vllm - 💡(How to fix) Fix [Feature]: unsupported architecture [1 comments, 2 participants]

We are using version v0.20.0 and encountering an "unsupported architecture" error (4090) during deployment. When should we begin adaptation, or is it already adapted? Is our startup command incorrect? The correct startup command is: vllm serve /root/xinglin-data/api_model/DeepSeek-V4-Flash --trust-remote-code --tensor-parallel-size 8 --enable-expert-parallel --block-size 128 --max-model-len 8192 --gpu-memory-utilization 0.90 --max-num-seqs 8 --max-num-batched-tokens 4096 --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 --reasoning-parser deepseek_v4 --enable-auto-tool-choice --host 0.0.0.0 --port 12800 --kv-cache-dtype fp8 --enforce-eager

TL;DR

The "unsupported architecture" error (4090) during deployment with version v0.20.0 may be resolved by checking the compatibility of the deployment environment with the version being used.

Guidance

• Verify that the deployment environment's architecture is supported by version v0.20.0.
• Check the startup command for any incorrect or unsupported parameters that might be causing the error.
• Review the documentation for version v0.20.0 to ensure that all requirements for deployment are met.
• Consider reaching out to the support team or checking the documentation page for more information on resolving the "unsupported architecture" error.

Notes

The provided information does not specify the exact architecture of the deployment environment, making it difficult to provide a more specific solution.

Recommendation

Apply workaround: Check the compatibility of the deployment environment and review the startup command for any incorrect parameters, as the issue seems to be related to the environment or command rather than the version itself.