Skip to content

[1/n] Adds skip-softmax calibration through the vLLM serving path - #1992

Open
kaix-nv wants to merge 14 commits into
mainfrom
kaix/vllm_skip_calib_upstream
Open

[1/n] Adds skip-softmax calibration through the vLLM serving path#1992
kaix-nv wants to merge 14 commits into
mainfrom
kaix/vllm_skip_calib_upstream

test: make vLLM calibration cache fixtures portable

b84cc4c
Select commit
Loading
Failed to load commit list.
Codecov / codecov/project succeeded Aug 29, 2026 in 0s

78.59% (-0.37%) compared to 022767c

View this Pull Request on Codecov

78.59% (-0.37%) compared to 022767c

Details

Codecov Report

❌ Patch coverage is 92.05882% with 27 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.59%. Comparing base (022767c) to head (b84cc4c).

Files with missing lines Patch % Lines
...lopt/torch/kernels/sparsity/attention/calibrate.py 69.23% 12 Missing ⚠️
.../torch/sparsity/attention_sparsity/plugins/vllm.py 92.39% 7 Missing ⚠️
...parsity/attention_sparsity/plugins/vllm_runtime.py 92.40% 6 Missing ⚠️
...ention_sparsity/plugins/sparse_attn_calibration.py 97.77% 2 Missing ⚠️
Additional details and impacted files 
@@            Coverage Diff             @@
##             main    #1992      +/-   ##
==========================================
- Coverage   78.95%   78.59%   -0.37%     
==========================================
  Files         524      525       +1     
  Lines       60866    61172     +306     
==========================================
+ Hits        48058    48079      +21     
- Misses      12808    13093     +285

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
View more details on Codecov