diff --git a/.github/actions/install-ci-dependencies/action.yml b/.github/actions/install-ci-dependencies/action.yml
index e84c949..b03d8bc 100644
--- a/.github/actions/install-ci-dependencies/action.yml
+++ b/.github/actions/install-ci-dependencies/action.yml
@@ -37,28 +37,37 @@ runs:
     # - requirements.txt excludes torch (dependencies resolved against nightly)
     # - Skip on macOS ARM64 - nightly +cpu wheels aren't available, use stable torch
     # - Skip for "oldest" builds - they use stable torch from lock file
-    - name: Install torch nightly
+    - name: Install torch prerelease (if configured)
       if: inputs.use_oldest != 'true'
       shell: bash
       run: |
-        NIGHTLY_FILE="requirements/ci/torch-nightly.txt"
-        if [[ -f "${NIGHTLY_FILE}" ]]; then
-          TORCH_VERSION=$(grep -v '^#' "${NIGHTLY_FILE}" | grep -v '^$' | head -1 || true)
-          if [[ -n "${TORCH_VERSION}" ]]; then
+        PRE_FILE="requirements/ci/torch-pre.txt"
+        if [[ -f "${PRE_FILE}" ]]; then
+          # Read 3-line config: version, CUDA target, channel type
+          # Use while loop for bash 3.2 compatibility (macOS)
+          PRE_CONFIG=()
+          while IFS= read -r line; do
+            PRE_CONFIG+=("$line")
+          done < <(grep -v '^#' "${PRE_FILE}" | grep -v '^$')
+          TORCH_VERSION="${PRE_CONFIG[0]}"
+          CUDA_TARGET="${PRE_CONFIG[1]}"
+          CHANNEL_TYPE="${PRE_CONFIG[2]}"
+
+          if [[ -n "${TORCH_VERSION}" && -n "${CHANNEL_TYPE}" ]]; then
             if [[ "${{ runner.os }}" == "macOS" ]]; then
-              echo "Skipping torch nightly on macOS ARM64 (no +cpu wheels available)"
+              echo "Skipping torch ${CHANNEL_TYPE} on macOS ARM64 (no +cpu wheels available)"
               echo "Will install stable torch with --torch-backend=auto"
             else
-              echo "Installing torch nightly: ${TORCH_VERSION}+cpu"
+              echo "Installing torch ${CHANNEL_TYPE}: ${TORCH_VERSION}+cpu"
               uv pip install --prerelease=allow "torch==${TORCH_VERSION}+cpu" \
-                --index-url https://download.pytorch.org/whl/nightly/cpu
+                --index-url "https://download.pytorch.org/whl/${CHANNEL_TYPE}/cpu"
             fi
           fi
         fi
 
     # Install FTS and all dependencies
     # - UV_OVERRIDE: applies Lightning commit pin from overrides.txt
-    # - When nightly configured (and not macOS): torch already installed, just install rest
+    # - When prerelease configured (and not macOS): torch already installed, just install rest
     # - For macOS: use --torch-backend=auto for MPS-compatible stable torch
     # - For stable builds: use --torch-backend=cpu
     - name: Install FTS and all dependencies
@@ -74,27 +83,32 @@ runs:
           echo "Installing with latest versions..."
         fi
 
-        # Check if torch nightly is configured
-        NIGHTLY_FILE="requirements/ci/torch-nightly.txt"
-        TORCH_NIGHTLY="false"
-        if [[ -f "${NIGHTLY_FILE}" ]]; then
-          TORCH_VERSION=$(grep -v '^#' "${NIGHTLY_FILE}" | grep -v '^$' | head -1 || true)
+        # Check if torch prerelease is configured
+        PRE_FILE="requirements/ci/torch-pre.txt"
+        TORCH_PRERELEASE="false"
+        if [[ -f "${PRE_FILE}" ]]; then
+          # Use while loop for bash 3.2 compatibility (macOS)
+          PRE_CONFIG=()
+          while IFS= read -r line; do
+            PRE_CONFIG+=("$line")
+          done < <(grep -v '^#' "${PRE_FILE}" | grep -v '^$')
+          TORCH_VERSION="${PRE_CONFIG[0]}"
           if [[ -n "${TORCH_VERSION}" ]]; then
-            TORCH_NIGHTLY="true"
+            TORCH_PRERELEASE="true"
           fi
         fi
 
         # Determine install command based on platform and torch configuration
         # - Oldest builds: use --torch-backend=cpu for stable torch
-        # - macOS: use --torch-backend=auto (MPS), nightly not available
-        # - Linux/Windows with nightly: torch already installed, no backend flag needed
+        # - macOS: use --torch-backend=auto (MPS), prerelease not available
+        # - Linux/Windows with prerelease: torch already installed, no backend flag needed
         # - Linux/Windows with stable: use --torch-backend=cpu
         if [[ "${{ inputs.use_oldest }}" == "true" ]]; then
           INSTALL_CMD="uv pip install -e . -r ${REQ_FILE} --torch-backend=cpu"
         elif [[ "${{ runner.os }}" == "macOS" ]]; then
           INSTALL_CMD="uv pip install -e . -r ${REQ_FILE} --torch-backend=auto"
-        elif [[ "${TORCH_NIGHTLY}" == "true" ]]; then
-          # Torch nightly already installed, just install FTS and deps
+        elif [[ "${TORCH_PRERELEASE}" == "true" ]]; then
+          # Torch prerelease already installed, just install FTS and deps
           INSTALL_CMD="uv pip install -e . -r ${REQ_FILE}"
         else
           INSTALL_CMD="uv pip install -e . -r ${REQ_FILE} --torch-backend=cpu"
diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md
index a60a8ba..2129feb 100644
--- a/.github/copilot-instructions.md
+++ b/.github/copilot-instructions.md
@@ -199,6 +199,43 @@ source ${FTS_VENV_BASE}/${FTS_TARGET_VENV}/bin/activate
 PL_RUN_STANDALONE_TESTS=1 python -m pytest tests/test_specific.py::test_function -v
 ```
 
+### Building Documentation
+
+**Documentation build commands (needs activated venv):**
+
+```bash
+export FTS_VENV_BASE=/mnt/cache/${USER}/.venvs
+export FTS_TARGET_VENV=fts_latest
+export FTS_REPO_DIR=${HOME}/repos/finetuning-scheduler  # Example: adjust to your local repo path
+# Activate your environment first
+cd ${FTS_REPO_DIR} && \
+source ${FTS_VENV_BASE}/${FTS_TARGET_VENV}/bin/activate
+
+# Clean previous builds
+cd docs && make clean
+
+# Build HTML documentation with warnings as errors
+make html --debug SPHINXOPTS="-W --keep-going"
+
+# Run linkcheck to verify all links
+make linkcheck SPHINXOPTS="-W --keep-going"
+
+# Check for errors in linkcheck output
+grep -i "error\|broken" build/linkcheck/output.txt || echo "No errors found in linkcheck"
+```
+
+**Documentation requirements:**
+
+- All documentation must build without warnings when using `-W` flag
+- All internal and external links must be valid (verified by linkcheck)
+- RST cross-references should use appropriate directives:
+  - `:class:` for class references
+  - `:meth:` for method references
+  - `:func:` for function references
+  - `:doc:` for document references
+  - `:ref:` for section references (requires explicit label like `.. _label_name:`)
+- Sphinx autosummary generates API documentation from docstrings
+
 ## Project Layout and Architecture
 
 ### Source Code Structure
diff --git a/.gitignore b/.gitignore
index 658f257..02bc844 100644
--- a/.gitignore
+++ b/.gitignore
@@ -32,7 +32,8 @@ timit_data/
 grid_generated*
 grid_ori*
 
-
+# we don't have repo-specific prompts at this juncture
+.github/prompts/
 
 # C extensions
 *.so
diff --git a/README.md b/README.md
index 472d695..696acbf 100644
--- a/README.md
+++ b/README.md
@@ -226,7 +226,7 @@ See the [versioning documentation](https://finetuning-scheduler.readthedocs.io/e
 <details>
   <summary>Current build statuses for Fine-Tuning Scheduler </summary>
 
-| System / (PyTorch/Python ver) |                                                                                                        2.6.0/3.9                                                                                                         |                                                                                                             2.10.0/3.9, 2.10.0/3.12                                                                                                              |
+| System / (PyTorch/Python ver) |                                                                                                        2.6.0/3.10                                                                                                        |                                                                                                             2.10.0/3.10, 2.10.0/3.12                                                                                                             |
 | :---------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
 |      Linux \[GPUs\*\*\]       |                                                                                                            -                                                                                                             | [![Build Status](https://dev.azure.com//speediedan/finetuning-scheduler/_apis/build/status/Multi-GPU%20&%20Example%20Tests?branchName=main)](https://dev.azure.com/speediedan/finetuning-scheduler/_build/latest?definitionId=1&branchName=main) |
 |     Linux (Ubuntu 22.04)      | [![Test](https://github.com/speediedan/finetuning-scheduler/actions/workflows/ci_test-full.yml/badge.svg?branch=main&event=push)](https://github.com/speediedan/finetuning-scheduler/actions/workflows/ci_test-full.yml) |             [![Test](https://github.com/speediedan/finetuning-scheduler/actions/workflows/ci_test-full.yml/badge.svg?branch=main&event=push)](https://github.com/speediedan/finetuning-scheduler/actions/workflows/ci_test-full.yml)             |
diff --git a/dockers/base-cuda/Dockerfile b/dockers/base-cuda/Dockerfile
index caa8e41..3e63426 100644
--- a/dockers/base-cuda/Dockerfile
+++ b/dockers/base-cuda/Dockerfile
@@ -81,13 +81,11 @@ RUN \
     else \
         # or target a specific cuda build, by specifying a particular index url w/...
         # ... default channel
-        # uv pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128; \
-        # ... pytorch patch version
-        # uv pip install torch==1.11.1+cu113 torchvision==0.11.3+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html; \
+        # uv pip install torch --torch-backend=cu128; \
         # ... pytorch nightly dev version
-        uv pip install --prerelease=allow torch==2.10.0.dev20251124 --index-url https://download.pytorch.org/whl/nightly/cu128; \
+        # uv pip install --prerelease=allow torch==2.10.0.dev20251124 --index-url https://download.pytorch.org/whl/nightly/cu128; \
         # ... test channel
-        # uv pip install --prerelease=allow torch==2.10.0 --index-url https://download.pytorch.org/whl/test/cu128; \
+        uv pip install --prerelease=allow torch==2.10.0 --index-url https://download.pytorch.org/whl/test/cu128; \
     fi && \
     # We avoid installing Lightning and other dependencies here as they are usually upgraded anyway later in
     # CI but we may re-enable in the future.
diff --git a/docs/source/distributed/fsdp_scheduled_fine_tuning.rst b/docs/source/distributed/fsdp_scheduled_fine_tuning.rst
index 4c033f2..a9d886b 100644
--- a/docs/source/distributed/fsdp_scheduled_fine_tuning.rst
+++ b/docs/source/distributed/fsdp_scheduled_fine_tuning.rst
@@ -309,10 +309,6 @@ While not technically required, we add ``DebertaV2Embeddings`` separately as wel
 As always, if needed, one can alternatively override ``configure_model`` and manually wrap a given
 :external+pl:class:`~lightning.pytorch.core.module.LightningModule` to align with a desired fine-tuning schedule.
 
-.. warning::
-
-  :class:`~finetuning_scheduler.strategy_adapters.FSDPStrategyAdapter` is in BETA and subject to change. The
-  interface can bring breaking changes and new features with the next release of PyTorch.
 
 .. note::
 
diff --git a/docs/source/index.rst b/docs/source/index.rst
index ca6ff9a..57a3d4f 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -523,6 +523,13 @@ Footnotes
    advanced/lr_scheduler_reinitialization
    advanced/optimizer_reinitialization
 
+.. toctree::
+   :maxdepth: 1
+   :name: Plugins
+   :caption: Plugins
+
+   plugins/strategy_adapter_entry_points
+
 .. toctree::
    :maxdepth: 1
    :name: Basic Examples
diff --git a/docs/source/plugins/strategy_adapter_entry_points.rst b/docs/source/plugins/strategy_adapter_entry_points.rst
new file mode 100644
index 0000000..1554007
--- /dev/null
+++ b/docs/source/plugins/strategy_adapter_entry_points.rst
@@ -0,0 +1,252 @@
+.. _strategy_adapter_entry_points:
+
+Strategy Adapter Entry Points
+##############################
+
+.. warning::
+   This is an :ref:`experimental <versioning:API Stability Classifications>` feature which is
+   still in development. The entry point API and plugin discovery mechanism may change in future
+   releases.
+
+Overview
+********
+
+Fine-Tuning Scheduler (FTS) supports custom strategy adapters via Python entry points, enabling third-party
+packages to extend FTS with specialized adapters for custom training strategies, model architectures,
+or parameter naming conventions (e.g. dynamic parameter views for latent space analysis).
+
+This plugin mechanism allows packages like `Interpretune <https://github.com/speediedan/interpretune>`_
+to provide adapters that integrate seamlessly with FTS without requiring modifications to the FTS
+codebase itself.
+
+**Important Concepts:**
+
+- **Lightning Strategy Flags**: Built-in PyTorch Lightning strategy identifiers (e.g., ``single_device``, ``ddp``, ``fsdp``)
+- **Strategy Adapters**: Classes that extend FTS functionality for specific strategies or use cases
+- **Custom Strategy Adapter Mapping**: User-provided dictionary mapping Lightning strategy flags to adapter implementations
+
+Note that custom strategy adapters are meant to **adapt existing Lightning strategies**, not create
+wholly new ones. If you need a new strategy, register it with PyTorch Lightning first, then create
+an adapter to extend FTS support for it.
+
+Entry Point Specification
+**************************
+
+Entry Point Group
+=================
+
+Custom strategy adapters are registered under the ``finetuning_scheduler.strategy_adapters`` entry
+point group.
+
+Registration Format
+===================
+
+In your package's ``pyproject.toml``, register your adapter:
+
+.. code-block:: toml
+
+    [project.entry-points."finetuning_scheduler.strategy_adapters"]
+    adapter_name = "package.module:AdapterClass"
+
+The entry point name (``adapter_name``) will be lowercased and used to reference the adapter.
+The value should follow Python's standard entry point format: ``module:attribute`` or
+``module.submodule:attribute``.
+
+Discovery and Loading
+*********************
+
+Entry points are discovered lazily during strategy setup (at the start of training). The discovery process:
+
+1. Scans for all registered entry points in the ``finetuning_scheduler.strategy_adapters`` group
+2. Attempts to load each adapter class
+3. Adds successfully loaded adapters to the runtime ``STRATEGY_ADAPTERS`` mapping, keyed by the
+   lowercased entry point name
+4. Logs warnings for any adapters that fail to load (without preventing FTS initialization)
+
+Usage Example
+*************
+
+Real-World Example: TransformerBridge Adapter
+==============================================
+
+The Interpretune package provides a ``TransformerBridgeStrategyAdapter`` that enables clean
+TransformerLens-style parameter naming in FTS schedules. Here's how it's registered and used:
+
+**Registration** (in Interpretune's ``pyproject.toml``):
+
+.. code-block:: toml
+
+    [project.entry-points."finetuning_scheduler.strategy_adapters"]
+    transformerbridge = "interpretune.adapters.transformer_lens:TransformerBridgeStrategyAdapter"
+
+**Usage** (in training configuration):
+
+.. code-block:: python
+
+    from finetuning_scheduler import FinetuningScheduler
+
+    # Map Lightning strategy flags to the adapter
+    # Multiple strategy flags can use the same adapter
+    fts = FinetuningScheduler(
+        custom_strategy_adapters={
+            "single_device": "transformerbridge",  # Use entry point name
+            "ddp": "transformerbridge",  # Same adapter for DDP
+            # Or use fully qualified paths:
+            # "single_device": "interpretune.adapters.transformer_lens:TransformerBridgeStrategyAdapter",
+            # "ddp": "interpretune.adapters.transformer_lens.TransformerBridgeStrategyAdapter",
+        },
+        strategy_adapter_cfg={"use_tl_names": True},  # Adapter-specific config
+    )
+
+This allows fine-tuning schedules to use architecture-agnostic parameter names like ``blocks.9.attn.W_Q`` instead
+of verbose and architecture-dependent canonical names, while FTS handles the necessary translations automatically.
