Support vLLM embeddings (Qwen3) | Runpod | Page 1

sweet coral Jan 30, 2026, 11:40 AM

#

The default behaviour of vLLM on serverless is --task generate (for normal completions), but it would be awesome if we could use embeddings via --task embed.

Relevant issue: https://github.com/runpod-workers/worker-vllm/issues/170

GitHub

Feature request: environment variable for `--task` · Issue #170 ·...

Hello! We are trying to use serverless for embedders like BAAI/bge-multilingual-gemma2 that work very well on your pods. For this embedder, one has to pass explicitly the argument --task embed for ...

naive mauveBOT Jan 30, 2026, 11:40 AM

#

wanton lava Jan 30, 2026, 2:48 PM

#

KoboldCpp can generate embeddings if you want

#Support vLLM embeddings (Qwen3)