Comments (11)
Hmm ... fair point. I could that we register a new version only with the intention of using it. We have an API to set default version, in what you are proposing assuming I register a new version only with the intention of using it versus having it present in the model cache. In this case, every new version that I register will then also have to be registered as the new default model.
This flow:
- cur default = 1.0
- register version = 1.1 (cur default = 1.1)
- A/B testing still possible
That Flow:
- cur default = 1.0
- register version = 1.1 (cur default = 1.0)
- set default = 1.1 (cur default = 1.1)
Question boils down to:
"This Flow or That Flow?"
FWIW current impl is less intuitive.
from serve.
I would go with That Flow™. My thinking around this is that the endpoint will be serving models in production, but people will want to benchmark, A/B test, or otherwise evaluate new model versions before switching them to production use, and they won't want to take down a working production system in the meantime.
from serve.
Fix available in #51 branch do_not_set_default_with_register
.
from serve.
renamed branch to issue_43
. New PR is : #151
from serve.
There's a bug in the PR where the model version string is cast to a Double.
from serve.
I tested the stage_release
branch for comparison, and it is also stopping on an exception (which, by the way, does not show a call stack, nor show up in ts_log.log
).
The version number must either consistently be a string (and treated as such in all places in code), or it needs to be a double, and that needs to be enforced in the model archiver. (Right now, the model archiver takes any string for a version.) With the common version number schemes that people use, IMO a string makes a lot more sense.
from serve.
Now TorchServe accepts model version as string. The fix is available in the same PR #151
from serve.
One change to unit tests requested on #151
from serve.
Updated the version to represent a string.
from serve.
This looks good, and the unit tests cover the new case of non-Double-parseable model version strings. Please close this issue after merging.
from serve.
PR #171 merged to master
from serve.
Related Issues (20)
- Warning in regression test: test_gRPC_inference_api.py
- Warning in regression test: test_install_dependencies_to_target_directory_with_requirements
- enable test_install_dependencies_to_venv_with_requirements in docker regression HOT 2
- Cannot run the text_classification example HOT 4
- CPP build failed with errors
- Open Inference Protocol with nightly build not working HOT 6
- CPP backend debugging and troubleshooting
- KServe wrapper default configuration is faulty HOT 5
- Update documentation on deprecating mac x86 support
- It seems like `metrics.yaml` doesn't apply HOT 1
- Config to disable gpu system metrics collection HOT 1
- torch.compile benchmark nightlies failing because of dependency of simpy
- '503 Service Unavailable' for url 'http://0.0.0.0:8085/v1/models/mnist:predict' HOT 2
- Update token authentication doc with maven link for downloading prebuilt plugin HOT 1
- KServe nightly tests are failing
- Incomplete example about emitting metrics HOT 2
- Broken example for a custom Counter metrics HOT 2
- Metrics REST API doesn't return custom metric HOT 3
- Model results are inconsistent between preheating and after preheating HOT 1
- Model with name <name> is not ready. HOT 22
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from serve.