🚀 The feature, motivation and pitch

Due to its powerful computing capabilities, Ascend NPU is currently used by many customers. We hope that vLLM can run smoothly on Ascend NPU, thereby serving more users. We have also completed the adaptation of vLLM's v0.4.2 version on Ascend NPU hardware. The adapted Ascend-vLLM demonstrates good performance in terms of ease of use and high performance. Now we plan to contribute the code to the vLLM project. Additionally, we welcome everyone to participate in the joint construction and collaboratively build the framework capabilities for large models on Ascend NPU.

Alternatives

No response

Additional context

No response