test(vidur): cover AICB KV cache save predictions

[catyans]

Summary

Add a focused regression test for the AICB execution-time predictor to ensure it always publishes the attn_kv_cache_save table consumed by Vidur batch execution.

This protects the failure reported in #266. The current master implementation contains the prediction after the SimAI 1.6 merge, but the behavior had no automated coverage and the issue remains open.

Test coverage

• verifies the AICB table contains attn_kv_cache_save;
• verifies every token count in the configured prediction range is populated;
• verifies generated execution times are non-negative.

Validation

• git diff --check
• Python 3.11 remote H20 syntax smoke for the predictor and regression test (ede4acc)

A full pytest run requires the Vidur Python dependency environment; the remote H20 base OS does not carry those packages.

Fixes regression coverage for #266.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[catyans]

CLA follow-up: rewrote the single commit metadata (content unchanged) so author, committer, and Signed-off-by use the email linked to GitHub user catyans. New head: 82d8af3a.

[catyans]

recheck

[Copilot]

Pull request overview

Adds a regression test to prevent a recurrence of SimAI issue #266 by ensuring the AICB execution-time predictor publishes attn_kv_cache_save predictions across the configured token range.

Changes:

• Introduces a new pytest regression test validating attn_kv_cache_save is present in AICB compute predictions.
• Asserts predictions are populated for every token count in the range and that values are non-negative.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.