+
+**Key Points:**
+
+- Strategy flags (``single_device``, ``ddp``, etc.) refer to Lightning's built-in strategies
+- The same adapter can be mapped to multiple strategy flags
+- Three formats are supported for referencing adapters:
+
+  1. Entry point name: ``\"transformerbridge\"``
+  2. Colon-separated path: ``\"interpretune.adapters.transformer_lens:TransformerBridgeStrategyAdapter\"``
+  3. Dot-separated path: ``\"interpretune.adapters.transformer_lens.TransformerBridgeStrategyAdapter\"``
+
+Creating Custom Adapters
+*************************
+
+Base Requirements
+=================
+
+Custom strategy adapters should:
+
+1. Inherit from :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`
+2. Implement required methods for your specific use case (see :doc:`/api/finetuning_scheduler.strategy_adapters`)
+3. Follow the adapter lifecycle hooks (``:meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.connect``,
+   ``:meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.on_before_init_fts``, etc.)
+
+Override Points
+===============
+
+Strategy adapters can customize FTS behavior at multiple levels of abstraction accommodating a variety of use cases:
+
+**Parameter Naming**
+  Override :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.get_named_params_for_schedule_validation`
+  to provide custom parameter names while using default validation logic.
+
+**Full Validation**
+  Override :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.validate_ft_sched`
+  to completely customize schedule validation.
+
+**Schedule Generation**
+  Override :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.gen_ft_schedule`
+  to customize how default schedules are generated.
+
+**Checkpoint Handling**
+  Override ``:meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.before_restore_model()``,
+  ``lightning_module_state_dict()``, and ``load_model_state_dict()`` for custom checkpoint translation logic.
+
+See :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter` for the complete API.
+
+Best Practices
+**************
+
+Robust Loading
+==============
+
+Entry point loading is wrapped in exception handling to prevent adapter failures from breaking FTS
+initialization. However, adapters should:
+
+- Validate dependencies and raise clear errors during ``__init__()`` if requirements aren't met
+- Use meaningful exception messages to help users diagnose configuration issues
+- Document any required dependencies in your package documentation
+
+Naming Conventions
+==================
+
+- Use descriptive, lowercase entry point names (e.g., ``transformerbridge``, ``custom_fsdp``)
+- Avoid generic names that might conflict with other packages
+- Consider prefixing with your package name for uniqueness (e.g., ``mypackage_adapter``)
+
+Configuration
+=============
+
+Custom Adapter Mapping Format
+------------------------------
+
+The :paramref:`~finetuning_scheduler.fts.FinetuningScheduler.custom_strategy_adapters` parameter
+accepts a dictionary mapping PyTorch Lightning strategy flags (canonical strategy names like
+``"single_device"``, ``"auto"``, ``"ddp"``, etc.) to adapter references. The adapter reference
+can be:
+
+1. **An entry point name** (lowercased) registered under ``finetuning_scheduler.strategy_adapters``
+2. **A fully qualified class path** in the format ``"module.path:ClassName"`` or
+   ``"module.path.ClassName"``
+
+This allows multiple strategy flags to be associated with the same adapter. For example:
+
+.. code-block:: python
+
+    from finetuning_scheduler import FinetuningScheduler
+
+    # Multiple strategies can use the same registered plugin adapter
+    fts = FinetuningScheduler(
+        custom_strategy_adapters={
+            "single_device": "transformerbridge",  # Plugin entry point name
+            # We can use the same plugin for multiple strategies, here we use a fully qualified path format as well
+            "auto": "interpretune.adapters.transformer_lens.TransformerBridgeStrategyAdapter",
+        },
+        strategy_adapter_cfg={
+            "use_tl_names": True,  # Configuration passed to the adapter
+        },
+    )
+
+**Native FTS Adapters**: FTS includes built-in adapters in the ``STRATEGY_ADAPTERS`` mapping
+that are always available:
+
+- ``"fsdp"`` - :class:`~finetuning_scheduler.strategy_adapters.FSDPStrategyAdapter`
+- ``"modelparallelstrategy"`` - :class:`~finetuning_scheduler.strategy_adapters.ModelParallelStrategyAdapter`
+
+These can be referenced directly without requiring plugin registration.
+
+Adapter-Specific Configuration
+-------------------------------
+
+If your adapter accepts configuration, use the ``strategy_adapter_cfg`` parameter:
+
+.. code-block:: python
+
+    fts = FinetuningScheduler(
+        custom_strategy_adapters={"target_strategy": "my_adapter"},
+        strategy_adapter_cfg={
+            "option1": value1,
+            "option2": value2,
+        },
+    )
+
+Testing
+=======
+
+Test your adapter with FTS by:
+
+1. Creating test fixtures that instantiate FTS with your adapter
+2. Verifying schedule validation works with your parameter naming
+3. Testing checkpoint save/restore if you override those methods
+4. Ensuring your adapter works with both explicit and implicit schedules
+
+Future Directions
+*****************
+
+This plugin system may be extended in future releases to support:
+
+- Versioned adapter APIs
+- Additional extension points beyond strategy adapters
+
+Community adapters and feedback on the plugin system are welcome! Please share your use cases
+and suggestions on the `GitHub repository <https://github.com/speediedan/finetuning-scheduler/issues>`_.
+
+See Also
+********
+
+- :doc:`/api/finetuning_scheduler.strategy_adapters`
+- :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`
+- :ref:`versioning:API Stability Classifications`
diff --git a/pyproject.toml b/pyproject.toml
index d8282e7..15f5b93 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -13,6 +13,12 @@ authors = [{name = "Daniel Dale", email = "danny.dale@gmail.com"}]
 license = "Apache-2.0"
 license-files = ["LICENSE*"]
 requires-python = ">=3.10"
+
+# NOTE: Minimum versions for torch and Lightning are managed dynamically by setup.py
+# See src/finetuning_scheduler/dynamic_versioning/utils.py for version specifications:
+#   - torch: defined in BASE_DEPENDENCIES
+#   - Lightning: defined in LIGHTNING_PACKAGES (unified/standalone)
+# The 'dependencies' field above is marked as 'dynamic' and populated at build time.
 keywords = [
     "deep learning",
     "pytorch",
@@ -118,6 +124,17 @@ all = [
     "traitlets>=5.0.0",
 ]
 
+# -----------------------------------------------------------------------------
+# FTS Metadata - Minimum Version Requirements (informational only)
+# -----------------------------------------------------------------------------
+# These values are for documentation purposes and are NOT used during installation.
+# Actual minimum versions are enforced in src/finetuning_scheduler/dynamic_versioning/utils.py
+# The build system (setup.py) reads from utils.py to populate the dynamic dependencies.
+[tool.fts.min-versions]
+torch = ">=2.6.0"
+lightning = ">=2.6.0,<2.6.1"
+python = "3.10"  # Defined in pyproject.toml requires-python
+
 [project.urls]
 "Homepage" = "https://github.com/speediedan/finetuning-scheduler"
 "Bug Tracker" = "https://github.com/speediedan/finetuning-scheduler/issues"
diff --git a/requirements/ci/requirements-oldest.txt b/requirements/ci/requirements-oldest.txt
index 20e13f0..1e8d69e 100644
--- a/requirements/ci/requirements-oldest.txt
+++ b/requirements/ci/requirements-oldest.txt
@@ -1,12 +1,12 @@
 # This file was autogenerated by uv via the following command:
-#    uv pip compile /home/speediedan/repos/finetuning-scheduler/pyproject.toml --extra all --group dev --group test --output-file /home/speediedan/repos/finetuning-scheduler/requirements/ci/requirements-oldest.txt --no-strip-extras --resolution lowest-direct --universal --python-version 3.10
+#    uv pip compile pyproject.toml --extra all --group dev --group test --output-file /home/speediedan/repos/finetuning-scheduler/requirements/ci/requirements-oldest.txt --no-strip-extras --resolution lowest-direct --universal --python-version 3.10
 aiohappyeyeballs==2.6.1
     # via aiohttp
-aiohttp==3.13.2
+aiohttp==3.13.3
     # via fsspec
 aiosignal==1.4.0
     # via aiohttp
-alembic==1.17.2
+alembic==1.18.0
     # via mlflow
 annotated-doc==0.0.4
     # via fastapi
@@ -16,7 +16,7 @@ antlr4-python3-runtime==4.9.3
     # via
     #   hydra-core
     #   omegaconf
-anyio==4.12.0
+anyio==4.12.1
     # via starlette
 appnope==0.1.4 ; sys_platform == 'darwin'
     # via
@@ -49,10 +49,8 @@ bleach==6.3.0
 blinker==1.9.0
     # via flask
 cachetools==5.5.2
-    # via
-    #   google-auth
-    #   mlflow-skinny
-certifi==2025.11.12
+    # via mlflow-skinny
+certifi==2026.1.4
     # via requests
 cffi==2.0.0
     # via
@@ -92,13 +90,13 @@ cryptography==46.0.3 ; sys_platform == 'linux'
     # via secretstorage
 cycler==0.12.1
     # via matplotlib
-databricks-sdk==0.73.0
+databricks-sdk==0.77.0
     # via mlflow-skinny
 datasets==4.0.0
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   evaluate
-debugpy==1.8.17
+debugpy==1.8.19
     # via ipykernel
 decorator==5.2.1
     # via ipython
@@ -115,7 +113,7 @@ docker==7.1.0
     # via mlflow
 docstring-parser==0.17.0
     # via jsonargparse
-docutils==0.22.3
+docutils==0.22.4
     # via readme-renderer
 entrypoints==0.4
     # via
@@ -125,9 +123,9 @@ evaluate==0.3.0
     # via finetuning-scheduler (pyproject.toml)
 exceptiongroup==1.3.1 ; python_full_version < '3.11'
     # via anyio
-fastapi==0.124.0
+fastapi==0.128.0
     # via mlflow-skinny
-filelock==3.20.0
+filelock==3.20.3
     # via
     #   datasets
     #   huggingface-hub
@@ -136,7 +134,7 @@ filelock==3.20.0
     #   virtualenv
 flask==3.1.2
     # via mlflow
-fonttools==4.61.0
+fonttools==4.61.1
     # via matplotlib
 frozenlist==1.8.0
     # via
@@ -152,9 +150,9 @@ fsspec[http]==2025.3.0
     #   torch
 gitdb==4.0.12
     # via gitpython
-gitpython==3.1.45
+gitpython==3.1.46
     # via mlflow-skinny
-google-auth==2.43.0
+google-auth==2.47.0
     # via databricks-sdk
 graphene==3.4.3
     # via mlflow
@@ -186,7 +184,7 @@ idna==3.11
     #   anyio
     #   requests
     #   yarl
-importlib-metadata==8.7.0
+importlib-metadata==8.7.1
     # via
     #   keyring
     #   mlflow-skinny
@@ -217,9 +215,9 @@ itsdangerous==2.2.0
     # via flask
 jaraco-classes==3.4.0
     # via keyring
-jaraco-context==6.0.1
+jaraco-context==6.0.2
     # via keyring
-jaraco-functools==4.3.0
+jaraco-functools==4.4.0
     # via keyring
 jedi==0.19.2
     # via ipython
@@ -233,7 +231,7 @@ jinja2==3.1.6
     #   nbconvert
     #   notebook
     #   torch
-joblib==1.5.2
+joblib==1.5.3
     # via scikit-learn
 jsonargparse[signatures]==4.27.7
     # via finetuning-scheduler (pyproject.toml)
@@ -282,7 +280,7 @@ markupsafe==3.0.3
     #   jinja2
     #   mako
     #   werkzeug
-matplotlib==3.10.7
+matplotlib==3.10.8
     # via mlflow
 matplotlib-inline==0.1.7
     # via
@@ -337,11 +335,11 @@ nest-asyncio==1.6.0
     #   nbclient
 networkx==3.4.2 ; python_full_version < '3.11'
     # via torch
-networkx==3.6 ; python_full_version >= '3.11'
+networkx==3.6.1 ; python_full_version >= '3.11'
     # via torch
 nh3==0.3.2
     # via readme-renderer
-nodeenv==1.9.1
+nodeenv==1.10.0
     # via
     #   pre-commit
     #   pyright
@@ -362,7 +360,7 @@ numpy==2.2.6 ; python_full_version < '3.11'
     #   tensorboardx
     #   torchmetrics
     #   transformers
-numpy==2.3.5 ; python_full_version >= '3.11'
+numpy==2.4.0 ; python_full_version >= '3.11'
     # via
     #   contourpy
     #   datasets
@@ -414,14 +412,14 @@ omegaconf==2.2.3
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   hydra-core
-opentelemetry-api==1.39.0
+opentelemetry-api==1.39.1
     # via
     #   mlflow-skinny
     #   opentelemetry-sdk
     #   opentelemetry-semantic-conventions
-opentelemetry-sdk==1.39.0
+opentelemetry-sdk==1.39.1
     # via mlflow-skinny
-opentelemetry-semantic-conventions==0.60b0
+opentelemetry-semantic-conventions==0.60b1
     # via opentelemetry-sdk
 packaging==25.0
     # via
@@ -452,7 +450,7 @@ pexpect==4.9.0 ; sys_platform != 'win32'
     # via ipython
 pickleshare==0.7.5
     # via ipython
-pillow==12.0.0
+pillow==12.1.0
     # via matplotlib
 pip==21.0
     # via finetuning-scheduler (pyproject.toml)
@@ -511,7 +509,7 @@ pygments==2.19.2
     #   nbconvert
     #   readme-renderer
     #   rich
-pyparsing==3.2.5
+pyparsing==3.3.1
     # via matplotlib
 pyright==1.1.390
     # via
@@ -604,7 +602,7 @@ scipy==1.16.3 ; python_full_version >= '3.11'
     #   scikit-learn
 secretstorage==3.5.0 ; sys_platform == 'linux'
     # via keyring
-send2trash==1.8.3
+send2trash==2.0.0
     # via notebook
 sentencepiece==0.2.0
     # via finetuning-scheduler (pyproject.toml)
@@ -621,11 +619,11 @@ six==1.17.0
     #   python-dateutil
 smmap==5.0.2
     # via gitdb
-sqlalchemy==2.0.44
+sqlalchemy==2.0.45
     # via
     #   alembic
     #   mlflow
-sqlparse==0.5.4
+sqlparse==0.5.5
     # via mlflow-skinny
 starlette==0.50.0
     # via fastapi
@@ -661,7 +659,7 @@ torchmetrics==1.8.2
     # via
     #   lightning
     #   pytorch-lightning
-tornado==6.5.2
+tornado==6.5.4
     # via
     #   ipykernel
     #   jupyter-client
@@ -729,16 +727,16 @@ typing-extensions==4.15.0
     #   virtualenv
 typing-inspection==0.4.2
     # via pydantic
-tzdata==2025.2
+tzdata==2025.3
     # via pandas
-urllib3==2.6.0
+urllib3==2.6.3
     # via
     #   docker
     #   requests
     #   responses
-uvicorn==0.38.0
+uvicorn==0.40.0
     # via mlflow-skinny
-virtualenv==20.35.4
+virtualenv==20.36.1
     # via pre-commit
 waitress==3.0.2 ; sys_platform == 'win32'
     # via mlflow
@@ -746,7 +744,7 @@ wcwidth==0.2.14
     # via prompt-toolkit
 webencodings==0.5.1
     # via bleach
-werkzeug==3.1.4
+werkzeug==3.1.5
     # via flask
 widgetsnbextension==3.5.2
     # via ipywidgets
diff --git a/requirements/ci/requirements.txt b/requirements/ci/requirements.txt
index 6a0e957..24c21b4 100644
--- a/requirements/ci/requirements.txt
+++ b/requirements/ci/requirements.txt
@@ -1,12 +1,12 @@
 # This file was autogenerated by uv via the following command:
-#    uv pip compile /home/speediedan/repos/finetuning-scheduler/pyproject.toml --extra all --group dev --group test --output-file /home/speediedan/repos/finetuning-scheduler/requirements/ci/requirements.txt --no-strip-extras --resolution highest --universal --python-version 3.10 --prerelease=if-necessary-or-explicit --override /tmp/tmp.CNVCWjH0nH --index-strategy unsafe-best-match --no-emit-package torch
