Qwen releases FP8 package for 27B vision-language model

A Qwen repository provides FP8-quantized weights and configuration files for Qwen3.8-27B in the Hugging Face Transformers format. The repository says the 27-billion-parameter vision-language model is compatible with Transformers, vLLM, SGLang and TokenSpeed, and has a native context length of 262,144 tokens that can be extended to 1 million tokens. It says the weights use fine-grained FP8 quantization with a block size of 128. The repository also says a hosted Qwen Cloud version with a 1-million-token default context and built-in tools is planned, but that service is coming soon.