[Speculative Decoding] Support draft model on different tensor-parallel size than target model (#5414)
Attach a file by drag & drop or click to upload