+#    uv pip compile pyproject.toml --extra all --group dev --group test --output-file /home/speediedan/repos/finetuning-scheduler/requirements/ci/requirements.txt --no-strip-extras --resolution highest --universal --python-version 3.10 --prerelease=if-necessary-or-explicit --override /tmp/tmp.qrV2Vucuzc --index-strategy unsafe-best-match --no-emit-package torch
 aiohappyeyeballs==2.6.1
     # via aiohttp
-aiohttp==3.13.2
+aiohttp==3.13.3
     # via fsspec
 aiosignal==1.4.0
     # via aiohttp
-alembic==1.17.2
+alembic==1.18.0
     # via mlflow
 annotated-doc==0.0.4
     # via fastapi
@@ -16,7 +16,7 @@ antlr4-python3-runtime==4.9.3
     # via
     #   hydra-core
     #   omegaconf
-anyio==4.12.0
+anyio==4.12.1
     # via
     #   httpx
     #   jupyter-server
@@ -50,12 +50,11 @@ bleach[css]==6.3.0
     # via nbconvert
 blinker==1.9.0
     # via flask
-cachetools==6.2.2
+cachetools==6.2.4
     # via
-    #   google-auth
     #   mlflow-skinny
     #   mlflow-tracing
-certifi==2025.11.12
+certifi==2026.1.4
     # via
     #   httpcore
     #   httpx
@@ -90,7 +89,7 @@ contourpy==1.3.2 ; python_full_version < '3.11'
     # via matplotlib
 contourpy==1.3.3 ; python_full_version >= '3.11'
     # via matplotlib
-coverage==7.12.0
+coverage==7.13.1
     # via
     #   finetuning-scheduler (pyproject.toml:dev)
     #   finetuning-scheduler (pyproject.toml:test)
@@ -101,15 +100,15 @@ cryptography==46.0.3
     #   secretstorage
 cycler==0.12.1
     # via matplotlib
-databricks-sdk==0.73.0
+databricks-sdk==0.77.0
     # via
     #   mlflow-skinny
     #   mlflow-tracing
-datasets==4.4.1
+datasets==4.4.2
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   evaluate
-debugpy==1.8.17
+debugpy==1.8.19
     # via ipykernel
 decorator==5.2.1
     # via ipython
@@ -126,7 +125,7 @@ docker==7.1.0
     # via mlflow
 docstring-parser==0.17.0
     # via jsonargparse
-docutils==0.22.3
+docutils==0.22.4
     # via readme-renderer
 evaluate==0.4.6
     # via finetuning-scheduler (pyproject.toml)
@@ -137,11 +136,11 @@ exceptiongroup==1.3.1 ; python_full_version < '3.11'
     #   pytest
 executing==2.2.1
     # via stack-data
-fastapi==0.124.0
+fastapi==0.128.0
     # via mlflow-skinny
 fastjsonschema==2.21.2
     # via nbformat
-filelock==3.20.0
+filelock==3.20.3
     # via
     #   datasets
     #   huggingface-hub
@@ -152,9 +151,9 @@ flask==3.1.2
     # via
     #   flask-cors
     #   mlflow
-flask-cors==6.0.1
+flask-cors==6.0.2
     # via mlflow
-fonttools==4.61.0
+fonttools==4.61.1
     # via matplotlib
 fqdn==1.5.1
     # via jsonschema
@@ -172,9 +171,9 @@ fsspec[http]==2025.10.0
     #   torch
 gitdb==4.0.12
     # via gitpython
-gitpython==3.1.45
+gitpython==3.1.46
     # via mlflow-skinny
-google-auth==2.43.0
+google-auth==2.47.0
     # via databricks-sdk
 graphene==3.4.3
     # via mlflow
@@ -200,7 +199,7 @@ httpx==0.28.1
     # via
     #   datasets
     #   jupyterlab
-huey==2.5.5
+huey==2.6.0
     # via mlflow
 huggingface-hub==0.36.0
     # via
@@ -221,7 +220,7 @@ idna==3.11
     #   jsonschema
     #   requests
     #   yarl
-importlib-metadata==8.7.0
+importlib-metadata==8.7.1
     # via
     #   keyring
     #   mlflow-skinny
@@ -235,12 +234,12 @@ ipykernel==7.1.0
     #   finetuning-scheduler (pyproject.toml)
     #   jupyterlab
     #   nbval
-ipython==8.37.0 ; python_full_version < '3.11'
+ipython==8.38.0 ; python_full_version < '3.11'
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   ipykernel
     #   ipywidgets
-ipython==9.8.0 ; python_full_version >= '3.11'
+ipython==9.9.0 ; python_full_version >= '3.11'
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   ipykernel
@@ -255,9 +254,9 @@ itsdangerous==2.2.0
     # via flask
 jaraco-classes==3.4.0 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
     # via keyring
-jaraco-context==6.0.1 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
+jaraco-context==6.0.2 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
     # via keyring
-jaraco-functools==4.3.0 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
+jaraco-functools==4.4.0 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
     # via keyring
 jedi==0.19.2
     # via ipython
@@ -273,22 +272,22 @@ jinja2==3.1.6
     #   jupyterlab-server
     #   nbconvert
     #   torch
-joblib==1.5.2
+joblib==1.5.3
     # via scikit-learn
-json5==0.12.1
+json5==0.13.0
     # via jupyterlab-server
 jsonargparse[signatures]==4.41.0
     # via finetuning-scheduler (pyproject.toml)
 jsonpointer==3.0.0
     # via jsonschema
-jsonschema[format-nongpl]==4.25.1
+jsonschema[format-nongpl]==4.26.0
     # via
     #   jupyter-events
     #   jupyterlab-server
     #   nbformat
 jsonschema-specifications==2025.9.1
     # via jsonschema
-jupyter-client==8.6.3
+jupyter-client==8.8.0
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   ipykernel
@@ -318,7 +317,7 @@ jupyter-server==2.17.0
     #   notebook-shim
 jupyter-server-terminals==0.5.3
     # via jupyter-server
-jupyterlab==4.5.0
+jupyterlab==4.5.1
     # via notebook
 jupyterlab-pygments==0.3.0
     # via nbconvert
@@ -357,7 +356,7 @@ markupsafe==3.0.3
     #   mako
     #   nbconvert
     #   werkzeug
-matplotlib==3.10.7
+matplotlib==3.10.8
     # via mlflow
 matplotlib-inline==0.2.1
     # via
@@ -367,15 +366,15 @@ mdit-py-plugins==0.5.0
     # via jupytext
 mdurl==0.1.2
     # via markdown-it-py
-mistune==3.1.4
+mistune==3.2.0
     # via nbconvert
-mlflow==3.7.0
+mlflow==3.8.1
     # via
     #   finetuning-scheduler (pyproject.toml:dev)
     #   finetuning-scheduler (pyproject.toml:test)
-mlflow-skinny==3.7.0
+mlflow-skinny==3.8.1
     # via mlflow
-mlflow-tracing==3.7.0
+mlflow-tracing==3.8.1
     # via mlflow
 more-itertools==10.8.0 ; platform_machine != 'ppc64le' and platform_machine != 's390x'
     # via
@@ -389,7 +388,7 @@ multiprocess==0.70.18
     # via
     #   datasets
     #   evaluate
-nbclient==0.10.2
+nbclient==0.10.4
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   nbconvert
@@ -411,11 +410,11 @@ nest-asyncio==1.6.0
     # via ipykernel
 nh3==0.3.2
     # via readme-renderer
-nodeenv==1.9.1
+nodeenv==1.10.0
     # via
     #   pre-commit
     #   pyright
-notebook==7.5.0
+notebook==7.5.1
     # via finetuning-scheduler (pyproject.toml)
 notebook-shim==0.2.4
     # via
@@ -434,7 +433,7 @@ numpy==2.2.6 ; python_full_version < '3.11'
     #   tensorboardx
     #   torchmetrics
     #   transformers
-numpy==2.3.5 ; python_full_version >= '3.11'
+numpy==2.4.0 ; python_full_version >= '3.11'
     # via
     #   contourpy
     #   datasets
@@ -451,21 +450,21 @@ omegaconf==2.3.0
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   hydra-core
-opentelemetry-api==1.39.0
+opentelemetry-api==1.39.1
     # via
     #   mlflow-skinny
     #   mlflow-tracing
     #   opentelemetry-sdk
     #   opentelemetry-semantic-conventions
-opentelemetry-proto==1.39.0
+opentelemetry-proto==1.39.1
     # via
     #   mlflow-skinny
     #   mlflow-tracing
-opentelemetry-sdk==1.39.0
+opentelemetry-sdk==1.39.1
     # via
     #   mlflow-skinny
     #   mlflow-tracing
-opentelemetry-semantic-conventions==0.60b0
+opentelemetry-semantic-conventions==0.60b1
     # via opentelemetry-sdk
 overrides==7.7.0 ; python_full_version < '3.12'
     # via jupyter-server
@@ -506,7 +505,7 @@ parso==0.8.5
     # via jedi
 pexpect==4.9.0 ; sys_platform != 'emscripten' and sys_platform != 'win32'
     # via ipython
-pillow==12.0.0
+pillow==12.1.0
     # via matplotlib
 pip==25.3
     # via finetuning-scheduler (pyproject.toml)
@@ -516,7 +515,7 @@ platformdirs==4.5.1
     #   virtualenv
 pluggy==1.6.0
     # via pytest
-pre-commit==4.5.0
+pre-commit==4.5.1
     # via
     #   finetuning-scheduler (pyproject.toml:dev)
     #   finetuning-scheduler (pyproject.toml:test)
@@ -535,7 +534,7 @@ protobuf==6.33.2
     #   mlflow-tracing
     #   opentelemetry-proto
     #   tensorboardx
-psutil==7.1.3
+psutil==7.2.1
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   ipykernel
@@ -572,13 +571,13 @@ pygments==2.19.2
     #   pytest
     #   readme-renderer
     #   rich
-pyparsing==3.2.5
+pyparsing==3.3.1
     # via matplotlib
-pyright==1.1.407
+pyright==1.1.408
     # via
     #   finetuning-scheduler (pyproject.toml:dev)
     #   finetuning-scheduler (pyproject.toml:test)
-pytest==9.0.1
+pytest==9.0.2
     # via
     #   finetuning-scheduler (pyproject.toml:dev)
     #   finetuning-scheduler (pyproject.toml:test)
@@ -678,7 +677,11 @@ rsa==4.9.1
     # via google-auth
 safetensors==0.7.0
     # via transformers
-scikit-learn==1.7.2
+scikit-learn==1.7.2 ; python_full_version < '3.11'
+    # via
+    #   finetuning-scheduler (pyproject.toml)
+    #   mlflow
+scikit-learn==1.8.0 ; python_full_version >= '3.11'
     # via
     #   finetuning-scheduler (pyproject.toml)
     #   mlflow
@@ -692,7 +695,7 @@ scipy==1.16.3 ; python_full_version >= '3.11'
     #   scikit-learn
 secretstorage==3.5.0 ; platform_machine != 'ppc64le' and platform_machine != 's390x' and sys_platform == 'linux'
     # via keyring
-send2trash==1.8.3
+send2trash==2.0.0
     # via jupyter-server
 sentencepiece==0.2.1
     # via finetuning-scheduler (pyproject.toml)
@@ -707,13 +710,13 @@ six==1.17.0
     #   rfc3339-validator
 smmap==5.0.2
     # via gitdb
-soupsieve==2.8
+soupsieve==2.8.1
     # via beautifulsoup4
-sqlalchemy==2.0.44
+sqlalchemy==2.0.45
     # via
     #   alembic
     #   mlflow
-sqlparse==0.5.4
+sqlparse==0.5.5
     # via mlflow-skinny
 stack-data==0.6.3
     # via ipython
@@ -731,7 +734,7 @@ threadpoolctl==3.6.0
     # via scikit-learn
 tinycss2==1.4.0
     # via bleach
-tokenizers==0.22.1
+tokenizers==0.22.2
     # via transformers
 tomli==2.3.0 ; python_full_version < '3.11'
     # via
@@ -743,7 +746,7 @@ torchmetrics==1.8.2
     # via
     #   lightning
     #   pytorch-lightning
-tornado==6.5.2
+tornado==6.5.4
     # via
     #   ipykernel
     #   jupyter-client
@@ -817,20 +820,20 @@ typing-extensions==4.15.0
     #   virtualenv
 typing-inspection==0.4.2
     # via pydantic
-tzdata==2025.2
+tzdata==2025.3
     # via
     #   arrow
     #   pandas
 uri-template==1.3.0
     # via jsonschema
-urllib3==2.6.0
+urllib3==2.6.3
     # via
     #   docker
     #   requests
     #   twine
-uvicorn==0.38.0
+uvicorn==0.40.0
     # via mlflow-skinny
-virtualenv==20.35.4
+virtualenv==20.36.1
     # via pre-commit
 waitress==3.0.2 ; sys_platform == 'win32'
     # via mlflow
@@ -844,7 +847,7 @@ webencodings==0.5.1
     #   tinycss2
 websocket-client==1.9.0
     # via jupyter-server
-werkzeug==3.1.4
+werkzeug==3.1.5
     # via
     #   flask
     #   flask-cors
diff --git a/requirements/ci/torch-nightly.txt b/requirements/ci/torch-nightly.txt
deleted file mode 100644
index 1030b4a..0000000
--- a/requirements/ci/torch-nightly.txt
+++ /dev/null
@@ -1,11 +0,0 @@
-# PyTorch nightly version configuration
-# To enable nightly builds, uncomment and set the version and CUDA target below.
-# Leave commented/empty to use stable PyTorch releases.
-#
-# Format:
-#   Line 1: torch version (e.g., 2.10.0.dev20251203)
-#   Line 2: CUDA target for local builds (e.g., cu128) - CI always uses cpu
-#
-# Example (uncomment both lines to enable):
-2.10.0.dev20251124
-cu128
diff --git a/requirements/ci/torch-pre.txt b/requirements/ci/torch-pre.txt
new file mode 100644
index 0000000..3418e47
--- /dev/null
+++ b/requirements/ci/torch-pre.txt
@@ -0,0 +1,18 @@
+# PyTorch prerelease version configuration
+# To enable prerelease builds, uncomment and set the values below.
+# Leave commented/empty to use stable PyTorch releases.
+#
+# Format:
+#   Line 1: torch version (e.g., 2.10.0 for test/RC, 2.10.0.dev20251203 for nightly)
+#   Line 2: CUDA target for local builds (e.g., cu128) - CI always uses cpu
+#   Line 3: channel type: "test" or "nightly"
+#
+# Example for test/RC channel (currently enabled):
+2.10.0
+cu128
+test
+
+# Example for nightly (commented out):
+# 2.10.0.dev20251124
+# cu128
+# nightly
diff --git a/requirements/ci/torch_override.txt b/requirements/ci/torch_override.txt
index 3a916b2..b98382d 100644
--- a/requirements/ci/torch_override.txt
+++ b/requirements/ci/torch_override.txt
@@ -1,16 +1,16 @@
-# PyTorch nightly override for UV_OVERRIDE
-# Generated by lock_ci_requirements.sh from torch-nightly.txt
+# PyTorch prerelease override for UV_OVERRIDE
+# Generated by lock_ci_requirements.sh from torch-pre.txt
 #
-# Manual installation with nightly (two-step approach for security):
-#   Step 1: Install PyTorch nightly
-#   uv pip install --prerelease=if-necessary-or-explicit torch==2.10.0.dev20251124 --index-url https://download.pytorch.org/whl/nightly/cu128
+# Manual installation with prerelease (two-step approach for security):
+#   Step 1: Install PyTorch prerelease (edit torch-pre.txt to configure channel)
+#   uv pip install --prerelease=if-necessary-or-explicit torch==2.10.0 --index-url https://download.pytorch.org/whl/nightly/cu128
 #
 #   Step 2: Install FTS with Lightning commit pin (torch already installed, will be skipped)
 #   export UV_OVERRIDE=${PWD}/requirements/ci/overrides.txt
 #   uv pip install -e ".[all]"
 #
 # Or with locked requirements:
-#   uv pip install --prerelease=if-necessary-or-explicit torch==2.10.0.dev20251124 --index-url https://download.pytorch.org/whl/nightly/cu128
+#   uv pip install --prerelease=if-necessary-or-explicit torch==2.10.0 --index-url https://download.pytorch.org/whl/nightly/cu128
 #   UV_OVERRIDE=${PWD}/requirements/ci/overrides.txt uv pip install -e . -r requirements/ci/requirements.txt
 
-torch==2.10.0.dev20251124
+torch==2.10.0
diff --git a/requirements/utils/lock_ci_requirements.sh b/requirements/utils/lock_ci_requirements.sh
index e6ad2fa..4df0ac1 100755
--- a/requirements/utils/lock_ci_requirements.sh
+++ b/requirements/utils/lock_ci_requirements.sh
@@ -10,12 +10,12 @@
 # which allows uv to properly resolve oldest compatible versions.
 #
 # Torch handling:
-# - When torch-nightly.txt is configured:
-#   - Lock file is generated with torch pinned to the nightly version
-#   - Uses PyTorch nightly index for resolution
-#   - Docker image and CI both use the same nightly version
+# - When torch-pre.txt is configured:
+#   - Lock file is generated with torch pinned to the prerelease version
+#   - Uses PyTorch nightly or test index for resolution (based on channel type)
+#   - Docker image and CI both use the same prerelease version
 #   - Post-processing prunes torch-only dependencies (see prune_torch_only_deps)
-# - Without torch-nightly.txt:
+# - Without torch-pre.txt:
 #   - Uses stable torch from PyPI
 #   - CI uses --torch-backend=cpu for CPU variant
 set -eo pipefail
@@ -27,7 +27,7 @@ unset UV_OVERRIDE
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
 CI_DIR="${REPO_ROOT}/requirements/ci"
-TORCH_NIGHTLY_FILE="${CI_DIR}/torch-nightly.txt"
+TORCH_PRE_FILE="${CI_DIR}/torch-pre.txt"
 TORCH_OVERRIDE_FILE="${CI_DIR}/torch_override.txt"
 
 # Ensure output directory exists
@@ -35,14 +35,17 @@ mkdir -p "${CI_DIR}"
 
 echo "Generating locked CI requirements from pyproject.toml..."
