no-archive model cannot be loaded

Context

• torchserve version: 0.2.0
• torch version: 1.6.0
• java version: 11.0.9
• Operating System and version: GNU/Linux

Your Environment

• Installed using source? [no]:
• Are you planning to deploy it using docker container? [no]:
• Is it a CPU or GPU environment?: GPU
• Using a default/custom handler? [custom]:
• What kind of model is it e.g. vision, text, audio?: text
• Are you planning to use local models from model-store or public url being used e.g. from S3 bucket etc.? [If public url then provide link.]: local

Expected Behavior

No archive model should be able to be loaded

Current Behavior

Model not found error when loading a no-archive model

Steps to Reproduce

1. Create a no-archive version of the model using model-archiver

   torch-model-archiver --model-name "BartQgen" --version 1.0 --serialized-file models/sqgen/pytorch_model.bin --handler sqgen/handler.py --extra-files "models/sqgen/config.json,models/sqgen/special_tokens_map.json,models/sqgen/tokenizer_config.json,models/sqgen/vocab.json,models/sqgen/merges.txt" --archive-format no-archive --export-path model_store

2. Try to load it using torchserve: torchserve --start --model-store model_store --models sqgen=BartQgen --no-config-snapshots

Failure Logs [if any]

2020-10-09 20:12:08,346 [INFO ] main org.pytorch.serve.ModelServer - Loading initial models: BartQgen
2020-10-09 20:12:08,348 [WARN ] main org.pytorch.serve.ModelServer - Failed to load model: BartQgen
org.pytorch.serve.archive.ModelNotFoundException: Model not found at: BartQgen
        at org.pytorch.serve.archive.ModelArchive.downloadModel(ModelArchive.java:84)
        at org.pytorch.serve.wlm.ModelManager.createModelArchive(ModelManager.java:137)
        at org.pytorch.serve.wlm.ModelManager.registerModel(ModelManager.java:115)
        at org.pytorch.serve.ModelServer.initModelStore(ModelServer.java:213)
        at org.pytorch.serve.ModelServer.start(ModelServer.java:308)
        at org.pytorch.serve.ModelServer.startAndWait(ModelServer.java:104)
        at org.pytorch.serve.ModelServer.main(ModelServer.java:85)

I also tried providing the absolute path to the model. That doesn’t work either.

[reubenwenisch]

Can you try and comment on what you get torchserve --start --model-store model_store --models sqgen=BartQgen.mar --no-config-snapshots

[Ark-kun]

The change breaks the error reporting. It’s very confusing when the system tells you that file does not exist even though it clearly does.