deepseek-ai / deepinfer
Code health
Explore this codebase
- Code health map326files scored
- Ownership316sole-owned files: one author, no second reader
- Architecture8modules, 1 entry points
- Decisions4recorded, of which 0 accepted and 4 awaiting review
- Refactoring54open opportunities, one per file
- TestsNo coverage report yet. Run your tests with coverage on, then add the report:
AlsoDocsChatFilesKnowledge graphChange couplingCommitsStatsCosts
Do next · 3
3 things worth doing in the week to Oct 21, the last indexed commit.
Worth planning
Break up _fused_norm_rope_hadamard_cast_store_kvcache (CCN 133, 359 lines)
deepinfer/model_executor/layers/tilelang_ops/cuda/fused_transform.py:612
_fused_norm_rope_hadamard_cast_store_kvcache nests deeply; easier to read and change once split, which also removes a duplicate; changed 1 time in 90 days.
- gain
- up to +2.0 health on this file(inferred)
- finding
- Nested complexity
- size
- CCN 133, 359 lines, nests 7 deep
- severity
- high
Medium effort, worth a checkBreak up _fused_sample (CCN 110, 319 lines)
deepinfer/model_executor/layers/tilelang_ops/cuda/fused_sample.py:326
_fused_sample nests deeply; easier to read and change once split, which also removes a duplicate; changed 1 time in 90 days.
- gain
- up to +0.9 health on this file(inferred)
- finding
- Nested complexity
- size
- CCN 110, 319 lines, nests 7 deep
- severity
- high
Medium effort, worth a checkBreak up _fused_sample (CCN 79, 445 lines)
deepinfer/model_executor/layers/tilelang_ops/asc/fused_sample.py:377
_fused_sample nests deeply; easier to read and change once split, which also removes a duplicate; changed 1 time in 90 days.
- gain
- up to +1.9 health on this file(inferred)
- finding
- Nested complexity
- size
- CCN 79, 445 lines, nests 8 deep
- severity
- high
Medium effort, worth a check
Improve what Repowise can see
Add a test coverage report
Without one, Repowise cannot tell tested code from untested code, so every test-related action says “unknown”.
- line coverage
- Unknown, no report
Small effort
Everything we found
5 areas
Every detector's open count by area. The actions above are the part worth doing.
- Code health550 open findingsMediumGod class · deepinfer/engine/llm_engine.py12 findings · 4.2 deducted
- Refactoring54 open opportunitiesLowscore_run+0.7 health, M effort · eval/score.py
- Decisions4 to reviewMediumBump DeepEP to f6aa38fe4 awaiting review, 0 drifting from the code
- Ownership316 sole-owned filesMediumbuild.shOne author, no second reader
- Dead code1Lowcp_broadcastUnreachable function (6 lines)