[sergereview]

Original CI failure

• Failure group: 2 integration tests for model cwmfailing withimport_or_config (other (2))
• tests/models/cwm/test_modeling_cwm.py::CwmIntegrationTest::test_cwm_integration [multi-gpu] (import_or_config, seen 7/7)

CI traceback — tests/models/cwm/test_modeling_cwm.py::CwmIntegrationTest::test_cwm_integration

(line 1967)  AttributeError: 'CwmDecoderLayer' object has no attribute 'attention_type'

Where to watch it:

• test_cwm_integration — Test dashboard

Relates to #48413

CwmIntegrationTest::test_cwm_integration fails because CwmDecoderLayer has no attention_type attribute, yet the test asserts layer.attention_type == model.config.layer_types[i].

Root cause: CwmDecoderLayer in modular_cwm.py only replaces the attention module; it never stores the per-layer attention type from config.layer_types.

Fix: set self.attention_type = config.layer_types[layer_idx] in CwmDecoderLayer.__init__. make fix-repo will propagate this to the generated modeling_cwm.py.

---

✅ Verified on GPU

serge ran the targeted @slow test(s) on a GPU runner and opened this PR only after they passed with this patch. Each targeted test was run 5× on both the pre-patch and patched trees to rule out flakiness — the result held on every run.

• Passes with this patch: https://github.com/huggingface/transformers/actions/runs/33279265651
• Failed before the fix (base commit): https://github.com/huggingface/transformers/actions/runs/33278229789

---

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: moonshotai/Kimi-K2.7-Code · 30 LLM turns · 30 tool calls · 47.8s · 784337 in / 10025 out tokens

[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.

[zucchini-nlp]

It was moved under CwmAttention.layer_type which we can use instead, and fix the test assertion

I really don't know how to ask Serge to re-do it, so cc @tarekziade

[tarekziade]

It was moved under CwmAttention.layer_type which we can use instead, and fix the test assertion

I really don't know how to ask Serge to re-do it, so cc @tarekziade

We can't ask for new edits for now, but that's a good idea I will add that feature. In the interim, fixing within the branch

[tarekziade]

run-slow: cwm

[github-actions]

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

run-slow: cwm

[github-actions]

Workflow Run ⚙️

This comment contains run-slow, running the specified jobs:

models: ["models/cwm"]
quantizations: []

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 33376462572:1
Result: success | Jobs: 8 | Tests: 379 | Failures: 0 | Duration: 4m 34s

[zucchini-nlp]

Yep thanks

[github-actions]

CI Results

Workflow Run ⚙️

Commit Info

Context	Commit	Description
RUN	bfccd9ef	workflow commit (merge commit)
PR	3323f777	branch commit (from PR)
main	8e35c981	base commit (on main)

✅ No failing test specific to this PR 🎉 👏 !

[tarekziade]

worked, left failures on that model are OOMs we will address in another patch