 
-# Check if torch nightly is configured
-# Returns: torch_version if nightly is enabled, empty string otherwise
-get_torch_nightly_version() {
-    if [[ -f "${TORCH_NIGHTLY_FILE}" ]]; then
-        # Read first non-comment, non-empty line as torch version
-        local version=$(grep -v '^#' "${TORCH_NIGHTLY_FILE}" | grep -v '^$' | head -1)
-        if [[ -n "${version}" ]]; then
-            echo "${version}"
+# Check if torch prerelease is configured
+# Returns: "version:channel" if prerelease is enabled, empty string otherwise
+get_torch_pre_config() {
+    if [[ -f "${TORCH_PRE_FILE}" ]]; then
+        # Read 3-line config: version, CUDA target, channel type
+        readarray -t PRE_CONFIG < <(grep -v '^#' "${TORCH_PRE_FILE}" | grep -v '^$')
+        local version="${PRE_CONFIG[0]}"
+        local cuda="${PRE_CONFIG[1]}"
+        local channel="${PRE_CONFIG[2]}"
+        if [[ -n "${version}" && -n "${channel}" ]]; then
+            echo "${version}:${channel}"
             return
         fi
     fi
@@ -65,8 +68,15 @@ get_torch_nightly_version() {
 # which ensures uv can properly resolve oldest compatible versions with
 # --resolution=lowest-direct without needing external constraint files.
 
-# Check if torch nightly is configured
-TORCH_NIGHTLY_VERSION=$(get_torch_nightly_version)
+# Check if torch prerelease is configured
+TORCH_PRE_CONFIG=$(get_torch_pre_config)
+if [[ -n "${TORCH_PRE_CONFIG}" ]]; then
+    TORCH_PRE_VERSION="${TORCH_PRE_CONFIG%%:*}"  # Extract version before ':'
+    TORCH_PRE_CHANNEL="${TORCH_PRE_CONFIG##*:}"  # Extract channel after ':'
+else
+    TORCH_PRE_VERSION=""
+    TORCH_PRE_CHANNEL=""
+fi
 
 # Prune packages that are ONLY dependencies of torch from the lockfile.
 # This reduces the dependency confusion attack surface when using unsafe-best-match
@@ -77,33 +87,33 @@ prune_torch_only_deps() {
     python "${SCRIPT_DIR}/prune_torch_deps.py" "${lockfile}"
 }
 
-# Generate/update torch_override.txt if nightly is configured, remove if not
+# Generate/update torch_override.txt if prerelease is configured, remove if not
 generate_torch_override() {
-    if [[ -n "${TORCH_NIGHTLY_VERSION}" ]]; then
+    if [[ -n "${TORCH_PRE_VERSION}" ]]; then
         cat > "${TORCH_OVERRIDE_FILE}" << EOF
-# PyTorch nightly override for UV_OVERRIDE
-# Generated by lock_ci_requirements.sh from torch-nightly.txt
+# PyTorch prerelease override for UV_OVERRIDE
+# Generated by lock_ci_requirements.sh from torch-pre.txt
 #
-# Manual installation with nightly (two-step approach for security):
-#   Step 1: Install PyTorch nightly
-#   uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_NIGHTLY_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128
+# Manual installation with prerelease (two-step approach for security):
+#   Step 1: Install PyTorch prerelease (edit torch-pre.txt to configure channel)
+#   uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_PRE_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128
 #
 #   Step 2: Install FTS with Lightning commit pin (torch already installed, will be skipped)
 #   export UV_OVERRIDE=\${PWD}/requirements/ci/overrides.txt
 #   uv pip install -e ".[all]"
 #
 # Or with locked requirements:
-#   uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_NIGHTLY_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128
+#   uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_PRE_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128
 #   UV_OVERRIDE=\${PWD}/requirements/ci/overrides.txt uv pip install -e . -r requirements/ci/requirements.txt
 
-torch==${TORCH_NIGHTLY_VERSION}
+torch==${TORCH_PRE_VERSION}
 EOF
-        echo "✓ Generated ${TORCH_OVERRIDE_FILE} (torch==${TORCH_NIGHTLY_VERSION})"
+        echo "✓ Generated ${TORCH_OVERRIDE_FILE} (torch==${TORCH_PRE_VERSION})"
     else
-        # Remove stale override file if nightly is disabled
+        # Remove stale override file if prerelease is disabled
         if [[ -f "${TORCH_OVERRIDE_FILE}" ]]; then
             rm -f "${TORCH_OVERRIDE_FILE}"
-            echo "✓ Removed ${TORCH_OVERRIDE_FILE} (nightly disabled)"
+            echo "✓ Removed ${TORCH_OVERRIDE_FILE} (prerelease disabled)"
         fi
     fi
 }
@@ -112,14 +122,17 @@ generate_lockfile() {
     local resolution=$1
     local output_file=$2
     local python_version=$3
-    local use_nightly=$4  # "true" to use torch nightly
+    local use_prerelease=$4  # "true" to use torch prerelease
 
     echo "Generating ${output_file} with resolution=${resolution}, python=${python_version}..."
 
+    # Change to repo root for dependency group resolution
+    pushd "${REPO_ROOT}" > /dev/null
+
     # Build the base compile command
     local compile_cmd=(
         uv pip compile
-        "${REPO_ROOT}/pyproject.toml"
+        pyproject.toml
         --extra all
         --group dev
         --group test
@@ -131,42 +144,45 @@ generate_lockfile() {
         --python-version "${python_version}"
     )
 
-    # When using torch nightly:
-    # 1. Create a temporary override file to pin torch to the nightly version for dependency resolution
+    # When using torch prerelease (nightly or test):
+    # 1. Create a temporary override file to pin torch to the prerelease version for dependency resolution
     # 2. Use --prerelease=if-necessary-or-explicit to only allow prereleases for explicitly specified packages (torch)
     #    or where all versions of the package are pre-release
-    # 3. Use --index with nightly CPU index for torch resolution
+    # 3. Use --index with prerelease CPU index for torch resolution (nightly or test channel)
     # 4. Use --index-strategy=unsafe-best-match for lockfile GENERATION only
     #    This is required because with first-index (default), uv would either:
     #      - Find torch on PyPI first (no nightly version), or
     #      - Find scipy/etc on nightly index first (missing versions)
     #
     #    Security rationale for using unsafe-best-match during lockfile generation:
-    #    a) User INSTALLATION uses a secure two-step approach, only ever installing torch nightly
-    #       from the explicitly specified nightly index (no unsafe-best-match at install time)
+    #    a) User INSTALLATION uses a secure two-step approach, only ever installing torch prerelease
+    #       from the explicitly specified prerelease index (no unsafe-best-match at install time)
     #    b) The marginal dependency confusion attack surface is limited to the closely monitored
-    #       PyTorch nightly index, which is maintained by PyTorch team. Post-processing prunes any packages that are
+    #       PyTorch prerelease index, which is maintained by PyTorch team. Post-processing prunes any packages that are
     #       ONLY dependencies of torch, eliminating potential attack vectors from torch-exclusive dependencies that
-    #       might only exist on the nightly index. If a package is shared with other dependencies, it's already
+    #       might only exist on the prerelease index. If a package is shared with other dependencies, it's already
     #       being resolved from PyPI and subject to normal security scanning.
     #    c) Lockfile generation runs on maintainer machines, not user machines
     #    d) Generated lockfile pins exact package versions from PyPI
     #
     # 5. Use --no-emit-package=torch to exclude torch from output (installed separately with backend)
     # 6. Post-process to prune torch-only dependencies (see prune_torch_only_deps)
-    if [[ "${use_nightly}" == "true" && -n "${TORCH_NIGHTLY_VERSION}" ]]; then
+    if [[ "${use_prerelease}" == "true" && -n "${TORCH_PRE_VERSION}" ]]; then
+        # Determine channel from global variable
+        local channel="${TORCH_PRE_CHANNEL:-nightly}"  # Default to nightly if not set
+
         local torch_override_file=$(mktemp)
-        echo "torch==${TORCH_NIGHTLY_VERSION}" > "${torch_override_file}"
+        echo "torch==${TORCH_PRE_VERSION}" > "${torch_override_file}"
 
         compile_cmd+=(
             --prerelease=if-necessary-or-explicit
             --override "${torch_override_file}"
-            --index "https://download.pytorch.org/whl/nightly/cpu"
+            --index "https://download.pytorch.org/whl/${channel}/cpu"
             --index-strategy unsafe-best-match  # for lockfile generation only, see comment above
             --no-emit-package torch
         )
 
-        echo "  Using torch nightly: ${TORCH_NIGHTLY_VERSION} (excluded from output, dependencies resolved)"
+        echo "  Using torch ${channel}: ${TORCH_PRE_VERSION} (excluded from output, dependencies resolved)"
         "${compile_cmd[@]}"
 
         rm -f "${torch_override_file}"
@@ -174,11 +190,14 @@ generate_lockfile() {
         # Prune torch-only dependencies to minimize dependency confusion attack surface
         prune_torch_only_deps "${output_file}"
 
-        echo "✓ Generated ${output_file} (torch ${TORCH_NIGHTLY_VERSION} excluded, torch-only deps pruned)"
+        echo "✓ Generated ${output_file} (torch ${TORCH_PRE_VERSION} excluded, torch-only deps pruned)"
     else
         "${compile_cmd[@]}"
         echo "✓ Generated ${output_file}"
     fi
+
+    # Return to original directory
+    popd > /dev/null
 }
 
 # Generate both lock files
@@ -186,21 +205,25 @@ generate_lockfile() {
 #   packages like contourpy that have Python version requirements
 # - Oldest: Python 3.10 (minimum supported), lowest resolution
 #
-# When torch nightly is configured:
+# When torch prerelease (nightly or test) is configured:
 # - requirements.txt excludes torch (installed separately with appropriate backend)
-# - torch dependencies are still resolved against the nightly version
+# - torch dependencies are still resolved against the prerelease version
 # - requirements-oldest.txt uses stable torch (for minimum version testing)
 
-USE_NIGHTLY="false"
-if [[ -n "${TORCH_NIGHTLY_VERSION}" ]]; then
-    USE_NIGHTLY="true"
-    echo "Torch nightly mode: ${TORCH_NIGHTLY_VERSION}"
+USE_PRERELEASE="false"
+if [[ -n "${TORCH_PRE_VERSION}" ]]; then
+    USE_PRERELEASE="true"
+    echo "Torch ${TORCH_PRE_CHANNEL} mode: ${TORCH_PRE_VERSION}"
 fi
 
 # Generate torch override file for manual installation
 generate_torch_override
 
-generate_lockfile "highest" "${CI_DIR}/requirements.txt" "3.10" "${USE_NIGHTLY}"
+# Sync metadata from utils.py to pyproject.toml before generating lockfiles
+echo "Syncing metadata from utils.py to pyproject.toml..."
+python "${SCRIPT_DIR}/sync_metadata.py"
+
+generate_lockfile "highest" "${CI_DIR}/requirements.txt" "3.10" "${USE_PRERELEASE}"
 generate_lockfile "lowest-direct" "${CI_DIR}/requirements-oldest.txt" "3.10" "false"
 
 echo ""
@@ -208,29 +231,29 @@ echo "Generated lock files:"
 echo "  - ${CI_DIR}/requirements.txt (highest resolution, for latest tests)"
 echo "  - ${CI_DIR}/requirements-oldest.txt (lowest resolution, for oldest tests)"
 echo ""
-if [[ -n "${TORCH_NIGHTLY_VERSION}" ]]; then
-    echo "⚠️  Torch nightly mode: ${TORCH_NIGHTLY_VERSION}"
-    echo "   requirements.txt excludes torch (dependencies resolved against nightly)"
+if [[ -n "${TORCH_PRE_VERSION}" ]]; then
+    echo "⚠️  Torch ${TORCH_PRE_CHANNEL} mode: ${TORCH_PRE_VERSION}"
+    echo "   requirements.txt excludes torch (dependencies resolved against ${TORCH_PRE_CHANNEL})"
     echo ""
     echo "Generated override file:"
-    echo "  - ${CI_DIR}/torch_override.txt (for manual nightly installation reference)"
+    echo "  - ${CI_DIR}/torch_override.txt (for manual prerelease installation reference)"
     echo ""
-    echo "Manual installation with nightly (two-step approach):"
-    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_NIGHTLY_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128"
+    echo "Manual installation with prerelease (two-step approach):"
+    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_PRE_VERSION} --index-url https://download.pytorch.org/whl/${TORCH_PRE_CHANNEL}/cu128"
     echo "  2. UV_OVERRIDE=requirements/ci/overrides.txt uv pip install -e \".[all]\""
     echo ""
     echo "Or with locked requirements:"
-    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_NIGHTLY_VERSION} --index-url https://download.pytorch.org/whl/nightly/cu128"
+    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_PRE_VERSION} --index-url https://download.pytorch.org/whl/${TORCH_PRE_CHANNEL}/cu128"
     echo "  2. UV_OVERRIDE=requirements/ci/overrides.txt uv pip install -e . -r requirements/ci/requirements.txt"
     echo ""
     echo "Docker image installation (CUDA):"
-    echo "  Ensure Dockerfile installs: torch==${TORCH_NIGHTLY_VERSION} from nightly/cu128 index"
+    echo "  Ensure Dockerfile installs: torch==${TORCH_PRE_VERSION} from ${TORCH_PRE_CHANNEL}/cu128 index"
     echo ""
     echo "Azure Pipelines (Docker with pre-installed torch):"
     echo "  UV_OVERRIDE=requirements/ci/overrides.txt uv pip install -e . -r requirements/ci/requirements.txt"
     echo ""
     echo "GitHub Actions (CPU):"
-    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_NIGHTLY_VERSION}+cpu --index-url https://download.pytorch.org/whl/nightly/cpu"
+    echo "  1. uv pip install --prerelease=if-necessary-or-explicit torch==${TORCH_PRE_VERSION}+cpu --index-url https://download.pytorch.org/whl/${TORCH_PRE_CHANNEL}/cpu"
     echo "  2. UV_OVERRIDE=requirements/ci/overrides.txt uv pip install -e . -r requirements/ci/requirements.txt"
 else
     echo "Standard installation (stable torch):"
diff --git a/requirements/utils/sync_metadata.py b/requirements/utils/sync_metadata.py
new file mode 100755
index 0000000..3c14b10
--- /dev/null
+++ b/requirements/utils/sync_metadata.py
@@ -0,0 +1,164 @@
+#!/usr/bin/env python3
+"""Sync minimum version metadata from utils.py to pyproject.toml.
+
+This script reads the actual version constraints from
+src/finetuning_scheduler/dynamic_versioning/utils.py and updates
+the informational [tool.fts.min-versions] section in pyproject.toml
+to match.
+
+This ensures the user-facing metadata stays synchronized with the
+actual version constraints used during installation.
