flashruntime 0.3.0__tar.gz

flashruntime-0.3.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   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.

flashruntime-0.3.0/PKG-INFO

@@ -0,0 +1,365 @@
+Metadata-Version: 2.4
+Name: flashruntime
+Version: 0.3.0
+Summary: Open fault-tolerant distributed ML runtime: strategy planning, job protocol, leases, checkpointing, and recovery across heterogeneous compute.
+Author: Zolli Labs
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/Zolli-Labs/flashruntime
+Project-URL: Documentation, https://zolli-labs.github.io/flashruntime/
+Project-URL: Repository, https://github.com/Zolli-Labs/flashruntime
+Project-URL: Issues, https://github.com/Zolli-Labs/flashruntime/issues
+Project-URL: Changelog, https://github.com/Zolli-Labs/flashruntime/blob/main/CHANGELOG.md
+Keywords: machine-learning,distributed-training,fault-tolerance,checkpointing,pytorch,ray,mlops,reliability,hyperparameter-search
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: System :: Distributed Computing
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pydantic>=2.7
+Provides-Extra: sklearn
+Requires-Dist: numpy>=1.26; extra == "sklearn"
+Requires-Dist: scikit-learn>=1.3; extra == "sklearn"
+Provides-Extra: k8s
+Requires-Dist: kubernetes>=29; extra == "k8s"
+Provides-Extra: artifacts
+Requires-Dist: minio>=7.2; extra == "artifacts"
+Provides-Extra: oss
+Requires-Dist: oss2>=2.18; extra == "oss"
+Provides-Extra: monitor
+Requires-Dist: psutil>=5.9; extra == "monitor"
+Provides-Extra: service
+Requires-Dist: fastapi>=0.111; extra == "service"
+Requires-Dist: uvicorn>=0.30; extra == "service"
+Requires-Dist: httpx>=0.27; extra == "service"
+Requires-Dist: kubernetes>=29; extra == "service"
+Requires-Dist: minio>=7.2; extra == "service"
+Requires-Dist: pyyaml>=6; extra == "service"
+Provides-Extra: dev
+Requires-Dist: pytest; extra == "dev"
+Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
+Requires-Dist: numpy>=1.26; extra == "dev"
+Requires-Dist: pyyaml>=6; extra == "dev"
+Requires-Dist: markdown>=3.6; extra == "dev"
+Dynamic: license-file
+
+# FlashRuntime
+
+> **The open fault-tolerant distributed ML runtime.** FlashRuntime *operates*
+> your training job — it never rewrites it. You keep the model, the training
+> loop, the loss, the data, and the framework you already have. FlashRuntime
+> wraps the reliability layer around them: it launches your command, injects
+> the environment it promised, tracks your metrics, validates your checkpoints,
+> **recovers from crashes in one call**, and collects your artifacts. The
+> distributed computation itself is always done by established libraries
+> (PyTorch DDP/FSDP2/torchrun, Ray, Hugging Face, scikit-learn).
+
+FlashRuntime is one of three components in the FlashML system by
+[Zolli Labs](https://github.com/Zolli-Labs):
+
+- **[flashnode](https://github.com/Zolli-Labs/flashnode)** — open host agent
+  installed by resource contributors.
+- **flashruntime** (this repo) — the open workload protocol and execution
+  layer. Self-hostable: useful without the cloud.
+- **flashml-cloud** (private) — the managed control plane, marketplace, and
+  dashboard.
+
+Read [`docs/SYSTEM_OVERVIEW.md`](docs/SYSTEM_OVERVIEW.md) for the full product
+architecture, and [`AGENTS.md`](AGENTS.md) if you are an AI coding agent working
+in this repo.
+
+## Install
+
+The core is deliberately tiny — `pip install flashruntime` brings **only
+pydantic**. Every core module (the `flash.submit()` SDK, planner, leases,
+checkpoints, recovery) works with zero infrastructure:
+
+```bash
+pip install flashruntime
+```
+
+Torch is **not** a dependency. FlashRuntime *launches* PyTorch; it never imports
+it. Install it yourself to run the DDP example — a CPU-only build is enough,
+since DDP works over the `gloo` backend with no GPU:
+
+```bash
+pip install torch          # CPU build is fine
+```
+
+## 60-second run
+
+`flash.submit()` operates any command. The one convention your script owes
+FlashRuntime is to write a `metrics.json` (a flat JSON object) into its working
+directory; FlashRuntime collects it and records it as a trial:
+
+```python
+import flashruntime as flash
+
+run = flash.submit(
+    flash.CommandWorkload(
+        command="python train.py --epochs 5",
+        source=flash.Source(path="~/my-project"),
+        outputs=flash.OutputSpec(collect=["metrics.json"]),
+    ),
+    max_restarts=2,   # a crashed attempt relaunches from the last VALID
+                      # checkpoint — up to twice (see below)
+    watch=True,       # opens the live run page and prints its URL
+)
+
+print(run.state.value)   # "SUCCEEDED" (or "FAILED")
+print(run.trials)        # [{'accuracy': 0.91, ...}]  — parsed from metrics.json
+print(run.artifacts)     # [PosixPath('.../metrics.json'), ...]
