Skip to content

Latest commit

 

History

History
16 lines (13 loc) · 1.16 KB

todo.md

File metadata and controls

16 lines (13 loc) · 1.16 KB
Raw
  • should not put the model in a docker container, use the file storage instead and make it available as a volume to the container
  • use a model registry to store models, build one with mlflow.
  • Run different services for each model, and use a load balancer to route the requests to the right model.

torch-model-archiver --model-name MasaknaneEnSwaRelNews
--version 1.0
--serialized-file src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/pytorch_model.bin
--handler src/torchserve/transformer_handler.py
--extra-files "src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/config.json, src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/special_tokens_map.json, src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/tokenizer_config.json, src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/vocab.json, src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/generation_config.json, src/torchserve/transformer_models/masakhane/m2m100_418M_en_swa_rel_news/sentencepiece.bpe.model"