+"""
+
+import re
+import sys
+from pathlib import Path
+
+
+def extract_torch_version(utils_content: str) -> str:
+    """Extract torch version constraint from BASE_DEPENDENCIES.
+
+    Args:
+        utils_content: Content of utils.py
+
+    Returns:
+        Full torch version constraint (e.g., ">=2.6.0")
+    """
+    # Look for: BASE_DEPENDENCIES = [..., "torch>=X.Y.Z", ...]
+    match = re.search(r'BASE_DEPENDENCIES\s*=\s*\[.*?"torch(>=?[0-9.]+(?:,[<>=0-9.]+)?)"', utils_content, re.DOTALL)
+    if not match:
+        raise ValueError("Could not find torch version in BASE_DEPENDENCIES")
+    return match.group(1)
+
+
+def extract_lightning_version(utils_content: str) -> str:
+    """Extract lightning version constraint from LIGHTNING_VERSION.
+
+    Args:
+        utils_content: Content of utils.py
+
+    Returns:
+        Full lightning version constraint (e.g., ">=2.6.0,<2.6.1")
+    """
+    # Look for: LIGHTNING_VERSION = ">=X.Y.Z,<X.Y.Z+1"
+    match = re.search(r'LIGHTNING_VERSION\s*=\s*"(>=?[0-9.]+(?:,[<>=0-9.]+)?)"', utils_content)
+    if not match:
+        raise ValueError("Could not find LIGHTNING_VERSION")
+    return match.group(1)
+
+
+def extract_python_version(pyproject_content: str) -> str:
+    """Extract minimum Python version from pyproject.toml requires-python.
+
+    Args:
+        pyproject_content: Content of pyproject.toml
+
+    Returns:
+        Minimum python version (e.g., "3.10")
+    """
+    # Look for: requires-python = ">=X.Y"
+    match = re.search(r'requires-python\s*=\s*">=([0-9.]+)"', pyproject_content)
+    if not match:
+        raise ValueError("Could not find requires-python in pyproject.toml")
+    return match.group(1)
+
+
+def update_metadata_section(pyproject_content: str, torch_ver: str, lightning_ver: str, python_ver: str) -> str:
+    """Update the [tool.fts.min-versions] section with new versions.
+
+    Args:
+        pyproject_content: Content of pyproject.toml
+        torch_ver: Minimum torch version
+        lightning_ver: Minimum lightning version
+        python_ver: Minimum python version
+
+    Returns:
+        Updated pyproject.toml content
+    """
+    # Use a line-by-line approach for safer replacement
+    lines = pyproject_content.split('\n')
+    updated_lines = []
+    in_metadata_section = False
+    metadata_updated = False
+
+    for i, line in enumerate(lines):
+        if '[tool.fts.min-versions]' in line:
+            in_metadata_section = True
+            updated_lines.append(line)
+        elif in_metadata_section:
+            # Update torch, lightning, python lines
+            if line.strip().startswith('torch'):
+                updated_lines.append(f'torch = "{torch_ver}"')
+                metadata_updated = True
+            elif line.strip().startswith('lightning'):
+                updated_lines.append(f'lightning = "{lightning_ver}"')
+            elif line.strip().startswith('python'):
+                updated_lines.append(f'python = "{python_ver}"  # Defined in pyproject.toml requires-python')
+                in_metadata_section = False  # End of section
+            else:
+                updated_lines.append(line)
+        else:
+            updated_lines.append(line)
+
+    if not metadata_updated:
+        raise ValueError("Could not find [tool.fts.min-versions] section to update")
+
+    return '\n'.join(updated_lines)
+
+
+def main():
+    """Main entry point."""
+    # Locate files
+    script_dir = Path(__file__).parent
+    repo_root = script_dir.parent.parent
+
+    utils_path = repo_root / "src" / "finetuning_scheduler" / "dynamic_versioning" / "utils.py"
+    pyproject_path = repo_root / "pyproject.toml"
+
+    if not utils_path.exists():
+        print(f"Error: Could not find {utils_path}", file=sys.stderr)
+        return 1
+
+    if not pyproject_path.exists():
+        print(f"Error: Could not find {pyproject_path}", file=sys.stderr)
+        return 1
+
+    # Read files
+    utils_content = utils_path.read_text()
+    pyproject_content = pyproject_path.read_text()
+
+    # Extract versions
+    try:
+        torch_ver = extract_torch_version(utils_content)
+        lightning_ver = extract_lightning_version(utils_content)
+        python_ver = extract_python_version(pyproject_content)
+    except ValueError as e:
+        print(f"Error extracting versions: {e}", file=sys.stderr)
+        return 1
+
+    print("Extracted versions from utils.py:")
+    print(f"  torch: {torch_ver}")
+    print(f"  lightning: {lightning_ver}")
+    print(f"  python: {python_ver} (from pyproject.toml requires-python)")
+
+    # Update pyproject.toml
+    try:
+        updated_content = update_metadata_section(pyproject_content, torch_ver, lightning_ver, python_ver)
+    except ValueError as e:
+        print(f"Error updating pyproject.toml: {e}", file=sys.stderr)
+        return 1
+
+    # Check if anything changed
+    if updated_content == pyproject_content:
+        print("✓ Metadata already up to date")
+        return 0
+
+    # Write updated content
+    pyproject_path.write_text(updated_content)
+    print(f"✓ Updated {pyproject_path}")
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/build_fts_env.sh b/scripts/build_fts_env.sh
index 61efa4f..28481de 100755
--- a/scripts/build_fts_env.sh
+++ b/scripts/build_fts_env.sh
@@ -10,17 +10,14 @@
 #   ./build_fts_env.sh --repo_home=${HOME}/repos/finetuning-scheduler --target_env_name=fts_oldest --oldest
 # build release:
 #   ./build_fts_env.sh --repo_home=${HOME}/repos/fts-release --target_env_name=fts_release
-# build latest with torch test channel:
-#   ./build_fts_env.sh --repo_home=~/repos/finetuning-scheduler --target_env_name=fts_latest --torch_test_channel
 # build latest from a package from source:
 #   ./build_fts_env.sh --repo_home=${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --from-source="lightning:${HOME}/repos/lightning:pytorch"
 #
-# To use a specific PyTorch nightly, edit requirements/ci/torch-nightly.txt
+# To configure PyTorch version (nightly/test/stable), edit requirements/ci/torch-pre.txt
 set -eo pipefail
 
 unset repo_home
 unset target_env_name
-unset torch_test_channel
 unset uv_install_flags
 unset no_commit_pin
 unset venv_dir
@@ -28,6 +25,7 @@ unset oldest
 declare -a from_source_specs=()
 
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
 source "${SCRIPT_DIR}/infra_utils.sh"
 
 usage(){
@@ -36,7 +34,6 @@ Usage: $0
    [ --repo_home input]
    [ --target_env_name input ]
    [ --oldest ]                # Use oldest CI requirements (Python 3.10, requirements-oldest.txt)
-   [ --torch_test_channel ]    # Use PyTorch test/RC channel
    [ --uv_install_flags "flags" ]
    [ --no_commit_pin ]
    [ --venv-dir input ]
@@ -51,8 +48,6 @@ Usage: $0
     #   ./build_fts_env.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_oldest --oldest --venv-dir=/mnt/cache/\${USER}/.venvs
     # build release:
     #   ./build_fts_env.sh --repo_home=\${HOME}/repos/fts-release --target_env_name=fts_release
-    # build latest with torch test channel:
-    #   ./build_fts_env.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --torch_test_channel
     # build latest with no cache:
     #   ./build_fts_env.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --uv_install_flags="--no-cache"
     # build latest without using CI commit pinning:
@@ -60,14 +55,15 @@ Usage: $0
     # build latest from Lightning source:
     #   ./build_fts_env.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --from-source="lightning:\${HOME}/repos/lightning:pytorch"
 
-    # To use a specific PyTorch nightly, edit requirements/ci/torch-nightly.txt:
-    #   Line 1: torch version (e.g., 2.10.0.dev20251124)
+    # To configure PyTorch version, edit requirements/ci/torch-pre.txt:
+    #   Line 1: torch version (e.g., 2.10.0 for test, 2.10.0.dev20251124 for nightly)
     #   Line 2: CUDA target (e.g., cu128)
+    #   Line 3: channel type (test or nightly)
 EOF
 exit 1
 }
 
-args=$(getopt -o '' --long repo_home:,target_env_name:,oldest,torch_test_channel,uv_install_flags:,no_commit_pin,venv-dir:,from-source:,help -- "$@")
+args=$(getopt -o '' --long repo_home:,target_env_name:,oldest,uv_install_flags:,no_commit_pin,venv-dir:,from-source:,help -- "$@")
 if [[ $? -gt 0 ]]; then
   usage
 fi
@@ -79,7 +75,6 @@ do
     --repo_home)  repo_home=$2    ; shift 2  ;;
     --target_env_name)  target_env_name=$2  ; shift 2 ;;
     --oldest)   oldest=1 ; shift  ;;
-    --torch_test_channel)   torch_test_channel=1 ; shift  ;;
     --uv_install_flags)   uv_install_flags=$2 ; shift 2 ;;
     --no_commit_pin)   no_commit_pin=1 ; shift  ;;
     --venv-dir)   venv_dir=$2 ; shift 2 ;;
@@ -109,32 +104,7 @@ if [[ ${#from_source_specs[@]} -gt 0 ]]; then
     from_source_spec=$(IFS=';'; echo "${from_source_specs[*]}")
 fi
 
-# Read torch nightly configuration from requirements/ci/torch-nightly.txt
-# Returns two values via global variables: TORCH_NIGHTLY_VERSION and TORCH_NIGHTLY_CUDA
-read_torch_nightly_config() {
-    local nightly_file="${repo_home}/requirements/ci/torch-nightly.txt"
-    TORCH_NIGHTLY_VERSION=""
-    TORCH_NIGHTLY_CUDA=""
-
-    if [[ -f "${nightly_file}" ]]; then
-        # Read non-comment, non-empty lines
-        local lines=()
-        while IFS= read -r line || [[ -n "$line" ]]; do
-            # Skip comments and empty lines
-            [[ "$line" =~ ^# ]] && continue
-            [[ -z "$line" ]] && continue
-            lines+=("$line")
-        done < "${nightly_file}"
-
-        # First line is torch version, second is CUDA target
-        if [[ ${#lines[@]} -ge 1 ]]; then
-            TORCH_NIGHTLY_VERSION="${lines[0]}"
-        fi
-        if [[ ${#lines[@]} -ge 2 ]]; then
-            TORCH_NIGHTLY_CUDA="${lines[1]}"
-        fi
-    fi
-}
+# Read torch prerelease configuration (now handled by infra_utils.sh::read_torch_pre_config)
 
 clear_activate_env(){
     local python_version=$1
@@ -159,33 +129,29 @@ base_env_build(){
 
     clear_activate_env ${python_version}
 
-    # Check for torch nightly configuration (skip for oldest builds)
+    # Check for torch prerelease configuration (skip for oldest builds)
     if [[ -z ${oldest} ]]; then
-        read_torch_nightly_config
+        read_torch_pre_config
     fi
 
     # Handle PyTorch version selection (pre-install before FTS dependencies)
-    # Priority: oldest (stable from lock) > torch nightly from config > torch test channel > stable (via --torch-backend in fts_install)
+    # Priority: oldest (stable from lock) > torch prerelease from config > stable (via --torch-backend in fts_install)
     if [[ -n ${oldest} ]]; then
         # For oldest builds, torch is installed from requirements-oldest.txt (stable version)
         echo "Using torch stable from requirements-oldest.txt for oldest build"
-    elif [[ -n "${TORCH_NIGHTLY_VERSION}" ]]; then
-        # Nightly version from torch-nightly.txt with specified CUDA backend
-        local cuda_target="${TORCH_NIGHTLY_CUDA:-cu128}"  # Default to cu128 if not specified
-        local torch_pkg="torch==${TORCH_NIGHTLY_VERSION}"
-        local torch_index_url="https://download.pytorch.org/whl/nightly/${cuda_target}"
-        echo "Pre-installing PyTorch nightly from torch-nightly.txt: ${torch_pkg}"
+    elif [[ -n "${TORCH_PRE_VERSION}" ]]; then
+        # Prerelease (nightly or test) configured in torch-pre.txt
+        local cuda_target="${TORCH_PRE_CUDA:-cu128}"  # Default to cu128 if not specified
+        local torch_pkg="torch==${TORCH_PRE_VERSION}"
+        local torch_index_url=$(get_torch_index_url "${TORCH_PRE_CHANNEL}" "${cuda_target}")
+
+        echo "Pre-installing PyTorch ${TORCH_PRE_CHANNEL} from torch-pre.txt: ${torch_pkg}"
+        echo "  Channel: ${TORCH_PRE_CHANNEL}"
         echo "  CUDA target: ${cuda_target}"
         echo "  Index URL: ${torch_index_url}"
+
         uv pip install ${uv_install_flags} --prerelease=allow "${torch_pkg}" --index-url "${torch_index_url}"
-        log_torch_version "after PyTorch nightly pre-install"
-    elif [[ -n ${torch_test_channel} ]]; then
-        # Test/RC channel - pre-install torch with test index and auto backend for GPU detection
-        local torch_index_url="https://download.pytorch.org/whl/test"
-        echo "Pre-installing PyTorch from test channel: ${torch_index_url}"
-        echo "  Using --torch-backend=auto for GPU auto-detection"
-        uv pip install ${uv_install_flags} --prerelease=allow torch --index-url ${torch_index_url} --torch-backend=auto
-        log_torch_version "after PyTorch test channel pre-install"
+        log_torch_version "after PyTorch ${TORCH_PRE_CHANNEL} pre-install"
     fi
     # For stable builds, torch will be installed via FTS dependencies with --torch-backend=auto
 }
@@ -218,16 +184,10 @@ fts_install(){
         echo "Using oldest requirements file: ${req_file}"
         # Oldest builds use torch stable from lock file, need --torch-backend=auto
         torch_backend_flag="--torch-backend=auto"
-    elif [[ -n "${TORCH_NIGHTLY_VERSION}" || -n ${torch_test_channel} ]]; then
-        # Torch already pre-installed (nightly or test channel)
-        # When nightly: requirements.txt already has torch filtered during lock generation
-        # When test channel: filter at runtime
-        if [[ -n ${torch_test_channel} ]]; then
-            echo "Torch test channel pre-installed, filtering torch from requirements..."