+print(run.viewer_url)    # http://127.0.0.1:<port> — the live page
+```
+
+`command` is `shlex`-split (there is no shell — for a pipe, pass
+`command="bash -c '...'"`), and `source` is a `flash.Source`, so `~` is expanded
+for you. The same thing from the shell:
+
+```bash
+flashruntime submit "python train.py --epochs 5" \
+    --source ~/my-project --max-restarts 2 --watch
+```
+
+A script that already reads its hyperparameters from `argparse` and writes a
+small JSON of results needs **zero FlashRuntime imports** to be operated. The
+contract at the boundary is thin on purpose: arguments in, `metrics.json` out.
+
+## One import, fault tolerant
+
+`max_restarts=N` is the automatic recovery budget. On a **FAILED** attempt
+FlashRuntime turns the exit into failure signals, classifies them
+(`recovery.classify`), and consults a *versioned, deterministic* policy
+(`recovery.decide`): a deterministic application bug fails fast — a retry would
+only re-hit it — while a transient failure relaunches the same spec from the
+job-scoped checkpoint. Same failure + same policy version ⇒ same action, every
+time. **No LLM in the loop.** Every decision is recorded on the run as
+`FAILURE_CLASSIFIED` / `RECOVERY_ACTION_SELECTED` events.
+
+Kill-and-resume is proven end to end. `tests/test_examples_e2e.py` kills a
+2-process DDP run mid-training and asserts the resumed run's final loss
+matches a never-crashed baseline to 1e-6 — recovery restores only a verified,
+topology-compatible checkpoint (parts-first / manifest-last: a half-written
+checkpoint can never look valid).
+
+For a script you *are* willing to touch, `import flashruntime.torch as ft` adds
+`ft.prepare` / `ft.checkpoint` / `ft.log_metrics`: torch's own DDP wrapped under
+the same checkpoint contract, with correct per-rank device placement, so a
+killed run resumes with a loss curve indistinguishable from the uninterrupted
+one (1e-6 in the e2e). It is optional sugar on the launch-only
+contract, never required.
+
+**GPU-validated (2026-07-23).** The CUDA paths — `nccl` DDP, per-rank device
+placement, and a GPU kill-and-resume — are validated end to end on real
+hardware: **2×NVIDIA GeForce RTX 4090** (RunPod community cloud), torch
+2.7.1+cu128, CUDA 12.8. Two runs, **$0.0725 total**. See `tests/test_gpu_e2e.py`
+and the harness `scripts/runpod_gpu_e2e.py`.
+
+## Watch it run
+
+Pass `watch=True` (the default at an interactive terminal, off in pipes/CI) and
+`flash.submit()` opens a live run page in your browser and prints its URL. It
+draws the run's topology, its loss curve, its verified checkpoints, and every
+recovery decision, refreshing every couple of seconds — served entirely from a
+loopback server with **no external assets**, so it renders with the network cut.
+These docs are served from that same viewer at `/docs`.
+
+> _screenshot: run any `flashruntime submit … --watch` to see the live page._
+
+## Documentation
+
+A PyTorch-style docs site is built from [`docs/site/`](docs/site/) by
+`scripts/build_docs.py` (every byte inline — no CDN, web font, or remote image).
+It is served offline by the viewer at `/docs`, and deploys to GitHub Pages on
+release:
+
+- **<https://zolli-labs.github.io/flashruntime/>** — _live after the first
+  Pages deploy._
+
+Start with [Get started](docs/site/get-started.md) (install → first job → first
+DDP run on CPU). The strategy planner's code walkthrough lives in
+[`docs/planner/README.md`](docs/planner/README.md).
