Add back inference.inference API #126601
Add back inference.inference API #126601
Conversation
pquentin
added
>non-issue
:ml
|
Pinging @elastic/ml-core (Team:ML) |
| "url": { | ||
| "paths": [ | ||
| { | ||
| "path": "/_inference/{inference_id}", |
There was a problem hiding this comment.
I think in most places we're directing users to include the task_type in the URL. Would it cause problems if we remove this path?
There was a problem hiding this comment.
Allowing to call the Inference API without a task type is how Kibana deploys a model. Enforcing the task type apparently makes things more complex here.
| } | ||
| }, | ||
| { | ||
| "path": "/_inference/{task_type}/{inference_id}", |
There was a problem hiding this comment.
Do you need this form as it is covered by inference.text_embedding, inference.sparse_embedding etc?
There was a problem hiding this comment.
I mainly wanted to include it for backward compatibility. It could prove helpful if a new task type is added to Elasticsearch but not added immediately in the specification, but a missing task type sounds much less likely than a missing provider. We could deprecate this form.
There was a problem hiding this comment.
It could prove helpful if a new task type is added to Elasticsearch but not added immediately in the specification
True, let's leave it in for now if it becomes problematic it can be removed without breaking any existing usage as the inference.{TASK_TYPE} cases cover everything.
| } | ||
| }, | ||
| { | ||
| "path": "/_inference/{task_type}/{inference_id}", |
There was a problem hiding this comment.
It could prove helpful if a new task type is added to Elasticsearch but not added immediately in the specification
True, let's leave it in for now if it becomes problematic it can be removed without breaking any existing usage as the inference.{TASK_TYPE} cases cover everything.
See elastic/elasticsearch-specification#4256 for details.