haipipe-end-deploy-mlflow

name	haipipe-end-deploy-mlflow
description	MLflow deploy specialist for haipipe-end. STATUS: DEFERRED -- no platform-mlflow-inference repo backs it yet; placeholder. Would register an Endpoint_Set into an MLflow Model Registry and serve via `mlflow models serve`. Reads (never modifies) Endpoint_Sets from haipipe-end-endpointset; the umbrella's no-args dashboard skips it while deferred.
argument-hint	[function] [endpoint_set_or_id] [args...]
allowed-tools	Bash, Read, Write, Edit, Grep, Glob
metadata	{"version":"0.1.2","last_updated":"2026-07-08","summary":"MLflow deploy specialist for haipipe-end."}

Skill: haipipe-end-deploy-mlflow

MLflow registry + serving specialist. Wraps an Endpoint_Set into an MLflow pyfunc model, logs and registers it, and (optionally) serves it via mlflow models serve. The registered model can also be the input artifact for downstream MLflow-aware deployers.

Status: DEFERRED — SKILL.md scaffolded as a placeholder; not yet exercised (no backing repo). Tracking server URI, registry name, and stage transition policy below are placeholders for the project to fill in.

Function axis: dashboard | deploy | test | monitor | teardown | review

---

Commands

/haipipe-end-deploy-mlflow                              -> dashboard: registered models + versions
/haipipe-end-deploy-mlflow dashboard                    -> same
/haipipe-end-deploy-mlflow deploy <endpoint_set>        -> log + register, optionally `mlflow models serve`
/haipipe-end-deploy-mlflow deploy <es> --register-only  -> register but do not start a server
/haipipe-end-deploy-mlflow test <endpoint_id>           -> hit local `mlflow models serve` server
/haipipe-end-deploy-mlflow monitor <endpoint_id>        -> tail mlflow serve logs
/haipipe-end-deploy-mlflow teardown <endpoint_id>       -> stop server, archive model version
/haipipe-end-deploy-mlflow review <endpoint_id>         -> audit registered model + signature

---

Dispatch Table

Invocation     Ref file(s)                              Function block
-------------- ---------------------------------------- -----------------------------------
dashboard      ../haipipe-end/ref/deploy-overview.md                          dashboard procedure
deploy         ../haipipe-end/ref/deploy-overview.md +
               ../haipipe-end/ref/
                 0-overview.md                          deploy procedure
test           ../haipipe-end/ref/deploy-overview.md                          test procedure
monitor        ../haipipe-end/ref/deploy-overview.md                          monitor procedure
teardown       ../haipipe-end/ref/deploy-overview.md                          teardown procedure
review         ../haipipe-end/ref/deploy-overview.md                          review procedure

---

Step-by-Step Protocol

Step 0: Read ../haipipe-end/ref/deploy-overview.md for MLflow tracking URI, registry conventions, and stage transition policy.

Step 1: Parse args. Required arg per function: deploy: <endpoint_set_name> [--register-only] [--port N] test/monitor/teardown/review: <endpoint_id> (model name + version)

Step 2: Verify MLflow context: - MLFLOW_TRACKING_URI set - Registry credentials available (if using a remote backend)

Step 3: Execute the function.

Step 4: Emit the structured tail:

status:    ok | blocked | failed
summary:   2-3 sentences on the deploy / test / etc.
artifacts: [model URI, model name + version, serve URL if started]
next:      suggested next command

---

Procedures (placeholder — fill from project conventions)

Deploy:

1. Read Endpoint_Set at _WorkSpace/6-EndpointStore/<endpoint_set>/. (input contract, all deploy skills: canonical input = the folder; a .tar.gz twin is a wire form only)
2. Wrap fn_endpoint/ + ModelInstance into an mlflow.pyfunc.PythonModel. (logical bundle name; physically materialized as code/ + model/ in the set)
3. mlflow.start_run() and log:
   • The pyfunc model with signature inferred from Endpoint_Set's manifest.
   • The Endpoint_Set's manifest as a run artifact.
   • Hyperparameters / training config from ModelInstance.
4. Register the logged model into the MLflow Model Registry under <endpoint_set> name.
5. Transition new version to the configured stage (e.g. Staging).
6. If not --register-only: spawn mlflow models serve on the chosen port.
7. Smoke-test the served endpoint or the registry record.
8. Record endpoint_id (model_name + version) in the project's deploy log.

Test, Monitor, Teardown, Review: See ../haipipe-end/ref/deploy-overview.md for the mlflow CLI patterns the project uses.

---

Target Scope

Owns:

• MLflow pyfunc wrapping of the Endpoint_Set
• MLflow tracking run + registered model version
• Stage transitions (Staging / Production / Archived)
• mlflow models serve local serving
• Model URI emission for downstream MLflow-aware deployers

Does NOT own:

• Endpoint_Set content (read-only input from /haipipe-end-endpointset)
• Downstream platform deploy (Databricks Model Serving uses /haipipe-end-deploy-databricks even though it consumes MLflow registry under the hood — keep concerns split for clarity)

If a deploy fails because of an Endpoint_Set issue, escalate to /haipipe-end-endpointset review.