[sergereview]

Original CI failure

• Failure group: 2 integration tests for model mamba2failing withoutput_mismatch (other (2))
• tests/models/mamba2/test_modeling_mamba2.py::Mamba2IntegrationTest::test_simple_generate [single-gpu] (output_mismatch, seen 21/21)

CI traceback — tests/models/mamba2/test_modeling_mamba2.py::Mamba2IntegrationTest::test_simple_generate

(line 468)  AssertionError: '<s>[[81 chars]rld!" program written in C++:\n\n```cpp\n#include <iostream>\n' != '<s>[[81 chars]rld!" program in C++:\n\n```cpp\n#include <iostream>\n\n'

Where to watch it:

• test_simple_generate — Test dashboard

The A10G (compute capability 8.6) produces the "written in C++" variant, identical to the existing ("xpu", 3) entry. Without a ("cuda", 8) key, Expectations falls back to ("cuda", 7) (score 1, the only cuda key), which has the "in C++" variant — a stale expectation.

Adds a ("cuda", 8) key with the value the runner actually produces. The difference from ("cuda", 7) is a single word ("written") in the generated text — normal hardware-level variation for a language model, not a regression.

Relates to #49943

---

⚠️ Not verified — this patch changed the expectation

This patch changes only expected values in test files — the assertions were rewritten, not the code under test.

The GPU run below is not evidence that this change is correct: the tests were re-run after the patch rewrote what they assert, so they pass by construction. Judge the new value on its merits.

• Passes with this patch: https://github.com/huggingface/transformers/actions/runs/40044056764
• Failed before the fix (base commit): https://github.com/huggingface/transformers/actions/runs/40041691984

Possibly related

Existing issues/PRs mentioning test_simple_generate (keyword match — not verified to share a root cause):

• #46773 — Fix padding mask skipped for batch size 1 in linear-attention models (PR, closed, updated 2026-10-06)
• #48978 — Contract the mamba2 chunk scan with einsum instead of broadcast-then-sum (PR, closed, updated 2026-09-21)
• #47630 — 🚨 [Kernels] Refactor all linear attn models & native kernels fallback (PR, closed, updated 2026-08-05)
• #44950 — 🚨 [Cache] Native mamba & hybrid cache (PR, closed, updated 2026-03-31)
• #43590 — Remove many output_attentions and other traced outputs on 100+ models (PR, closed, updated 2026-03-12)

---

This change was produced automatically by serge from a CI failure report. The patch was generated by an LLM and applied by serge; review before merging.

serge v0.1.0 · model: Qwen/Qwen3.8-27B:deepinfra · 14 LLM turns · 13 tool calls · 391.8s · 288606 in / 6105 out tokens

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: mamba2

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.