chore: bump Llama Stack version to 0.2.21

[nathan-weinberg]

to align with d/s change

Summary by CodeRabbit

• New Features

  • Added a local filesystem-backed file store for distribution meta-reference data in runtime configuration.

• Chores

  • Upgraded core stack dependency to 0.2.21 across build and tooling.
  • Updated container build to install CPU PyTorch wheels, include torchao (>=0.12), and bump datasets (>=4.0).
  • Refreshed pre-commit hook to align with the new stack version.

[coderabbitai]

Walkthrough

Dependency and build script updates plus a runtime provider addition: llama-stack bumped to 0.2.21 across config/build files; Containerfile switches to PyTorch CPU wheels via an extra-index-url, adds torchao>=0.12.0, and upgrades datasets>=4.0.0; distribution/build.py now treats --extra-index-url like --index-url when composing install commands; distribution/run.yaml adds a local filesystem-backed provider for meta-reference files with a SQLite metadata store.

Changes

Cohort / File(s)	Change Summary
Llama-stack version bump \.pre-commit-config.yaml, distribution/Containerfile.in, distribution/build.py, distribution/Containerfile	Update llama-stack from 0.2.18 to 0.2.21 in multiple build/config files; no API/signature changes.
Container build dependency updates distribution/Containerfile	Switch PyTorch installation to CPU wheels via --extra-index-url (installing torch, torchvision, torchao>=0.12.0); remove previous PyTorch --index-url line; set datasets>=4.0.0; install llama-stack 0.2.21 with --no-cache.
Dependency command composition distribution/build.py	Broadened handling so lines containing --extra-index-url are treated like --index-url when constructing full pip install commands (now forming single-line commands for those cases).
Runtime provider addition distribution/run.yaml	Add new tool_runtime provider group files containing provider meta-reference-files of type inline::localfs with storage_dir: /opt/app-root/src/.llama/distributions/rh/files and sqlite metadata_store at /opt/app-root/src/.llama/distributions/rh/files_metadata.db. Existing registry metadata_store unchanged.

Sequence Diagram(s)

sequenceDiagram
  participant Build as distribution/build.py
  participant ReqFile as requirements / Containerfile deps
  participant Pip as pip installer

  Note over Build,ReqFile: build.py composes install commands
  ReqFile->>Build: provide dependency lines (may include --index-url or --extra-index-url)
  Build->>Pip: emit single-line pip install (treats --extra-index-url like --index-url)
  Pip-->>Build: packages installed (torch, torchvision, torchao, datasets, llama-stack)

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

I nibble versions, stitch the build’s new thread,
Wheels swap shelves, and files find a bed.
SQLite whispers, metadata tucked away,
A rabbit hops—CI munches on the day. 🐇✨

Pre-merge checks (3 passed)

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title Check	✅ Passed	The title "chore: bump Llama Stack version to 0.2.21" is concise and accurately reflects the primary change in the diff (llama-stack version updated in .pre-commit-config.yaml, distribution/build.py, and Containerfile(.in)). Although the PR also contains related build/config updates (PyTorch index change, datasets bump, and a new local file provider), those appear secondary to the stated version bump, so the title appropriately communicates the main intent.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

✨ Finishing touches

• 📝 Generate Docstrings

🧪 Generate unit tests

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

distribution/Containerfile (1)

8-13: Use --index-url (not --extra-index-url) for PyTorch CPU wheels.

This avoids mixed resolution and follows PyTorch install guidance for CPU wheels.

-RUN pip install \
-    --extra-index-url \
-    https://download.pytorch.org/whl/cpu \
-    torch \
-    torchao>=0.12.0 \
-    torchvision
+RUN pip install \
+    --index-url https://download.pytorch.org/whl/cpu \
+    torch \
+    torchao>=0.12.0 \
+    torchvision

Reference install docs show CPU installs via --index-url. (docs.pytorch.org)

📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 4d50ebe and 21fe35b.

