vllm - 💡(How to fix) Fix [CI Failure]: mi300_1: DeepSeek V2-Lite Prefetch Offload Accuracy (H100-MI300) [3 comments, 2 participants]

• python3 - Traceback (most recent call last): File "<stdin>", line 3, in <module> AssertionError: deepseek-ai/DeepSeek-V2-Lite prefetch_offload accuracy 0.015

TL;DR

The failing test can be mitigated by investigating the deepseek_v2_lite_prefetch_offload.sh script and the AssertionError related to the deepseek-ai/DeepSeek-V2-Lite model's prefetch offload accuracy.

Guidance

• Review the deepseek_v2_lite_prefetch_offload.sh script to understand the test workflow and identify potential issues.
• Investigate the AssertionError and the expected accuracy value of 0.015 to determine if it's a valid threshold.
• Check the model's configuration and training data to ensure they are correct and up-to-date.
• Run the test with additional logging or debugging to gather more information about the failure.

Example

No code snippet is provided as the issue lacks specific code details.

Notes

The issue seems to be related to a specific test case and model configuration. The provided information is limited, and further investigation is needed to determine the root cause.

Recommendation

Apply workaround: Investigate and adjust the deepseek_v2_lite_prefetch_offload.sh script and model configuration to resolve the AssertionError and improve test reliability.