+
+## Benchmarks
+
+An honest benchmark suite (`python -m benchmarks run`) records every result with
+the host that produced it and reports missing comparators as skipped rows, never
+faked. A measured baseline is committed under
+[`benchmarks/results/`](benchmarks/results/). On **Apple M4 (CPU)**:
+
+| Scenario | Measured | Read it honestly |
+|---|---|---|
+| Launch overhead | **+0.04 s** | `flash.submit` vs bare `torchrun`, paired. A one-time fixed cost, not per-step. |
+| Checkpoint cost | **−1.2 ms / checkpoint** (p90 6.2 ms) | A few-KB state dict; the median delta is *below* the run-to-run noise floor at this size (it goes slightly negative), so checkpoints are effectively free here — the p90 bounds it. A larger model surfaces a real positive cost. |
+| Auto-resume | **40 steps not recomputed** | The size-*independent* guarantee: resume never re-does work past the last valid checkpoint. |
+| Adoption | **7 lines** | To make a vanilla script framework-ready (difflib vs each project's own docs). |
+| Resilience | **5/5 faults handled · 20/20 integrity under `kill -9`** | Every fault type routed to the right typed recovery; a checkpoint survives a mid-write `SIGKILL` on 20/20 kills (naive `torch.save`: 20/20 corrupted). 16-trial storm, half crash-armed: 16/16 completed, all 8 crashed trials auto-resumed, 0 manual. Dead-worker MTTD **3.05 s**, MTTR **~3.5 ms**. |
+
+The *timing* rows are smoke-scale on one host: the models are tiny, so the
+*wall-clock* saved by recovery is at the noise floor here. The figure that
+scales is `steps_not_recomputed`, not seconds — the seconds-saved grows with the
+compute between the last checkpoint and the crash (negligible at smoke size,
+hours on a real job). The resilience counts are exact, not smoke-diluted: they
+are *counted* from real fault injection (`kill -9` in the checkpoint-write
+window, a storm of crash-armed trials, a killed lease worker), and the full
+Resilience table with per-scenario method lives in
+[the benchmarks page](docs/site/benchmarks.md). Full provenance and caveats live
+in the result file.
+
+---
+
+## Also: plan a job (no cluster required)
+
+Before you launch, `flash.plan()` turns model + hardware + budget + deadline
+into a ranked, explained execution plan — deterministic, framework-import-free:
+
+```python
+import flashruntime as flash
+
+report = flash.plan(flash.PlanRequest(
+    workload=flash.TransformerFineTune(model="Qwen/Qwen2.5-7B", method="lora",
+                                       train_tokens_m=25),
+    resources=flash.Resources(gpus=4, gpu_type="RTX4090",
+                              hourly_cost_usd_per_gpu=0.44),
+    objective=flash.Objective(mode="balanced", max_cost_usd=20,
+                              deadline_minutes=240),
+))
+print(flash.render(report))
+# → SELECTED: ddp + lora(r=16), 4 workers, torchrun, 21.6 GB/GPU, ~96 min,
+#   $2.82 — plus every rejected candidate with its arithmetic
+```
+
+The report names the distributed method, the libraries it is built from
+(torchrun, DDP/FSDP2, transformers+peft, bitsandbytes, PyTorch DCP) and their
+roles, the knobs, and *why* — including why the alternatives lost. Details:
+[`docs/planner/README.md`](docs/planner/README.md).
+
+## Also: leases, checkpoints, recovery
+
+The reliability core is an embeddable library, not a service you must run:
+
+```python
+from flashruntime.leases import LeaseManager
+from flashruntime.recovery import FailureSignals, classify, decide
+from flashruntime.protocol.v1alpha1 import TaskSpec
+
+mgr = LeaseManager()
+mgr.add_task(TaskSpec(task_id="trial-01", job_id="sweep", commit_key="sweep/trial-01"))
+lease = mgr.claim(node_id="laptop-1")            # pull, never push
+mgr.heartbeat(lease.lease_id)                    # prove liveness
+mgr.complete(lease.lease_id, output_sha256="…")  # first valid commit wins
+
+decision = decide(classify(FailureSignals(heartbeat_lost=True)),
+                  mode="independent_tasks")      # → RETRY_TASK, cordon node
+```
+
+The FlashRuntime service exposes these same objects over HTTP; FlashNode's
+device executor is their remote client. The lease coordinator's durable
+`SqliteLeaseStore` means in-flight leases survive a coordinator restart.
+
+## Testing
+
+```bash
+pytest                  # unit tests — pure Python, no infrastructure
+pytest -m integration   # opt-in: Docker / Kubernetes / MinIO environments
+```
+
+Integration tests live in [`tests/integration/`](tests/integration/README.md)
+and skip themselves (with instructions) when their environment is absent. The
+one GPU test (`tests/test_gpu_e2e.py`) is CUDA-gated and skips without a GPU. The
+local kind + KubeRay + MinIO stack is owned by the workspace Makefile, not this
+repo — the library stays `pip install`-clean.