📒 Files selected for processing (4)

• .pre-commit-config.yaml (1 hunks)
• distribution/Containerfile (2 hunks)
• distribution/Containerfile.in (1 hunks)
• distribution/build.py (1 hunks)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: build-test-push (linux/amd64)

🔇 Additional comments (5)

.pre-commit-config.yaml (1)

59-59: LLM stack pin updated in local hook — verified.
No remaining llama-stack==0.2.18 occurrences; all pins are 0.2.21 in distribution/Containerfile.in:6, distribution/Containerfile:48, distribution/build.py:16.

distribution/build.py (1)

16-16: Pin bumped to 0.2.21 — aligned with repo.

Matches .pre-commit and Containerfiles; version guard will enforce consistency.

distribution/Containerfile.in (1)

6-6: Container build pin updated — OK.

Keeps the generated image in lockstep with tooling.

distribution/Containerfile (2)

18-18: datasets>=4.0.0 update — fine with Python 3.12 base.

No concerns from the container side.

---

48-48: llama-stack 0.2.21 in image — good.

Pin matches tooling; reproducible build maintained.

[coderabbitai]

Actionable comments posted: 0

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

distribution/build.py (1)

79-96: Bug: sorting tokens breaks --index-url/--extra-index-url flag–argument pairing.

Sorting the tail “packages” fragment reorders --index-url|--extra-index-url and their URL arguments, which can corrupt the install command. Preserve original ordering when these flags are present.

Apply this minimal fix:

-                    packages = sorted(
-                        set(parts[3].split())
-                    )  # Sort the package names and remove duplicates
+                    rest = parts[3].strip()
@@
-                    if ("--index-url" in line) or ("--extra-index-url" in line):
-                        full_cmd = " ".join(cmd_parts + [" ".join(packages)])
+                    if ("--index-url" in line) or ("--extra-index-url" in line):
+                        # Preserve original ordering to keep flag-argument pairs intact
+                        full_cmd = " ".join(cmd_parts + [rest])
                         torch_deps.append(full_cmd)
-                    elif "--no-deps" in line:
-                        full_cmd = " ".join(cmd_parts + [" ".join(packages)])
+                    elif "--no-deps" in line:
+                        full_cmd = " ".join(cmd_parts + [rest])
                         no_deps.append(full_cmd)
-                    elif "--no-cache" in line:
-                        full_cmd = " ".join(cmd_parts + [" ".join(packages)])
+                    elif "--no-cache" in line:
+                        full_cmd = " ".join(cmd_parts + [rest])
                         no_cache.append(full_cmd)
                     else:
-                        formatted_packages = " \\\n    ".join(packages)
+                        packages = sorted(set(rest.split()))
+                        formatted_packages = " \\\n    ".join(packages)
                         full_cmd = f"{' '.join(cmd_parts)} \\\n    {formatted_packages}"
                         standard_deps.append(full_cmd)

Optional hardening (future): tokenize with shlex.split and keep a small set of “option-with-arg” flags to preserve their adjacency.

🧹 Nitpick comments (1)

distribution/run.yaml (1)

109-116: Confirm schema: ‘files’ providers may require listing ‘files’ in top-level apis (or nesting under the correct API).

If “files” is its own API, add it to apis:; if it’s intended under tool_runtime, restructure accordingly. Validate with llama stack validate/build.

Proposed addition (if “files” is an API):

 apis:
   - agents
   - datasetio
   - eval
   - inference
   - safety
   - scoring
   - telemetry
   - tool_runtime
   - vector_io
+  - files

📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between a86a399 and 7763ed7.

📒 Files selected for processing (3)

• distribution/Containerfile (2 hunks)
• distribution/build.py (2 hunks)
• distribution/run.yaml (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

• distribution/Containerfile

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: build-test-push (linux/amd64)

🔇 Additional comments (1)

distribution/build.py (1)

16-17: Approve — bump to llama-stack==0.2.21 applied repo‑wide.
All 'llama-stack==' occurrences point to 0.2.21: .pre-commit-config.yaml:59; distribution/Containerfile.in:6; distribution/Containerfile:43; distribution/build.py:16 (detection at 42).

[Elbehery]

/lgtm