fix: update model recipe for llama-3 70b to match with common recipe template

[biswapanda]

Overview:

When running the perf job with the specified command, the script fails due to an incorrect repo id format error during the tokenizer creation process. The expected result is for the perf job to finish successfully.

closes: DEP-526

Details:

Where should the reviewer start?

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

• closes GitHub issue: #xxx

Summary by CodeRabbit

• Refactor

  • Switched to a more efficient model download method to improve speed and reliability.
  • Pinned the model version to ensure consistent behavior across deployments.

• Chores

  • Simplified setup by removing the need for a token-based configuration.
  • Standardized the model storage location for easier management and persistence across runs.

[coderabbitai]

Walkthrough

Replaces Hugging Face CLI + HF_TOKEN–based model download with an hf-transfer flow. Introduces HF_HOME, HF_HUB_ENABLE_HF_TRANSFER, and MODEL_REVISION. Pins download to a specific revision. Changes storage path and volume mount to /model-store. Removes secretKeyRef for HF_TOKEN and updates the download command and working directory.

Changes

Cohort / File(s)	Summary
Model cache workflow recipes/llama-3-70b/model-cache/model-download.yaml	Switch from huggingface-cli with HF_TOKEN to hf download using hf-transfer; set HF_HOME=/model-store, enable HF transfer, add MODEL_REVISION pin; change workingDir and volume mount to /model-store; remove HF_TOKEN secret and secretKeyRef.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  participant J as K8s Job/Container
  participant HF as Hugging Face Hub
  participant FS as /model-store (Volume)

  Note over J: Env: HF_HOME=/model-store<br/>HF_HUB_ENABLE_HF_TRANSFER=1<br/>MODEL_REVISION=<pin>
  J->>FS: Set working directory to /model-store
  J->>HF: hf download <model> --revision $MODEL_REVISION
  HF-->>J: Stream artifacts (hf-transfer)
  J->>FS: Write model files to /model-store
  Note over J,FS: No HF_TOKEN secret used

Loading

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

I hopped to the hub with a whisk of my ear,
No secrets in pockets—just paths crystal clear.
A pinned little revision, a speedy transfer cheer,
Models in /model-store, neatly landing here.
Thump-thump: cache is tidy, the carrots are near! 🥕✨

Pre-merge checks

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description Check	⚠️ Warning	The pull request description includes the required section headings but leaves the Details and Where should the reviewer start sections entirely blank with placeholder comments, and the Related Issues section uses a placeholder “#xxx” rather than an actual issue reference. The Overview also misplaces “closes: DEP-526” instead of reserving the Related Issues section for issue closure. These omissions mean the description does not fully adhere to the repository template or provide sufficient context about the changes.	Please fill in the Details section with a clear summary of the actual code and configuration changes, specify which files or sections the reviewer should inspect under “Where should the reviewer start?”, correct the Related Issues entry to reference the actual issue number, and remove or relocate the misplaced “closes: DEP-526” from the Overview.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changes. Docstring coverage check skipped.
Title Check	✅ Passed	The title specifically highlights updating the llama-3 70b model recipe to conform to the common template, which aligns with the primary change of rewriting the model download flow and adjusting environment variables. It is concise, clear, and directly reflects the main purpose of the changeset without unnecessary detail.

---

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.}

[biswapanda]

/ok to test 9f3f7ed

[grahamking]

Could you extend the title from "update model path" to be more specific? Which model path(s)?

[biswapanda]

Could you extend the title from "update model path" to be more specific? Which model path(s)?

Thanks, updated the title