Refactor: replace deprecated pl.incore/pl.auto_incore with pl.at syntax

[zhangqi-chen]

Summary

• Replace all pl.incore() with pl.at(level=pl.Level.CORE_GROUP) across examples and docs
• Replace all pl.auto_incore() with pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) across examples and docs
• Remove obsolete qwen3_32b_decode_scope12.py

Related Issues

N/A

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

This PR systematically replaces pl.auto_incore() and pl.incore() context managers with explicit placement directives using pl.at(level=pl.Level.CORE_GROUP, ...) or pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) across documentation and example code, standardizing how execution scopes are specified.

Changes

Cohort / File(s)	Summary
Documentation Updates .claude/rules/coding-style.md, docs/para_for.md, docs/pto2_rt.md, docs/pypto-frontend-coding-style.md	Updated guidance and examples to use explicit pl.at(level=pl.Level.CORE_GROUP, ...) placement instead of pl.auto_incore() and pl.incore() patterns for anonymous InCore scopes.
Beginner Examples examples/beginner/hello_world.py, examples/beginner/matmul.py	Replaced pl.auto_incore() with pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) in loop context managers.
Intermediate Examples examples/intermediate/gemm.py, examples/intermediate/layer_norm.py, examples/intermediate/rms_norm.py, examples/intermediate/rope.py, examples/intermediate/softmax.py	Standardized execution context from pl.auto_incore() to explicit pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) for tiled kernel implementations.
DeepSeek V3.2 Models examples/models/deepseek_v3_2/deepseek_v3_2_decode_back.py, examples/models/deepseek_v3_2/deepseek_v3_2_decode_front.py, examples/models/deepseek_v3_2/deepseek_v3_2_prefill_back.py, examples/models/deepseek_v3_2/deepseek_v3_2_prefill_front.py	Updated execution scopes in decode/prefill layer implementations to use explicit core-group placement with chunked-loop optimization.
Kimi K2 Decode examples/models/kimi/kimi_k2_decode.py	Replaced three pl.auto_incore() regions with explicit pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) around RMSNorm+QKV, flash attention, and output projection scopes.
MILM Decode examples/models/milm/milm_decode.py	Converted three pl.auto_incore() execution contexts to explicit pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) in RMSNorm, attention, and MLP sections.
Qwen3 Basic Decode/Prefill examples/models/qwen3/qwen3-32b.py, examples/models/qwen3/qwen3_32b_prefill.py, examples/models/qwen3/qwen3_32b_training_forward_and_backward.py	Replaced pl.auto_incore() with explicit pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) across layer scopes.
Qwen3 Scope-based Implementations examples/models/qwen3/qwen3_32b_decode.py, examples/models/qwen3/qwen3_32b_decode_mixed.py, examples/models/qwen3/qwen3_32b_decode_scope1.py, examples/models/qwen3/qwen3_32b_decode_scope2.py, examples/models/qwen3/qwen3_32b_decode_scope3.py	Systematically replaced pl.incore() with pl.at(level=pl.Level.CORE_GROUP) and pl.auto_incore() with pl.at(level=pl.Level.CORE_GROUP, optimization=pl.chunked_loop_optimizer) across multiple scopes (RMSNorm, attention stages, MLP projections).
Qwen3 Tilelet & Prefill Variants examples/models/qwen3/qwen3_32b_prefill_tilelet.py	Updated token-tile and output-projection scopes to use explicit pl.at(...) placement with core-group level and chunked-loop optimization.
Qwen3 Scope12 Deletion examples/models/qwen3/qwen3_32b_decode_scope12.py	Removed entire file (638 lines) containing a two-stage transformer attention pipeline implementation with RMSNorm, RoPE, and online-softmax attention logic.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related issues

• [A2A3/A5 E2E] qwen3_32b_prefill_tilelet — Full Prefill Layer Validation #71: Directly modifies the same examples/models/qwen3/qwen3_32b_prefill_tilelet.py scopes and incore context management discussed in the validation issue.

Possibly related PRs

• PR #40: Modifies examples/models/qwen3/qwen3_32b_decode.py with overlapping changes to the same qwen3_decode_layer implementation's execution scopes.
• PR #93: Updates qwen3_32b_decode_scope2.py and qwen3_32b_decode_scope12.py, directly overlapping with this PR's scope2/scope12 modifications.
• PR #66: Widespread changes across multiple Qwen3 example modules and the qwen3_32b_decode_scope12.py file, indicating substantial code-level overlap with scope1/scope12/scope2 additions and incore region updates.

Poem

🐰 Old incore magic fades to light,
Explicit placement shines so bright!
From auto to pl.at() we hop,
With CORE_GROUP optimization,
The code's now crystal clear—top to top! ✨

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 9.09% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the main refactoring objective: replacing deprecated pl.incore/pl.auto_incore with pl.at syntax across the entire changeset.
Description check	✅ Passed	The description clearly outlines the three main changes: replacing pl.incore, replacing pl.auto_incore, and removing an obsolete file, all of which are present in the changeset.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

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.}

[gemini-code-assist]

Code Review

This pull request updates the PyPTO API by replacing deprecated pl.auto_incore() and pl.incore() calls with the new pl.at() scope across documentation and model examples. The transition to with pl.at(level=pl.Level.CORE_GROUP, ...) provides more explicit control over execution levels and enables optimizations like the chunked loop optimizer. These changes are reflected in core documentation and a wide range of model implementations, including DeepSeek v3.2, Kimi K2, MiLM, and Qwen3-32B. Additionally, a redundant Qwen3 decode script was removed. I have no feedback to provide as there are no review comments.