-            grep -v '^torch==' "${req_file}" > /tmp/requirements_no_torch.txt
-            req_file="/tmp/requirements_no_torch.txt"
-        fi
-        echo "Using requirements without torch (pre-installed)"
+    elif [[ -n "${TORCH_PRE_VERSION}" ]]; then
+        # Torch prerelease already pre-installed (nightly or test channel)
+        # requirements.txt already has torch filtered during lock generation
+        echo "Using requirements without torch (pre-installed ${TORCH_PRE_CHANNEL})"
     else
         # Use auto torch backend for GPU detection
         torch_backend_flag="--torch-backend=auto"
diff --git a/scripts/gen_fts_coverage.sh b/scripts/gen_fts_coverage.sh
index 68b856f..1c14249 100755
--- a/scripts/gen_fts_coverage.sh
+++ b/scripts/gen_fts_coverage.sh
@@ -5,8 +5,6 @@ set -eo pipefail
 
 unset repo_home
 unset target_env_name
-unset torch_dev_ver
-unset torch_test_channel
 unset no_rebuild_base
 unset include_experimental
 unset uv_install_flags
@@ -26,8 +24,6 @@ Usage: $0
    [ --repo_home input]
    [ --target_env_name input ]
    [ --oldest ]                  # Use oldest CI requirements (Python 3.10, requirements-oldest.txt)
-   [ --torch_dev_ver input ]
-   [ --torch_test_channel ]
    [ --no_rebuild_base ]
    [ --no-special ]              # Skip special tests (standalone/experimental), run only main test suite
    [ --include_experimental ]
@@ -42,25 +38,20 @@ Usage: $0
 	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --no_rebuild_base
 	# generate oldest CI build coverage (matches CI oldest matrix):
 	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_oldest --oldest --no-special --venv-dir=/mnt/cache/\${USER}/.venvs
-	# generate fts_latest coverage with a given torch_dev_version:
-	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --torch_dev_ver=dev20240201
-    # generate fts_latest coverage, rebuilding base fts_latest with PyTorch test channel and run tests that require experimental patches:
-    #   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --torch_test_channel --include_experimental
-	# generate fts_release coverage, rebuilding the base fts_release environment with PyTorch stable channel:
+	# generate fts_release coverage, rebuilding the base fts_release environment:
 	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/fts-release --target_env_name=fts_release
 	# generate fts_release coverage, rebuilding the base fts_release environment with PyTorch test channel:
-	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/fts-release --target_env_name=fts_release --torch_test_channel
 	# generate fts_latest coverage with explicit venv directory (recommended for hardlink performance):
 	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --venv-dir=/mnt/cache/\${USER}/.venvs
 	# generate fts_release coverage without using CI commit pinning:
 	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/fts-release --target_env_name=fts_release --no_commit_pin
 	# dry-run mode: setup environment and show what tests would run without executing them:
-	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --torch_dev_ver=dev20240201 --dry-run
+	#   ./gen_fts_coverage.sh --repo_home=\${HOME}/repos/finetuning-scheduler --target_env_name=fts_latest --dry-run
 EOF
 exit 1
 }
 
-args=$(getopt -o '' --long repo_home:,target_env_name:,oldest,torch_dev_ver:,torch_test_channel,no_rebuild_base,no-special,include_experimental,uv_install_flags:,no_commit_pin,venv-dir:,from-source:,dry-run,help -- "$@")
+args=$(getopt -o '' --long repo_home:,target_env_name:,oldest,no_rebuild_base,no-special,include_experimental,uv_install_flags:,no_commit_pin,venv-dir:,from-source:,dry-run,help -- "$@")
 if [[ $? -gt 0 ]]; then
   usage
 fi
@@ -72,8 +63,6 @@ do
     --repo_home)  repo_home=$2    ; shift 2  ;;
     --target_env_name)  target_env_name=$2  ; shift 2 ;;
     --oldest)   oldest=1 ; shift  ;;
-    --torch_dev_ver)   torch_dev_ver=$2   ; shift 2 ;;
-    --torch_test_channel)   torch_test_channel=1 ; shift  ;;
     --no_rebuild_base)   no_rebuild_base=1 ; shift  ;;
     --no-special)   no_special=1 ; shift  ;;
     --include_experimental)   include_experimental=1 ; shift  ;;
@@ -159,18 +148,10 @@ env_rebuild(){
 
     case $1 in
         fts_latest|fts_oldest)
-            if [[ -n ${torch_dev_ver} ]]; then
-                cmd_args+=("--torch_dev_ver=${torch_dev_ver}")
-			elif [[ $torch_test_channel -eq 1 ]]; then
-                cmd_args+=("--torch_test_channel")
-            fi
             log_msg "Final build command: ${cmd_args[*]}"
             "${cmd_args[@]}"
             ;;
         fts_release)
-            if [[ $torch_test_channel -eq 1 ]]; then
-                cmd_args+=("--torch_test_channel")
-            fi
             log_msg "Final build command: ${cmd_args[*]}"
             "${cmd_args[@]}"
             ;;
diff --git a/scripts/infra_utils.sh b/scripts/infra_utils.sh
index ede420f..a7c174a 100755
--- a/scripts/infra_utils.sh
+++ b/scripts/infra_utils.sh
@@ -248,3 +248,56 @@ install_from_source_packages(){
         fi
     done
 }
+
+# Read torch prerelease configuration from requirements/ci/torch-pre.txt
+# Returns values via global variables:
+#   TORCH_PRE_VERSION - torch version to install
+#   TORCH_PRE_CUDA    - CUDA target for local builds
+#   TORCH_PRE_CHANNEL - channel type: "test" or "nightly"
+# Note: Requires REPO_ROOT to be set (usually via SCRIPT_DIR)
+read_torch_pre_config() {
+    TORCH_PRE_VERSION=""
+    TORCH_PRE_CUDA=""
+    TORCH_PRE_CHANNEL=""
+
+    # Determine repo root from SCRIPT_DIR if not already set
+    local repo_root="${REPO_ROOT:-$(cd "${SCRIPT_DIR}/.." && pwd)}"
+    local pre_file="${repo_root}/requirements/ci/torch-pre.txt"
+
+    if [[ ! -f "${pre_file}" ]]; then
+        return
+    fi
+
+    # Read non-comment, non-empty lines
+    local lines=($(grep -v '^#' "${pre_file}" | grep -v '^$' || true))
+
+    if [[ ${#lines[@]} -ge 3 ]]; then
+        TORCH_PRE_VERSION="${lines[0]}"
+        TORCH_PRE_CUDA="${lines[1]}"
+        TORCH_PRE_CHANNEL="${lines[2]}"
+
+        # Validate channel
+        if [[ "${TORCH_PRE_CHANNEL}" != "test" && "${TORCH_PRE_CHANNEL}" != "nightly" ]]; then
+            echo "ERROR: Invalid channel '${TORCH_PRE_CHANNEL}' in ${pre_file}" >&2
+            echo "Must be 'test' or 'nightly'" >&2
+            return 1
+        fi
+    fi
+}
+
+# Get torch index URL based on channel and CUDA target
+# Args: $1 = channel ("test" or "nightly"), $2 = cuda_target (e.g., "cu128" or "cpu")
+# Returns: PyTorch wheel index URL
+get_torch_index_url() {
+    local channel="$1"
+    local cuda_target="${2:-cpu}"
+
+    if [[ "${channel}" == "test" ]]; then
+        echo "https://download.pytorch.org/whl/test/${cuda_target}"
+    elif [[ "${channel}" == "nightly" ]]; then
+        echo "https://download.pytorch.org/whl/nightly/${cuda_target}"
+    else
+        echo "ERROR: Invalid channel: ${channel}" >&2
+        return 1
+    fi
+}
diff --git a/src/finetuning_scheduler/dynamic_versioning/utils.py b/src/finetuning_scheduler/dynamic_versioning/utils.py
index abf03b7..4f08578 100644
--- a/src/finetuning_scheduler/dynamic_versioning/utils.py
+++ b/src/finetuning_scheduler/dynamic_versioning/utils.py
@@ -18,6 +18,12 @@
 # -----------------------------------------------------------------------------
 # Lightning Configuration
 # -----------------------------------------------------------------------------
+#
+# These version constraints are the single source of truth for minimum versions.
+# They are used by setup.py to generate dynamic dependencies at build time.
+#
+# For visibility, these values are also documented in pyproject.toml under
+# [tool.fts.min-versions] (informational only - not used during installation).
 
 # Shared version constraint for all Lightning packages
 LIGHTNING_VERSION = ">=2.6.0,<2.6.1"
@@ -43,6 +49,9 @@
 
 # Base dependencies (torch + Lightning are handled dynamically)
 # These are the core dependencies that are always installed
+#
+# Note: For visibility, minimum versions are also documented in pyproject.toml
+# under [tool.fts.min-versions] (informational only).
 BASE_DEPENDENCIES = [
     "torch>=2.6.0",
 ]
diff --git a/src/finetuning_scheduler/fts.py b/src/finetuning_scheduler/fts.py
index 26b6202..f387307 100644
--- a/src/finetuning_scheduler/fts.py
+++ b/src/finetuning_scheduler/fts.py
@@ -9,12 +9,9 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-r"""
-Fine-Tuning Scheduler
-^^^^^^^^^^^^^^^^^^^^^
+r"""Fine-Tuning Scheduler.
 
 Used to implement flexible fine-tuning training schedules
-
 """
 import logging
 import os
@@ -41,6 +38,7 @@
     ScheduleImplMixin,
     ScheduleParsingMixin,
     STRATEGY_ADAPTERS,
+    _discover_strategy_adapters
 )
 from finetuning_scheduler.strategy_adapters.base import StrategyAdapter
 from finetuning_scheduler.types import ParamGroupAddable
@@ -111,7 +109,7 @@ def __init__(
         reinit_optim_cfg: Optional[Dict] = None,
         reinit_lr_cfg: Optional[Dict] = None,
         strategy_adapter_cfg: Optional[Dict] = None,
-        custom_strategy_adapter: Optional[Dict[str, str]] = None,
+        custom_strategy_adapters: Optional[Dict[str, str]] = None,
         allow_untested: bool = False,
         apply_lambdas_new_pgs: bool = False,
         logging_level: int = logging.INFO,
@@ -223,11 +221,15 @@ def __init__(
                 :external+pl:class:`~lightning.pytorch.strategies.Strategy`. See the relevant
                 :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter` documentation for strategy-specific
                 configuration options. Defaults to None.
-            custom_strategy_adapter: A dictionary associating the canonical ``strategy_flag`` associated with a
-                :external+pl:class:`~lightning.pytorch.strategies.Strategy` (potentially a custom user-registered one)
-                to the fully qualified path of a
-                :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter` subclass. This is an experimental
-                feature that is subject to change. Requires ``allow_untested`` to be set to ``True``. Defaults to None.
+            custom_strategy_adapters: A dictionary mapping PyTorch Lightning strategy flags (canonical strategy
+                names like ``"single_device"``, ``"auto"``, ``"ddp"``, etc.) to strategy adapter references. Multiple
+                ``strategy_flag`` keys can be associated with the same adapter. The adapter reference can be: (1) an
+                entry point name registered under ``finetuning_scheduler.strategy_adapters`` (see
+                :ref:`strategy_adapter_entry_points`) (2) a fully
+                qualified :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter` subclass path in the
+                format ``"module.path:ClassName"`` or (3) a fully qualified dot path in the format
+                ``"module.path.ClassName"``. This is an experimental feature that is subject to change.
+                Defaults to None.
             apply_lambdas_new_pgs: If ``True``, applies most recent lambda in ``lr_lambdas`` list to newly added
                 optimizer groups for lr schedulers that have a ``lr_lambdas`` attribute. Note this option only applies
                 to phases without reinitialized lr schedulers. Phases with defined lr scheduler reinitialization configs
@@ -270,7 +272,7 @@ def __init__(
         self.reinit_optim_cfg = reinit_optim_cfg
         self.reinit_lr_cfg = reinit_lr_cfg
         self.strategy_adapter_cfg = strategy_adapter_cfg or {}
-        self.custom_strategy_adapter = custom_strategy_adapter
+        self.custom_strategy_adapters = custom_strategy_adapters
         self.allow_untested = allow_untested
         self.apply_lambdas_new_pgs = apply_lambdas_new_pgs
         self.enforce_phase0_params = enforce_phase0_params
@@ -527,6 +529,19 @@ def restore_best_ckpt(self) -> None:
         except KeyError as ke:  # we may want to allow training to progress conditioned on context of restoration
             self._maybe_allow_incompatible_reinit_ckpt(ke)
         self.trainer._checkpoint_connector.restore_datamodule()
+
+        # Inspect state before strategy adapter transformation
+        loaded_ckpt = self.trainer._checkpoint_connector._loaded_checkpoint
+
+        # Allow strategy-specific adapters to transform the state dict before model restoration
+        if self.strategy_adapter is not None:
+            try:
+                loaded_ckpt = self.strategy_adapter.before_restore_model(loaded_ckpt)
+                # assign back in case adapter replaced or modified the checkpoint
+                self.trainer._checkpoint_connector._loaded_checkpoint = loaded_ckpt
+            except Exception as err:
+                rank_zero_warn(f"Strategy adapter before_restore_model hook raised: {err}")
+
         self.trainer._checkpoint_connector.restore_model()
         # we need to override checkpoint_connector.restore_training_state() to bypass loop restoration
         # if additional customizations are required, may make sense to subclass _CheckpointConnector at some point
@@ -656,6 +671,7 @@ def _strategy_setup(self, trainer: "pl.Trainer") -> None:
         connect_flg = getattr(trainer._accelerator_connector, "_strategy_flag", "")
         strategy_flag = getattr(connect_flg, "strategy_name", connect_flg.__class__.__name__.lower()) if \
             isinstance(connect_flg, Strategy) else connect_flg
+        _discover_strategy_adapters()  # Discover strategy adapter plugins lazily
         supported = [t.lower() for t in self._supported_strategy_flags()]
         if strategy_flag and strategy_flag not in supported:  # type: ignore[attr-defined]
             if not self.allow_untested:
@@ -671,8 +687,8 @@ def _strategy_setup(self, trainer: "pl.Trainer") -> None:
                     f" '{strategy}' because ``allow_untested`` is ``True``."  # type: ignore[attr-defined]
                 )
                 rank_zero_warn(warn_msg)
-        if self.custom_strategy_adapter:
-            strategy_cls = self._import_strategy_adapter(strategy_flag, self.custom_strategy_adapter)
+        if self.custom_strategy_adapters:
+            strategy_cls = self._resolve_strategy_adapter(strategy_flag, self.custom_strategy_adapters)
             rank_zero_info(
                 f"Imported custom strategy adapter class type `{strategy_cls}` associated with the current strategy"
                 f" `{strategy_flag}`."
@@ -703,13 +719,15 @@ def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: s
         Raises:
             SystemExit: Gracefully exit before training if only generating and not executing a fine-tuning schedule.
         """
+        # TODO: might not be necessary if we move introspection back to IT where it belongs
+        # Save pl_module/trainer refs early to allow strategy adapter selection to introspect the module
+        self.pl_module, self.trainer = pl_module, trainer  # save pl_module/trainer refs for downstream convenience
         self._callback_dep_setup(trainer, pl_module, stage)
         self._strategy_setup(trainer)
-        self.pl_module, self.trainer = pl_module, trainer  # save pl_module/trainer refs for downstream convenience
         if self.gen_ft_sched_only:
             if trainer.is_global_zero:
                 assert self.log_dir is not None, "log_dir must be set to generate a fine-tuning schedule"
-                _ = ScheduleImplMixin.gen_ft_schedule(self.pl_module, self.log_dir)
+                _ = self.strategy_adapter.gen_ft_schedule(self.log_dir)
                 log.info("Bypassing training, generating fine-tuning schedule for review and subsequent fine-tuning")
             raise SystemExit(0)
         if not self.epoch_transitions_only:
diff --git a/src/finetuning_scheduler/fts_supporters.py b/src/finetuning_scheduler/fts_supporters.py
index 2991acd..9583032 100644
--- a/src/finetuning_scheduler/fts_supporters.py
+++ b/src/finetuning_scheduler/fts_supporters.py
@@ -67,6 +67,55 @@
 STRATEGY_ADAPTERS = {"fsdp": FSDPStrategyAdapter, "modelparallelstrategy": ModelParallelStrategyAdapter}
 
 
+def _discover_strategy_adapters() -> None:
+    """Discover strategy adapter plugins via entry points.
+
+    This tries to discover user-contributed adapters registered under the
+    ``finetuning_scheduler.strategy_adapters`` entry point group and extends the
+    runtime ``STRATEGY_ADAPTERS`` mapping with any discovered adapters keyed by
+    the entry point name (lowercased).
+
+    Note:
+        This function is called lazily during strategy setup.
+
+    .. warning::
+        This is an :ref:`experimental <versioning:API Stability Classifications>` feature which is
+        still in development. The entry point API and plugin discovery mechanism may change in future
+        releases. See :ref:`strategy_adapter_entry_points` for documentation.
+    """
+    # We now require Python 3.10+ environments so can use the standard importlib.metadata API.
+    from importlib.metadata import entry_points
+    eps = entry_points(group="finetuning_scheduler.strategy_adapters")
+    for ep in eps:
+        try:
+            # prefer using the standard entrypoint loader which handles 'module:attr'
+            try:
+                cls = ep.load()
+            except Exception as ep_err:
+                # fall back to dot notation if a colon-separated import path was not provided
+                try:
+                    val = getattr(ep, 'value', '')
+                    if not val:
+                        raise ValueError(f"Entry point {ep.name} has no value attribute")
+                    module_name, attr = val.split(':', 1) if ':' in val else val.rsplit('.', 1)
+                    module = __import__(module_name, fromlist=[attr])
+                    cls = getattr(module, attr)
+                except Exception as fallback_err:
+                    # If fallback also fails, log both errors and re-raise
+                    rank_zero_warn(
+                        f"Failed to load entry point {ep.name} with ep.load(): {ep_err}. "
+                        f"Fallback import also failed: {fallback_err}"
+                    )
+                    raise
+            if hasattr(ep, "name") and isinstance(ep.name, str):
+                STRATEGY_ADAPTERS[ep.name.lower()] = cls
+                rank_zero_info(f"Discovered strategy adapter entrypoint '{ep.name}' -> {cls}")
+        except Exception as err:
+            rank_zero_warn(f"Failed to load strategy adapter entry point {ep}: {err}")
+
+# Note: _discover_strategy_adapters() is called lazily during strategy setup
+
+
 @dataclass
 class FTSState:
     """Dataclass to encapsulate the :class:`~finetuning_scheduler.fts.FinetuningScheduler` internal state."""
