[serge] Fix 8 integration tests for model deepseek_v32 failing with load_error (other (8))

[sergereview]

Original CI failure

• Failure group: 1/1: 8 integration tests for model deepseek_v32failing withload_error (other (8))
• tests/models/deepseek_v32/test_modeling_deepseek_v32.py::DeepseekV32IntegrationTest::test_batched_generation_padding [multi-gpu] (load_error, seen 5/7)

Relates to #46906

The DeepseekV32 integration tests load deepseek-ai/DeepSeek-V3.2-Exp with device_map="auto". On CI runners whose auto device_map offloads some weights to disk, loading fails because no offload_folder is provided. This is expected for large MoE models whose internal weight layout differs from the checkpoint layout.

Root cause: test calls missing the offload_folder argument required by the accelerate disk-offload path.

Fix: pass a temporary offload_folder to every DeepseekV32ForCausalLM.from_pretrained call in the integration test class.

---

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 · 25 LLM turns · 30 tool calls · 31.0s · 412737 in / 3828 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.

[github-actions]

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

run-slow: deepseek_v32

[github-actions]

CI Dashboard: View test results in Grafana