Repository navigation
Commit 1c7b1fc
committed
feat(serve): support fine-tuned models in deployment-config API
Make ModelBuilder.list_deployment_configs / get_deployment_config /
set_deployment_config work for fine-tuned (model-customization) models
in addition to base/JumpStart models, dispatching internally on
_is_model_customization(). For fine-tuned models the recipe's published
HostingConfigs are the source of truth; selection is by instance type
(recipe configs are largely unnamed).
The base "deployment config" vs recipe "hosting config" distinction is
internal and does not surface: fine-tuned configs are normalized to the
same shape the base response returns (DeploymentConfigName plus a nested
DeploymentArgs block with the base DeploymentArgs keys, present as None
when a recipe does not provide them — a compatible superset), so a
caller can iterate results from either pathway identically.
BenchmarkMetrics/AccelerationConfigs are None; an additive IsDefault
flag marks the recipe's Default config.
set_deployment_config(instance_type=...) applies the whole matching
config (image, environment, compute requirements) at build time and
raises on an unpublished or ambiguous instance type. A caller-provided
image_uri still takes precedence over the config's ImageUri (documented
on the method). The base/JumpStart branch of the setter continues to
require both config_name and instance_type — the instance_type-only
form is fine-tuned-only.
list_deployment_configs(instance_type=...): for fine-tuned models this
filters the recipe's per-instance bundles to the requested instance; for
base/JumpStart models the requested instance is threaded into JumpStart's
own config resolution rather than post-filtered, so a config that
supports the requested instance but is materialized at a different
default is not wrongly discarded. Configs published with only
DefaultInstanceType are matched consistently end to end so they are not
silently dropped at build.
Config discovery (recipe-level with a top-level HostingConfigs fallback)
is centralized in _extract_hosting_configs_from_hub() and used by BOTH
the selection API and the build path, so a top-level config that is
listable/selectable is also applied at build. Nova models are routed to
their dedicated build path first (per-tier SMI validation + Nova env
precedence), so the top-level fallback never diverts them. An explicit
selection stores a deep copy of the raw config, is applied exactly at
build (or raises if its instance type is no longer published), and the
endpoint instance type is re-synced to the selected bundle. Returned
configs are copied so caller mutation cannot corrupt internal state.
This is the SDK half of a paired change; the SageMaker agent
model-deployment skill consumes this unified API. Internal CR:
https://code.amazon.com/reviews/CR-2899036431 parent 3f810c6 commit 1c7b1fc
2 files changed
Lines changed: 1286 additions & 46 deletions
File tree
- sagemaker-serve
- src/sagemaker/serve
- tests/unit
0 commit comments