@@ -533,6 +582,7 @@ class ScheduleParsingMixin(ABC):
     ft_schedule: Optional[Union[str, dict]]
     reinit_optim_cfg: Optional[Dict]
     reinit_lr_cfg: Optional[Dict]
+    strategy_adapter: StrategyAdapter  # added to support adapter-based parameter naming
 
     def _validate_ft_sched(self) -> Tuple[int, int]:
         """Ensure the explicitly specified fine-tuning schedule has a valid configuration.
@@ -546,7 +596,8 @@ def _validate_ft_sched(self) -> Tuple[int, int]:
         max_phase = 0
         self._validate_schedule_keys()
         self._validate_reinit_cfg()
-        named_params = dict(self.pl_module.named_parameters()).keys()
+        # Use strategy adapter to get named params (allows TL-style or other custom naming)
+        named_params = self.strategy_adapter.get_named_params_for_schedule_validation().keys()
         model_shared_params = find_shared_parameters(self.pl_module)
         msp_ref = tuple((model_shared_params, set(itertools.chain(*model_shared_params))))
         for depth in self.ft_schedule.keys():  # type: ignore[union-attr]
@@ -1172,8 +1223,8 @@ def _import_reinit_class(
         return reinit_class
 
     @staticmethod
-    def _import_strategy_adapter(strategy_key: str, adapter_map: Dict[str, str]) -> Type[StrategyAdapter]:
-        """Import the custom strategy adapter specified in the ``custom_strategy_adapter`` configuration.
+    def _resolve_strategy_adapter(strategy_key: str, adapter_map: Dict[str, str]) -> Type[StrategyAdapter]:
+        """Resolve the custom strategy adapter specified in the ``custom_strategy_adapters`` configuration.
 
         Args:
             qualname (Dict): The user-provided custom strategy adapter fully qualified class name.
@@ -1190,13 +1241,29 @@ def _import_strategy_adapter(strategy_key: str, adapter_map: Dict[str, str]) ->
             if not qualname:
                 raise MisconfigurationException(
                     f"Current strategy name ({strategy_key}) does not map to a custom strategy adapter in the"
-                    f" provided `custom_strategy_adapter` mapping ({adapter_map})."
+                    f" provided `custom_strategy_adapters` mapping ({adapter_map})."
                 )
-            class_module, class_name = qualname.rsplit(".", 1)
+            # If a short entry point name was provided, check the discovered STRATEGY_ADAPTERS mapping
+            if qualname in STRATEGY_ADAPTERS:
+                return STRATEGY_ADAPTERS[qualname]
+
+            # Accept either 'module.Class' or 'module:Class' form to support both direct import paths and
+            # entry-point-like strings.
+            if ":" in qualname:
+                class_module, class_name = qualname.split(":", 1)
+            else:
+                # Require at least one '.' for module.Class format
+                if "." not in qualname:
+                    raise ValueError(
+                        f"Invalid adapter name '{qualname}'. Must be either a registered plugin name "
+                        f"(found in STRATEGY_ADAPTERS: {list(STRATEGY_ADAPTERS.keys())}) or a fully qualified "
+                        f"class name in 'module.Class' or 'module:Class' format."
+                    )
+                class_module, class_name = qualname.rsplit(".", 1)
             module = __import__(class_module, fromlist=[class_name])
             custom_strategy_adapter_cls = getattr(module, class_name)
             issubclass(custom_strategy_adapter_cls, StrategyAdapter)
-        except (ImportError, AttributeError) as err:
+        except (ImportError, AttributeError, ValueError) as err:
             error_msg = (
                 "Could not import the specified custom strategy adapter class using the provided fully qualified class"
                 f" name ({qualname}). Received the following error while importing: {err}. Please validate specified"
@@ -1373,7 +1440,8 @@ def init_ft_sched(self) -> None:
             self.max_depth = len(self.ft_schedule) - 1
         else:
             self.max_depth = min(self.max_depth, len(self.ft_schedule) - 1)
-        max_phase, max_epoch_wm = self._validate_ft_sched()  # type: ignore[attr-defined]
+        # Delegate schedule validation to strategy adapter (allows custom validation logic)
+        max_phase, max_epoch_wm = self.strategy_adapter.validate_ft_sched()
         # if the final phase is not using EarlyStopping, apply the maximum phase-specified epoch to global max_epochs
         if self.ft_schedule[max_phase]["max_transition_epoch"] >= 0:
             assert self.trainer is not None
@@ -1394,7 +1462,7 @@ def gen_implicit_schedule(self, sched_dir: Union[str, os.PathLike]) -> None:
             sched_dir: directory to which the generated schedule should be written. By default will be
                 ``Trainer.log_dir``.
         """
-        default_ft_schedule = ScheduleImplMixin.gen_ft_schedule(self.pl_module, sched_dir)
+        default_ft_schedule = self.strategy_adapter.gen_ft_schedule(sched_dir)
         assert default_ft_schedule is not None
         rank_zero_info(f"Generated default fine-tuning schedule '{default_ft_schedule}' for iterative fine-tuning")
         self.ft_schedule = self.load_yaml_schedule(default_ft_schedule)
@@ -1429,6 +1497,11 @@ def save_schedule(schedule_name: str, layer_config: Dict, dump_loc: Union[str, o
     def gen_ft_schedule(module: Module, dump_loc: Union[str, os.PathLike]) -> Optional[os.PathLike]:
         """Generate the default fine-tuning schedule using a naive, 2-parameters per-level heuristic.
 
+        .. deprecated:: 2.10.0
+            Direct calls to this static method are deprecated. Use the
+            :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.gen_ft_schedule` instance method
+            instead, which allows strategy adapters to customize schedule generation.
+
         Args:
             module (:class:`~torch.nn.Module`): The :class:`~torch.nn.Module` for which a fine-tuning schedule will be
                 generated
@@ -1438,11 +1511,33 @@ def gen_ft_schedule(module: Module, dump_loc: Union[str, os.PathLike]) -> Option
             :external+pl:class:`~lightning.pytorch.core.module.LightningModule` subclass in use with the suffix
             ``_ft_schedule.yaml``)
         """
+        rank_zero_warn(
+            "Direct calls to ScheduleImplMixin.gen_ft_schedule() are deprecated since v2.10.0 and will be "
+            "removed in v2.12.0. Use strategy_adapter.gen_ft_schedule() instead to allow strategy-specific "
+            "customization."
+        )
+        return ScheduleImplMixin._gen_ft_schedule_impl(module, dump_loc)
+
+    @staticmethod
+    def _gen_ft_schedule_impl(module: Module, dump_loc: Union[str, os.PathLike]) -> Optional[os.PathLike]:
+        """Internal implementation of default fine-tuning schedule generation.
+
+        This method contains the actual schedule generation logic shared between the deprecated static method
+        and the strategy adapter instance method.
+
+        Args:
+            module (:class:`~torch.nn.Module`): The :class:`~torch.nn.Module` for which a fine-tuning schedule will be
+                generated
+            dump_loc: The directory to which the generated schedule (.yaml) should be written
+
+        Returns:
+            os.PathLike: The path to the generated schedule
+        """
         # Note: This initial default fine-tuning schedule generation approach is intentionally simple/naive but is
-        # effective for a suprising fraction of models. Future versions of this callback may use module introspection to
-        # generate default schedules that better accommodate more complex structures and specific architectures if the
-        # callback proves sufficiently useful.
-        log.info(f"Proceeding with dumping default fine-tuning schedule for {module.__class__.__name__}")
+        # effective for a surprising fraction of models. Future versions of this callback may use module
+        # introspection to generate default schedules that better accommodate more complex structures and
+        # specific architectures if the callback proves sufficiently useful.
+        rank_zero_info(f"Proceeding with dumping default fine-tuning schedule for {module.__class__.__name__}")
         param_lists: List = []
         cur_group: List = []
         model_params = list(module.named_parameters())[::-1]
diff --git a/src/finetuning_scheduler/strategy_adapters/__init__.py b/src/finetuning_scheduler/strategy_adapters/__init__.py
index fba81f2..56f031d 100644
--- a/src/finetuning_scheduler/strategy_adapters/__init__.py
+++ b/src/finetuning_scheduler/strategy_adapters/__init__.py
@@ -12,6 +12,49 @@
 r"""
 Fine-Tuning Scheduler Strategy Adapters
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Strategy adapters extend Fine-Tuning Scheduler support for complex or custom training strategies.
+The built-in adapters (:class:`~finetuning_scheduler.strategy_adapters.FSDPStrategyAdapter`,
+:class:`~finetuning_scheduler.strategy_adapters.ModelParallelStrategyAdapter`) handle PyTorch's
+advanced distributed training strategies.
+
+Plugin Support
+**************
+
+.. warning::
+    This is an :ref:`experimental <versioning:API Stability Classifications>` feature which is
+    still in development.
+
+Fine-Tuning Scheduler supports custom strategy adapters via Python entry points. Third-party packages
+can register custom strategy adapters that will be automatically discovered at runtime.
+
+To register a custom strategy adapter, add an entry point in your package's ``pyproject.toml``:
+
+.. code-block:: toml
+
+    [project.entry-points."finetuning_scheduler.strategy_adapters"]
+    my_adapter = "my_package.adapters:MyStrategyAdapter"
+
+The entry point name (``my_adapter`` in this example) will be used to reference the adapter,
+automatically lowercased. Once registered, the adapter can be used by mapping Lightning strategy
+flags to the adapter via the ``custom_strategy_adapters`` parameter. You can use the entry point
+name, a fully qualified class path with colon separator (``module:Class``), or dot separator
+(``module.Class``):
+
+.. code-block:: python
+
+    from finetuning_scheduler import FinetuningScheduler
+
+    # Map strategy flags to adapters using entry point name
+    fts = FinetuningScheduler(
+        custom_strategy_adapters={
+            "single_device": "my_adapter",  # Entry point name
+            "ddp": "my_package.adapters:MyStrategyAdapter",  # Colon-separated
+            "fsdp": "my_package.adapters.MyStrategyAdapter",  # Dot-separated
+        }
+    )
+
+See :ref:`strategy_adapter_entry_points` for complete documentation and examples.
 """
 from finetuning_scheduler.strategy_adapters.base import StrategyAdapter
 from finetuning_scheduler.strategy_adapters.fsdp import FSDPStrategyAdapter
diff --git a/src/finetuning_scheduler/strategy_adapters/base.py b/src/finetuning_scheduler/strategy_adapters/base.py
index 8496183..b8ddd96 100644
--- a/src/finetuning_scheduler/strategy_adapters/base.py
+++ b/src/finetuning_scheduler/strategy_adapters/base.py
@@ -18,7 +18,9 @@
 """
 from functools import partialmethod
 from pprint import pformat as pfmt
-from typing import Callable, Iterable, List, Optional, Tuple, Dict, Union
+from typing import Callable, Iterable, List, Optional, Tuple, Dict, Union, Any
+import logging
+import os
 
 import torch
 from torch.optim.lr_scheduler import ReduceLROnPlateau
@@ -29,6 +31,8 @@
 from lightning.pytorch.strategies.strategy import Strategy
 from lightning.pytorch.utilities.rank_zero import rank_zero_debug
 
+log = logging.getLogger(__name__)
+
 
 class StrategyAdapter:
     r"""Base class for all strategy adapters. Implements the default
@@ -37,11 +41,6 @@ class StrategyAdapter:
     :external+pl:class:`~lightning.pytorch.strategies.Strategy` via an associated
     :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`.
 
-    .. warning::
-
-        :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter` is in BETA and subject to change. The interface
-        can bring breaking changes and new features with the next release of FTS.
-
     .. tip::
 
         If you want to extend FTS to use a custom, currently unsupported strategy or override current FTS behavior in
@@ -165,6 +164,20 @@ def fts_optim_transform(self, orig_pl: List, inspect_only: bool = False) -> List
         """
         return orig_pl
 
+    def before_restore_model(self, checkpoint: Dict[str, Any]) -> Dict[str, Any]:
+        """Adapter hook executed before model restore.
+
+        Strategy adapters can override this to modify or translate the checkpoint contents (e.g. for state-dict
+        translations) before the model's load path is executed.
+
+        Args:
+            checkpoint (Dict[str, Any]): The full checkpoint dict loaded by the Trainer.
+
+        Returns:
+            Dict[str, Any]: The checkpoint dictionary to be used for restore.
+        """
+        return checkpoint
+
     def logical_param_translation(self, param_names: List) -> List:
         """Effectively the reverse transformation of
         :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.fts_optim_transform`. Can be overridden by a
@@ -380,3 +393,90 @@ def _get_target_bn_modules(self, schedule_phase: int) -> List:
                 isinstance(m, torch.nn.modules.batchnorm._BatchNorm)]
 
     fts_optim_inspect = partialmethod(fts_optim_transform, inspect_only=True)
+
+    def get_named_params_for_schedule_validation(self) -> Dict[str, torch.nn.Parameter]:
+        """Get named parameters for schedule validation.
+
+        This method can be overridden by :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`
+        subclasses to customize parameter iteration for schedule validation (e.g., returning TL-style
+        parameter names instead of canonical names).
+
+        .. note::
+            Strategy adapters can override validation behavior at two levels of abstraction:
+
+            1. **Parameter naming only** (simpler): Override this method to provide custom parameter names
+               while using the default validation logic from
+               :meth:`~finetuning_scheduler.fts_supporters.ScheduleParsingMixin._validate_ft_sched`.
+
+            2. **Full validation logic** (more control): Override
+               :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.validate_ft_sched` to
+               completely customize the validation process.
+
+            Choose the approach that best suits your use case. Most adapters only need to override this
+            method to provide custom parameter names.
+
+        Returns:
+            Dict[str, torch.nn.Parameter]: A dictionary mapping parameter names to parameter tensors.
+                By default, returns the standard ``named_parameters()`` dict.
+        """
+        return dict(self.pl_module.named_parameters())
+
+    def validate_ft_sched(self) -> Tuple[int, int]:
+        """Validate the fine-tuning schedule configuration.
+
+        This method can be overridden by :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`
+        subclasses to customize schedule validation for specific strategies (e.g., strategies that
+        require substantially different validation logic beyond just custom parameter naming).
+
+        .. note::
+            Strategy adapters can override validation behavior at two levels of abstraction:
+
+            1. **Parameter naming only** (simpler): Override
+               :meth:`~finetuning_scheduler.strategy_adapters.StrategyAdapter.get_named_params_for_schedule_validation`
+               to provide custom parameter names while using the default validation logic from
+               :meth:`~finetuning_scheduler.fts_supporters.ScheduleParsingMixin._validate_ft_sched`.
+
+            2. **Full validation logic** (more control): Override this method to completely customize
+               the validation process.
+
+            Choose the approach that best suits your use case. Most adapters only need to override
+            :meth:`get_named_params_for_schedule_validation` to provide custom parameter names.
+
+        Returns:
+            Tuple[int, int]: A tuple of ints specifying:
+                1. The depth of the final scheduled phase
+                2. The maximum epoch watermark explicitly specified in the schedule
+        """
+        # Import here to avoid circular dependency
+        from finetuning_scheduler.fts_supporters import ScheduleParsingMixin
+        from typing import cast
+
+        rank_zero_debug(
+            f"[base StrategyAdapter.validate_ft_sched] Validating schedule for "
+            f"{self.pl_module.__class__.__name__}"
+        )
+        # Delegate to the mixin's implementation by default.
+        return ScheduleParsingMixin._validate_ft_sched(cast(ScheduleParsingMixin, self.fts_handle))
+
+    def gen_ft_schedule(self, dump_loc: Union[str, os.PathLike]) -> Optional[os.PathLike]:
+        """Generate the default fine-tuning schedule using a naive, 2-parameters per-level heuristic.
+
+        This method can be overridden by :class:`~finetuning_scheduler.strategy_adapters.StrategyAdapter`
+        subclasses to customize schedule generation for specific strategies (e.g., using strategy-specific
+        parameter naming conventions).
