feat(zk): add GPU ZK long-run tests and CI workflow by pdroalves · Pull Request #3453 · zama-ai/tfhe-rs

pdroalves · 2026-04-07T01:02:34Z

closes: https://github.com/zama-ai/tfhe-rs-internal/issues/1372

PR content/description

Check-list:

Tests for the changes have been added (for bug fixes / features)
Docs have been added / updated (for bug fixes / features)
Relevant issues are marked as resolved/closed, related issues are linked in the description
Check for breaking changes (including serialization changes) and add them to commit message following the conventional commit specification

andrei-stoian-zama

I see this output:




running 1 test
test gpu::tests::prove_verify_stress::test_pke_v2_gpu_cpu_equivalence_long_run ... test_pke_v2_gpu_cpu_equivalence_long_run: base_seed=0x7ca2d9fd80287902, rounds=3
  round 0/3: seed=0x61bbcfdbee87953a
  round 1/3: seed=0x2a5b618e3fcba7e0
  round 2/3: seed=0xce7587440ec88f6d
ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 54 filtered out; finished in 505.59s

   Doc-tests tfhe_zk_pok

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

Is this ok ? only 3 rounds of one test are run ?

can't we group this with some other workflow ? we're adding more CI costs just for 6 minutes of tests

andrei-stoian-zama · 2026-04-08T08:23:22Z

+    thread_rng().gen()
+}
+
+/// Run one full GPU-vs-CPU equivalence sweep for a given seed.


this comment isn't very clear. the values in the comment may change. the comment should explain what the function does not give the constants used.

andrei-stoian-zama · 2026-04-08T08:51:48Z

+///     they agree.
+#[test]
+fn test_pke_v2_gpu_cpu_equivalence() {
+    let seed: u64 = thread_rng().gen();


why don't we allow to use the user supplied seed here? why print the seed if we can't reproduce with it ?

Which user supplied seed?

The seed print was once requested by @nsarlin-zama. In case some test fails we can easily reproduce the execution path.

if we print the seed we may want to re-use it easily through "TFHE_RS_LONGRUN_TESTS_SEED", why don't we use that here ?

andrei-stoian-zama · 2026-04-08T08:52:08Z

+
+/// Multi-seed long-run variant of the GPU/CPU equivalence test.
+///
+/// Repeats the full sweep (3 CRS variants x 32 invalid-witness flags


the comment is too verbose. no need to explain environment variables again

pdroalves · 2026-04-08T15:16:34Z

@andrei-stoian-zama I removed most of the comments. Check again, please.

pdroalves · 2026-04-08T15:26:39Z

@andrei-stoian-zama I don't know why you got only 3 runs. Here I see 20.

➜  zk-long-tests git:(pa/chore/zk-long-tests) ✗ make test_integer_zk_experimental_long_run_gpu 
RUSTFLAGS="-C target-cpu=native" cargo test --profile release \
	-p tfhe-zk-pok --features experimental,gpu-experimental -- \
	test_pke_v2_gpu_cpu_equivalence_long_run --test-threads=1 --nocapture
   Compiling tfhe-cuda-backend v0.14.0 (/home/zama/Code/worktree/zk-long-tests/backends/tfhe-cuda-backend)
   Compiling tfhe-safe-serialize v0.1.0 (/home/zama/Code/worktree/zk-long-tests/utils/tfhe-safe-serialize)
   Compiling zk-cuda-backend v0.1.0 (/home/zama/Code/worktree/zk-long-tests/backends/zk-cuda-backend)
   Compiling tfhe-zk-pok v0.8.1 (/home/zama/Code/worktree/zk-long-tests/tfhe-zk-pok)
    Finished `release` profile [optimized] target(s) in 1m 31s
     Running unittests src/lib.rs (target/release/deps/tfhe_zk_pok-87f34bb3a8b87d8e)

running 1 test
test gpu::tests::prove_verify_stress::test_pke_v2_gpu_cpu_equivalence_long_run ... test_pke_v2_gpu_cpu_equivalence_long_run: base_seed=0x42140705ecf06134, rounds=20
  round 0/20: seed=0xedf04fa57d7e8096
  round 1/20: seed=0xc324af2d510cccea
  round 2/20: seed=0x29b5a233e97efed
  round 3/20: seed=0x586b8903a1d2caf2
  round 4/20: seed=0x6afbbe1eb01672f9

I'm not familiar with the other long run tests. Which one do you think would fit this test?

andrei-stoian-zama · 2026-04-13T08:16:42Z

+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 20 * * *"


let's only run it on Mondays (integer long run will run the same way)

andrei-stoian-zama · 2026-04-13T08:38:59Z

+  workflow_dispatch:
+  schedule:
+    - cron: "0 20 * * *"
+  pull_request:


please use an appropriate "should_run" filter - only changes to ZK trigger it

andrei-stoian-zama · 2026-04-13T08:40:20Z

+        });
+    }
+
+    let is_minimal =


if it's set and not "TRUE" then please assert. otherwise we'll make mistakes with the syntax

andrei-stoian-zama · 2026-04-13T08:43:09Z

+///     they agree.
+#[test]
+fn test_pke_v2_gpu_cpu_equivalence() {
+    let seed: u64 = thread_rng().gen();


if we print the seed we may want to re-use it easily through "TFHE_RS_LONGRUN_TESTS_SEED", why don't we use that here ?

cla-bot bot added the cla-signed label Apr 7, 2026

pdroalves added the approved label Apr 7, 2026

pdroalves force-pushed the pa/chore/zk-long-tests branch from 778b322 to 5f8fc4e Compare April 7, 2026 19:55

zama-bot removed the approved label Apr 7, 2026

andrei-stoian-zama requested changes Apr 8, 2026

View reviewed changes

feat(zk): add GPU ZK long-run tests and CI workflow

e0f2e3f

pdroalves force-pushed the pa/chore/zk-long-tests branch from 5f8fc4e to e0f2e3f Compare April 8, 2026 15:00

pdroalves added the approved label Apr 8, 2026

pdroalves requested a review from andrei-stoian-zama April 8, 2026 18:24

andrei-stoian-zama reviewed Apr 13, 2026

View reviewed changes

Conversation

pdroalves commented Apr 7, 2026

PR content/description

Check-list:

Uh oh!

andrei-stoian-zama left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pdroalves commented Apr 8, 2026

Uh oh!

pdroalves commented Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

andrei-stoian-zama left a comment •

edited

Loading

pdroalves commented Apr 8, 2026 •

edited

Loading