+
+## Package layout
+
+**Pure-Python core** — `pip install flashruntime` brings pydantic and nothing
+else; every module below works with zero infrastructure:
+
+```
+flashruntime/
+├── protocol/    # versioned public schemas (v1alpha1: JobSpec, Event, Lease,
+│                #   TaskAttempt, CheckpointManifest, RecoveryDecision, plans)
+├── sdk.py       # flash.submit(CommandWorkload) → Run: compile→launch→wait→
+│                #   collect, with the automatic max_restarts recovery loop
+├── workloads/   # CommandWorkload: framework-neutral "run this command"
+├── integrations/# one-file adapters (sklearn / pytorch DDP / HF) — no framework
+│                #   imports at module level; convention is CLI-flags-in/JSON-out
+├── torch/       # optional in-script helper: ft.prepare / checkpoint / log_metrics
+├── planner/     # strategy planner: estimators, curated menus, explained selector
+├── leases/      # Mode A core: claim / heartbeat / expiry sweep / idempotent
+│                #   commit; InMemory + SQLite stores (restart-durable)
+├── checkpoint/  # manifest catalog: parts-first/manifest-last, validation ladder
+├── recovery/    # failure-signal classifier + versioned deterministic policy
+└── viewer/      # stdlib live run page + the built docs site (zero external assets)
+```
+
+**Infrastructure integrations** (opt-in extras, never core imports):
+
+```
+├── service/     # [service] the coordinator: job submission + task expansion,
+│                #   lease/node/checkpoint HTTP endpoints, local artifact hosting,
+│                #   join codes, SQLite ledger, dashboard at GET /, the CLI
+├── launchers/   # LocalProcessLauncher (the concrete local launcher today)
+├── strategies/  # CommandWorkload → LaunchSpec compilation
+├── recipes/     # coordinator-side WorkloadRecipe registry (command jobs)
+├── backends/    # [k8s] ExecutionBackend contract + KubeRay backend (Mode B)
+└── artifacts/   # [artifacts]/[oss] MinIO/S3-compatible + Alibaba OSS stores
+
+flashml_workloads/   # runnable task modules (pure stdlib/sklearn):
+│                    #   sklearn_trial, kmeans_shard + kmeans_driver,
+│                    #   sgd_trainer (checkpointable, bit-identical resume)
+```
+
+Still to come (each lands with its vertical slice, no empty scaffolds):
+multi-node DDP rendezvous (`nnodes > 1` raises `NotImplementedError` today);
+FlashNode's argv runner so service-side command jobs execute remotely;
+checkpoint-manifest persistence (catalog is in-memory; checkpoint *files* are
+durable); Stage-8 ledger metrics (MTTD/MTTR/goodput); `flash.run(plan)` wiring;
+the cloud stage (Postgres, SSE, HF Trainer + PEFT LoRA recipes).
+
+Four orthogonal axes keep the architecture honest: **providers** get machines,
+**launchers** start processes, **strategies** configure execution, **recipes**
+integrate user code. Hugging Face lives in recipes — it is the workload layer,
+not an execution backend. The planner never imports framework code; it emits a
+backend-neutral `StrategyPlan` that strategy compilers translate.
+
+## What FlashRuntime builds on (and what it owns)
+
+| Layer | Libraries reused | What FlashRuntime adds |
+|---|---|---|
+| ML math & models | PyTorch, HF Transformers + PEFT, bitsandbytes, scikit-learn | Nothing — untouched, operated by recipes |
+| Distribution strategies | DDP, FSDP2 (`fully_shard`); DeepSpeed ZeRO later | The *choice* and its explanation, compiled from a StrategyPlan |
+| Launching | torchrun / Torch Elastic; Ray Core on clusters | Checkpoint-aware restart around torchrun; the **lease/heartbeat protocol** for machines outside any cluster — the layer with no existing library |
+| Checkpointing | PyTorch Distributed Checkpoint (parallel save/load, resharding) | The manifest **catalog**: written-last commit, validation status, compatibility, recovery-time selection |
+| Infrastructure | Kubernetes + KubeRay today; SkyPilot / Slurm / provider APIs later | Provider adapters + one cross-environment job model |
+
+Deliberately not built on: HF Accelerate (overlaps torchrun and strategy
+config — user scripts that use it still work via the launch-only contract) and
+Ray Train (mid V1→V2 migration).
+
+## The dependency rule
+
+`flashruntime` owns the public protocol and imports **neither** application
+repo. `flashnode` and `flashml-cloud` both depend on `flashruntime`; they never
+import each other.
+
+## Contributing, security, changelog
+
+- [`CONTRIBUTING.md`](CONTRIBUTING.md) — dev setup, the test/doc/audit gates, and
+  the Developer Certificate of Origin (`git commit -s`).
+- [`SECURITY.md`](SECURITY.md) — how to report a vulnerability.
+- [`CHANGELOG.md`](CHANGELOG.md) — release notes ([Keep a Changelog](https://keepachangelog.com/)).
+
+## License
+
+[Apache-2.0](LICENSE). Contributions via Developer Certificate of Origin
+(`git commit -s`).