CI: Narrow affected-models detection by resolving __init__.py re-exports

[tadani3]

CI: Narrow affected-models detection by resolving init.py re-exports

Changes

1. src/mobius/tasks/ reclassified from shared_infra → traceable.
   Task files no longer trigger run_all unconditionally; they are now traced
   through the import graph.
2. Re-export resolution for __init__.py hubs. A new _build_reexport_map()
   walks every __init__.py and builds a (package, symbol) → source_module
   map. _parse_imports() resolves from mobius.components import Foo to the
   actual source module (mobius.components._attention) instead of recording a
   dependency on the package itself. The package name is only added when a
   symbol can't be resolved (wildcards, missing entries).

The result: changes to a re-export hub (e.g. components/__init__.py) no
longer invalidate every importer — only models whose actually-used symbol's
source changed are affected.

Before vs. after (Gemma4 + Qwen test scenario)

Same input file list:

Detector	Affected models	run_all
Before	193	false
After	2 (gemma4, gemma4_text)	false

Verification

# Gemma4 + Qwen test scenario
cat <<'EOF' | python scripts/detect_affected_models.py --stdin
src/mobius/components/__init__.py
src/mobius/components/_gemma4_audio.py
src/mobius/models/gemma4.py
EOF
# → {"affected": ["gemma4", "gemma4_text"], "run_all": false}

# Attention component → all causal-LMs / VLMs that use shared Attention
echo "src/mobius/components/_attention.py" | python scripts/detect_affected_models.py --stdin
# → 154 models (vision-only / audio-only models correctly excluded)

# MoE component → only MoE architectures
echo "src/mobius/components/_moe.py" | python scripts/detect_affected_models.py --stdin
# → 33 models (Mixtral, DeepSeek V2/V3, Qwen MoE, Granite MoE, Jamba, ...)

All 49 tests in scripts/detect_affected_models_test.py pass.

Caveat

Tasks aren't directly imported by model files, so a task change currently
produces an empty affected set (no run_all). If you want task changes to
trigger specific models, a task → model_type mapping would need to be added.

[github-actions]

Performance Comparison

Comparing 8be8155 → c28b703

Model	Metric	Baseline	Current	Delta
bert (feature-extraction)	model_size_bytes	359 KB	359 KB	+0.0%	⚪
bert (feature-extraction)	num_nodes	60	60	+0.0%	⚪
falcon	model_size_bytes	364 KB	364 KB	+0.0%	⚪
falcon	num_nodes	66	66	+0.0%	⚪
gemma2	model_size_bytes	428 KB	428 KB	+0.0%	⚪
gemma2	num_nodes	107	107	+0.0%	⚪
gpt2	model_size_bytes	388 KB	388 KB	+0.0%	⚪
gpt2	num_nodes	53	53	+0.0%	⚪
llama	model_size_bytes	425 KB	425 KB	+0.0%	⚪
llama	num_nodes	61	61	+0.0%	⚪
llama (static-cache)	model_size_bytes	425 KB	425 KB	+0.0%	⚪
llama (static-cache)	num_nodes	58	58	+0.0%	⚪
mamba (ssm-text-generation)	model_size_bytes	360 KB	360 KB	+0.0%	⚪
mamba (ssm-text-generation)	num_nodes	98	98	+0.0%	⚪
phi3	model_size_bytes	421 KB	421 KB	+0.0%	⚪
phi3	num_nodes	59	59	+0.0%	⚪
phi3 (static-cache)	model_size_bytes	421 KB	421 KB	+0.0%	⚪
phi3 (static-cache)	num_nodes	56	56	+0.0%	⚪
qwen2	model_size_bytes	425 KB	425 KB	+0.0%	⚪
qwen2	num_nodes	61	61	+0.0%	⚪
qwen2 (static-cache)	model_size_bytes	425 KB	425 KB	+0.0%	⚪
qwen2 (static-cache)	num_nodes	58	58	+0.0%	⚪
qwen3_5_moe (hybrid-text-generation)	model_size_bytes	506 KB	506 KB	+0.0%	⚪
qwen3_5_moe (hybrid-text-generation)	num_nodes	275	275	+0.0%	⚪
qwen3_5_text (hybrid-text-generation)	model_size_bytes	458 KB	458 KB	+0.0%	⚪
qwen3_5_text (hybrid-text-generation)	num_nodes	129	129	+0.0%	⚪
qwen3_5_vl (hybrid-qwen-vl)	model_size_bytes	977 KB	977 KB	+0.0%	⚪
qwen3_5_vl (hybrid-qwen-vl)	num_nodes	408	408	+0.0%	⚪
t5 (seq2seq)	model_size_bytes	836 KB	836 KB	+0.0%	⚪
t5 (seq2seq)	num_nodes	166	166	+0.0%	⚪
whisper (speech-to-text)	model_size_bytes	1008 KB	1008 KB	+0.0%	⚪
whisper (speech-to-text)	num_nodes	128	128	+0.0%	⚪

No performance regressions.

[codecov]

The author of this PR, tadani3, is not an activated member of this organization on Codecov.
Please activate this user on Codecov to display this PR comment.
Coverage data is still being uploaded to Codecov.io for purposes of overall coverage calculations.
Please don't hesitate to email us at support@codecov.io with any questions.

[Copilot]

Pull request overview

This PR tightens CI affected-model detection by improving import-graph precision, especially for packages that act as re-export hubs (e.g., mobius.components), so CI runs model tests for the models actually impacted by a change.

Changes:

• Treat src/mobius/tasks/ as traceable (import-graph traced) instead of always triggering run_all.
• Add __init__.py re-export resolution: from mobius.components import X is attributed to the underlying defining module for X when resolvable.
• Update detector unit tests to reflect the new task classification and run-all behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
scripts/detect_affected_models.py	Implements re-export resolution for from pkg import sym and reclassifies tasks/ as traceable.
scripts/detect_affected_models_test.py	Updates unit tests for the new tasks/ classification and detector behavior.