+
+        Args:
+            dump_loc: The directory to which the generated schedule (.yaml) should be written
+
+        Returns:
+            os.PathLike: The path to the generated schedule, by default ``Trainer.log_dir`` and named after the
+            :external+pl:class:`~lightning.pytorch.core.module.LightningModule` subclass in use with the suffix
+            ``_ft_schedule.yaml``)
+        """
+        # Import here to avoid circular dependency
+        from finetuning_scheduler.fts_supporters import ScheduleImplMixin
+
+        rank_zero_debug(
+            f"[base StrategyAdapter.gen_ft_schedule] Generating schedule for {self.pl_module.__class__.__name__}"
+        )
+        return ScheduleImplMixin._gen_ft_schedule_impl(self.pl_module, dump_loc)
diff --git a/src/finetuning_scheduler/strategy_adapters/model_parallel.py b/src/finetuning_scheduler/strategy_adapters/model_parallel.py
index 0b5efb2..a9598c3 100644
--- a/src/finetuning_scheduler/strategy_adapters/model_parallel.py
+++ b/src/finetuning_scheduler/strategy_adapters/model_parallel.py
@@ -80,10 +80,6 @@ class ModelParallelStrategyAdapter(StrategyAdapter):
 
     See the :ref:`model-parallel-fine-tuning-examples` tutorial for a concrete example and additional guidance.
 
-    .. warning::
-        :class:`~finetuning_scheduler.strategy_adapters.ModelParallelStrategyAdapter` is in BETA and subject to change.
-        The interface can bring breaking changes and new features with the next release of PyTorch.
-
     .. note::
         ``fsdp_plan`` module name/pattern-based ``fully_shard`` directives are applied after any preceding Tensor
         Parallel or explicit ``fully_shard`` directives in ``LightningModule.configure_model``. FTS will only apply
diff --git a/src/fts_examples/profiling/memprofiler.py b/src/fts_examples/profiling/memprofiler.py
index 8ce1866..2e4e722 100644
--- a/src/fts_examples/profiling/memprofiler.py
+++ b/src/fts_examples/profiling/memprofiler.py
@@ -49,10 +49,6 @@ class MemProfiler:
     - `cuda memory snapshot and allocator history tracking <https://pytorch.org/docs/stable/torch_cuda_memory.html>`_
     - host-level memory tracking
     - custom memory hooks (e.g. for activation checkpoint memory tracking via ``saved_tensors_hooks`` etc.)
-
-    .. warning::
-        :class:`~fts_examples.profiling.memprofiler.MemProfiler` is in BETA and subject to change.
-        The interface can bring breaking changes and new features with the next release of Finetuning Scheduler.
     """
     def __init__(self, *args, **kwargs) -> None:
         """The MemProfiler is a powerful memory profiling utility that synthesizes numerous complementary profiling
diff --git a/tests/helpers/expected_warns.py b/tests/helpers/expected_warns.py
index 4d68190..d4666fa 100644
--- a/tests/helpers/expected_warns.py
+++ b/tests/helpers/expected_warns.py
@@ -22,7 +22,7 @@
     "`max_epochs` was not",
     "The dirpath has changed from",
     "unless they are explicitly allowlisted", # required for oldest pytorch (2.5.0) with Lightning 2.5.6
-    "Conversion of an array with ndim > 0",  # still needed with python 3.9 and torch 2.4.0
+    "Conversion of an array with ndim > 0",
     "Please use the new API settings to control TF32 behavior",  # TODO: temporarily required with 20250811 nightly
     "treespec, LeafSpec",  # TODO: required temporarily while lightning uses deprecated PT pytree API
     "torch.jit.script",  # TODO: required temporarily with PT 2.10 nightly 20251124 due to upstream import
diff --git a/tests/helpers/fake_adapter.py b/tests/helpers/fake_adapter.py
new file mode 100644
index 0000000..aca24ce
--- /dev/null
+++ b/tests/helpers/fake_adapter.py
@@ -0,0 +1,6 @@
+class FakeAdapter:
+    """A simple fake Strategy Adapter class for entry point discovery tests.
+
+    Exists solely so entry-points pointing into the package can be imported in unit tests.
+    """
+    pass
diff --git a/tests/test_finetuning_scheduler_callback.py b/tests/test_finetuning_scheduler_callback.py
index 4435cca..0938bac 100644
--- a/tests/test_finetuning_scheduler_callback.py
+++ b/tests/test_finetuning_scheduler_callback.py
@@ -327,7 +327,7 @@ def setup(self, trainer, pl_module, stage: Optional[str] = None) -> None:
         if self.mock_strategy:
             trainer._accelerator_connector._strategy_flag = MOCK_STRATEGY_MAPPING[self.mock_strategy][0]
             self.allow_untested = MOCK_STRATEGY_MAPPING[self.mock_strategy][1]
-            self.custom_strategy_adapter = MOCK_STRATEGY_MAPPING[self.mock_strategy][2]
+            self.custom_strategy_adapters = MOCK_STRATEGY_MAPPING[self.mock_strategy][2]
         super().setup(trainer, pl_module, stage)
         if self.mock_strategy and self.allow_untested:
             raise SystemExit(0)
@@ -1111,6 +1111,28 @@ def test_fts_gen_ft_schedule(tmpdir, model: "LightningModule", dist_mode: bool,
         assert test_schedule[1]["params"] == expected[1]
         assert test_schedule[next(reversed(list(test_schedule.keys())))]["params"] == expected[2]
 
+
+def test_fts_gen_ft_schedule_deprecation_warning(tmpdir):
+    """Validate that direct calls to ScheduleImplMixin.gen_ft_schedule() issue a deprecation warning."""
+    from finetuning_scheduler.fts_supporters import ScheduleImplMixin
+
+    model = FinetuningSchedulerBoringModel()
+
+    # Test that the deprecation warning is issued
+    with pytest.warns(
+        UserWarning,
+        match=r"Direct calls to ScheduleImplMixin\.gen_ft_schedule\(\) are deprecated since v2\.10\.0",
+    ):
+        schedule_path = ScheduleImplMixin.gen_ft_schedule(model, tmpdir)
+
+    # Verify the schedule was still generated correctly
+    assert schedule_path is not None
+    assert os.path.isfile(schedule_path)
+    with open(schedule_path) as f:
+        test_schedule = yaml.safe_load(f.read())
+    assert isinstance(test_schedule, Dict)
+    assert len(test_schedule) > 0
+
 @pytest.mark.skipif(not _MLFLOW_AVAILABLE, reason="test requires MLflow")
 @pytest.mark.parametrize("use_fts_log_dir", [True, False], ids=["fts_log_dir", "no_fts_log_dir"])
 def test_fts_log_dir(tmpdir, use_fts_log_dir):
@@ -2028,6 +2050,43 @@ def test_fts_unallowed_key_error():
         test_fts.restore_best_ckpt()
 
 
+def test_fts_strategy_adapter_restore_exception():
+    """Test that exceptions raised by strategy adapter's before_restore_model hook are properly caught and
+    warned."""
+    test_fts = FinetuningScheduler()
+    test_fts.pl_module, test_fts.trainer = mock.MagicMock(), mock.MagicMock()
+
+    # Mock the checkpoint connector and its loaded checkpoint
+    mock_checkpoint = {"state_dict": {}, "optimizer_states": [{}]}
+    test_fts.trainer._checkpoint_connector._loaded_checkpoint = mock_checkpoint
+    test_fts.trainer._checkpoint_connector.resume_start = mock.MagicMock()
+    test_fts.trainer._checkpoint_connector.restore_datamodule = mock.MagicMock()
+    test_fts.trainer._checkpoint_connector.restore_model = mock.MagicMock()
+    test_fts.trainer._checkpoint_connector.resume_end = mock.MagicMock()
+    test_fts.trainer.strategy.barrier = mock.MagicMock()
+    test_fts.trainer.checkpoint_callback = mock.MagicMock()
+    test_fts.trainer.checkpoint_callback.best_model_path = "/mock/path/checkpoint.ckpt"
+    test_fts.trainer.optimizers = [mock.MagicMock()]
+    test_fts._fts_state._fts_ckpt_metadata = {"best_ckpt_pgs": [[]]}
+
+    # Mock the strategy adapter to raise an exception in before_restore_model
+    mock_adapter = mock.MagicMock()
+    test_error_message = "Mock adapter transformation error"
+    mock_adapter.before_restore_model = mock.MagicMock(side_effect=RuntimeError(test_error_message))
+    mock_adapter.using_sharded_optimizer = False
+    test_fts.strategy_adapter = mock_adapter
+
+    # Mock _restore_training_state to prevent further execution
+    test_fts._restore_training_state = mock.MagicMock()
+
+    # Execute and verify warning is issued
+    with pytest.warns(UserWarning, match=f"Strategy adapter before_restore_model hook raised.*{test_error_message}"):
+        test_fts.restore_best_ckpt()
+
+    # Verify that restore_model was still called despite the exception
+    test_fts.trainer._checkpoint_connector.restore_model.assert_called_once()
+
+
 @pytest.mark.parametrize(
     "explicit_mode, lam_mode, w_expected",
     [
diff --git a/tests/test_strategy_adapter_discovery.py b/tests/test_strategy_adapter_discovery.py
new file mode 100644
index 0000000..5fab1ae
--- /dev/null
+++ b/tests/test_strategy_adapter_discovery.py
@@ -0,0 +1,165 @@
+from importlib.metadata import EntryPoint
+from unittest.mock import MagicMock
+import pytest
+
+from finetuning_scheduler import fts_supporters
+from finetuning_scheduler.fts import FinetuningScheduler
+from lightning.pytorch.utilities.exceptions import MisconfigurationException
+
+
+def test_discover_strategy_adapters(monkeypatch):
+    # Create fake entrypoints that point to a plugin adapter using two common formats
+    # Use a local fake adapter class for discovery tests to avoid importing external packages
+    ep_colon = EntryPoint(
+        name="fake_adapter_colon",
+        value="tests.helpers.fake_adapter:FakeAdapter",
+        group="finetuning_scheduler.strategy_adapters",
+    )
+    ep_dot = EntryPoint(
+        name="fake_adapter_dot",
+        value="tests.helpers.fake_adapter.FakeAdapter",
+        group="finetuning_scheduler.strategy_adapters",
+    )
+
+    def fake_entry_points(group=None):
+        if group == "finetuning_scheduler.strategy_adapters":
+            return [ep_colon, ep_dot]
+        return []
+
+    # Override the entry_points function in the fts_supporters module.
+    # Monkeypatch the importlib.metadata.entry_points function so our fake entrypoints are used by discovery
+    monkeypatch.setattr("importlib.metadata.entry_points", fake_entry_points)
+    # call discovery explicitly (no reload, since entry_points function is patched)
+    # call the function again - it will load entrypoints and register
+    fts_supporters._discover_strategy_adapters()
+    assert "fake_adapter_colon" in fts_supporters.STRATEGY_ADAPTERS
+    assert "fake_adapter_dot" in fts_supporters.STRATEGY_ADAPTERS
+
+
+def test_resolve_strategy_adapter_by_qualname():
+    adapter_map = {"single_device": "tests.helpers.fake_adapter:FakeAdapter"}
+    fts = FinetuningScheduler()
+    cls = fts._resolve_strategy_adapter("single_device", adapter_map)
+    assert cls.__name__ == "FakeAdapter"
+
+
+def test_resolve_strategy_adapter_by_dot_form():
+    adapter_map = {"single_device": "tests.helpers.fake_adapter.FakeAdapter"}
+    fts = FinetuningScheduler()
+    cls = fts._resolve_strategy_adapter("single_device", adapter_map)
+    assert cls.__name__ == "FakeAdapter"
+
+
+def test_resolve_strategy_adapter_by_plugin_name(monkeypatch):
+    """Test importing strategy adapter using discovered plugin entry point name."""
+    # register alias in STRATEGY_ADAPTERS to simulate a discovered plugin (monkeypatched so it's reverted)
+    monkeypatch.setitem(
+        fts_supporters.STRATEGY_ADAPTERS, "fakeplugin", fts_supporters.STRATEGY_ADAPTERS.get("fsdp")
+    )
+    adapter_map = {"single_device": "fakeplugin"}
+    fts = FinetuningScheduler()
+    cls = fts._resolve_strategy_adapter("single_device", adapter_map)
+    assert cls in fts_supporters.STRATEGY_ADAPTERS.values()
+
+
+def test_discover_strategy_adapters_ep_load_failure(monkeypatch):
+    """Test that ep.load() failure triggers fallback import and succeeds."""
+    # Create a mock entry point that will fail on ep.load() but succeed with fallback
+    ep = MagicMock()
+    ep.name = "fallback_adapter"
+    ep.value = "tests.helpers.fake_adapter:FakeAdapter"
+    ep.load.side_effect = ImportError("Simulated ep.load() failure")
+
+    def fake_entry_points(group=None):
+        if group == "finetuning_scheduler.strategy_adapters":
+            return [ep]
+        return []
+
+    monkeypatch.setattr("importlib.metadata.entry_points", fake_entry_points)
+
+    # Should succeed via fallback import despite ep.load() failing
+    fts_supporters._discover_strategy_adapters()
+    assert "fallback_adapter" in fts_supporters.STRATEGY_ADAPTERS
+
+
+def test_discover_strategy_adapters_both_imports_fail(monkeypatch):
+    """Test that both ep.load() and fallback import failures are handled gracefully."""
+    # Create a mock entry point with invalid value that will fail both import methods
+    ep = MagicMock()
+    ep.name = "broken_adapter"
+    ep.value = "nonexistent.module:NonexistentClass"
+    ep.load.side_effect = ImportError("Simulated ep.load() failure")
+
+    def fake_entry_points(group=None):
+        if group == "finetuning_scheduler.strategy_adapters":
+            return [ep]
+        return []
+
+    monkeypatch.setattr("importlib.metadata.entry_points", fake_entry_points)
+
+    # Should not raise, just log warnings and continue
+    with pytest.warns(UserWarning, match="Failed to load strategy adapter entry point"):
+        fts_supporters._discover_strategy_adapters()
+
+    # Adapter should not be registered
+    assert "broken_adapter" not in fts_supporters.STRATEGY_ADAPTERS
+
+
+def test_discover_strategy_adapters_no_value_attribute(monkeypatch):
+    """Test handling of entry point with missing value attribute."""
+    # Create a mock entry point without a value attribute
+    ep = MagicMock()
+    ep.name = "no_value_adapter"
+    ep.load.side_effect = AttributeError("No value attribute")
+    # Simulate missing value attribute
+    type(ep).value = property(lambda self: None)
+
+    def fake_entry_points(group=None):
+        if group == "finetuning_scheduler.strategy_adapters":
+            return [ep]
+        return []
+
+    monkeypatch.setattr("importlib.metadata.entry_points", fake_entry_points)
+
+    # Should handle gracefully and log warning
+    with pytest.warns(UserWarning, match="Failed to load strategy adapter entry point"):
+        fts_supporters._discover_strategy_adapters()
+
+    assert "no_value_adapter" not in fts_supporters.STRATEGY_ADAPTERS
+
+
+def test_resolve_strategy_adapter_invalid_format():
+    """Test that invalid adapter name format raises MisconfigurationException."""
+    # Test with a name that has no dots or colons (invalid format)
+    adapter_map = {"single_device": "invalidname"}
+    fts = FinetuningScheduler()
+
+    with pytest.raises(
+        MisconfigurationException,
+        match=r"Invalid adapter name 'invalidname'.*Must be either a registered plugin name.*or a fully qualified",
+    ):
+        fts._resolve_strategy_adapter("single_device", adapter_map)
+
+
+def test_resolve_strategy_adapter_import_error():
+    """Test that import errors are properly handled and re-raised as MisconfigurationException."""
+    adapter_map = {"single_device": "nonexistent.module:NonexistentClass"}
+    fts = FinetuningScheduler()
+
+    with pytest.raises(
+        MisconfigurationException,
+        match=r"Could not import the specified custom strategy adapter class.*nonexistent\.module:NonexistentClass",
+    ):
+        fts._resolve_strategy_adapter("single_device", adapter_map)
+
+
+def test_resolve_strategy_adapter_missing_strategy_key():
+    """Test that missing strategy key in adapter_map raises MisconfigurationException."""
+    adapter_map = {"ddp": "some_adapter"}  # Missing "single_device" key
+    fts = FinetuningScheduler()
+
+    with pytest.raises(
+        MisconfigurationException,
+        match=r"Current strategy name \(single_device\) does not map to a custom strategy adapter",
+    ):
+        fts._resolve_strategy_adapter("single_device", adapter_map)