Repository navigation
doc(runtime): describe when tesseract-jax hits the VJP cache - #941
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #941 +/- ##
==========================================
+ Coverage 87.29% 87.30% +0.01%
==========================================
Files 57 57
Lines 7855 7855
Branches 1213 1213
==========================================
+ Hits 6857 6858 +1
+ Misses 715 714 -1
Partials 283 283 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Benchmark ResultsBenchmarks use a no-op Tesseract to measure pure framework overhead. 🚀 0 faster, ✅ No significant performance changes detected. Full results
Ran 3 baseline and 3 current passes, alternating between the two. A benchmark is flagged only if every current pass differs from every baseline pass by more than 10% in the same direction. Times are the median over passes of each pass's median. Spread is the range of the per-pass medians on the noisier side, relative to its median.
|
jpbrodrick89
left a comment
There was a problem hiding this comment.
Generally good improvements except for the small nit comment above, thanks @singhharsh1708 !
23b17e1 to
0224ac8
Compare
The
set_jax_vjp_cache_sizedocstring said tesseract-jax goes throughcustom_vjpandfwd, and that the cache hits underjax.jacrev. tesseract-jax binds its own primitive: underjax.gradandjax.value_and_gradit callsapplyand thenvector_jacobian_producton the same inputs, which is why the cache hits. The docstring now names the case where the cache does not help: underjax.jit, JAX dropsapplywhen nothing the jitted function returns depends on its outputs (e.g.jax.gradof a loss linear in them). The comment injax_vjpthat repeated the jacrev claim is corrected too.