. Would be great to hear what you think about the categories or if there are other proposals.
- Tutorials Captum Workflows Captum Cloud_storage_stream_inference Cloudformation Custom_metrics Instruction_embedding Intel_extension_for_pytorch Micro_batching Nvidia_dali Stateful Pt2 Torch_tensorrt Torchrec_dlrm
- Basics Image_classifier Image_segmenter Text_classification object_detector
- Generative AI Diffusers Gpt_fast Segment_anything_fast Tp_llama LLM/llama2
- Speech Asr_rnnt_emformer Speech2text_wav2vec2 text_to_speech_synthesizer
- HuggingFace FasterTransformer_HuggingFace_Bert Huggingface_Transformers Huggingface_accelerate Huggingface_pippy
- DeepSpeed Deepspeed deepspeed_mii
- Legacy Dcgan_fashiongen Text_classification_with_scriptable_tokenizer MMF-activity-recognition
- Inferentia Llama2
- FAIR Nmt_transformer
- Benchmarking benchmarking/resnet50
Alternatives
No response
Additional context
See also https://github.com/pytorch/serve/issues/2040
