Remove sharding docs

[smahdavi4]

Summary by CodeRabbit

• Documentation

  • Removed obsolete sharded model conversion section.
  • Updated dataset guides to use the non-sharded DeepSeek-R1 path and revised server arguments to reflect the new setup.

• Chores

  • Updated solution generation configs to use the non-sharded DeepSeek-R1 path.
  • Switched server arguments to an explicit parallelism setting, aligning configs with the updated runtime.
  • Simplifies setup and removes reliance on sharded model artifacts.

[coderabbitai]

Walkthrough

Documentation removes sharded conversion instructions and updates example model paths and server args. Configs switch model paths from pre-sharded “tp16” variants to unsharded paths and replace “--load-format sharded_state” with “--ep-size 16” while preserving context lengths.

Changes

Cohort / File(s)	Summary
Docs: Open Math Reasoning docs/releases/openmathreasoning/dataset.md	Removed the “Model conversion” section describing sharded DeepSeek-R1 state conversion; remaining sections unchanged.
Docs: Open Reasoning docs/releases/openreasoning/dataset.md	Removed optional sharding block; updated example model path from /workspace/DeepSeek-R1-0528-tp16 to /workspace/DeepSeek-R1-0528; changed server_args from --load-format sharded_state --context-length {…} to --ep-size 16 --context-length {…}.
Configs: Open Code Reasoning R1 recipes/opencodereasoning/configs/solution_sdg/r1.yaml	In generate_solutions.stage_kwargs: model path /hf_models/DeepSeek-R1-tp16 → /hf_models/DeepSeek-R1; server_args --load-format sharded_state --context-length 32768 → --ep-size 16 --context-length 32768.
Configs: Open Math Reasoning R1 recipes/openmathreasoning/configs/solution_sdg/r1.yaml	In generate_solutions.stage_kwargs: model path /hf_models/DeepSeek-R1-tp16 → /hf_models/DeepSeek-R1; server_args --load-format sharded_state --context-length 18384 → --ep-size 16 --context-length 18384.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

I nudge my whiskers at config lines,
Hop from “tp16” to simpler signs.
No shards to juggle, neat and clean—
Just ep-size 16 in the scene.
Carrots compiled, docs aligned;
A tidy trail for buns behind. 🥕✨

Pre-merge checks and finishing touches

✅ 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 concisely describes the primary change in the pull request, which is the removal of sharding-related documentation, making it clear to reviewers without extraneous detail.
Docstring Coverage	✅ Passed	No functions found in the changes. Docstring coverage check skipped.

✨ Finishing touches

🧪 Generate unit tests

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch smahdavi/sharding-docs

---

📜 Recent review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between e342ee8 and 4ed7f6c.

📒 Files selected for processing (4)

• docs/releases/openmathreasoning/dataset.md (0 hunks)
• docs/releases/openreasoning/dataset.md (1 hunks)
• recipes/opencodereasoning/configs/solution_sdg/r1.yaml (1 hunks)
• recipes/openmathreasoning/configs/solution_sdg/r1.yaml (1 hunks)

💤 Files with no reviewable changes (1)

• docs/releases/openmathreasoning/dataset.md

⏰ 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). (2)

• GitHub Check: pre-commit
• GitHub Check: unit-tests

🔇 Additional comments (3)

recipes/openmathreasoning/configs/solution_sdg/r1.yaml (1)

53-53: LGTM! Model path and server args updated correctly.

The changes correctly switch from pre-sharded checkpoint loading to unsharded model with explicit parallelism. The --ep-size 16 matches the total GPU count (8 GPUs × 2 nodes), and the context length is properly preserved.

Also applies to: 57-57

recipes/opencodereasoning/configs/solution_sdg/r1.yaml (1)

36-36: LGTM! Consistent model path and server args updates.

The changes follow the same pattern as the math config, correctly switching to unsharded model loading with --ep-size 16 (matching 8 GPUs × 2 nodes). The higher context length of 32768 is appropriate for code generation tasks.

Also applies to: 40-40

docs/releases/openreasoning/dataset.md (1)

69-69: LGTM – docs/releases/openreasoning/dataset.md is free of any -tp16 or sharded_state references.

---

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.}

[Kipok]

thanks!