GitLab

Co-authored-by: Michael Goin <michael@neuralmagic.com>

Co-authored-by: Michael Goin <michael@neuralmagic.com>

Signed-off-by: Travis Johnson <tsjohnso@us.ibm.com>

Co-authored-by: hezhihui <hzh7269@modelbest.cn>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>

original title: [Distributed][Model] Rank-based Component Creation for Pipeline Parallelism Memory Optimization

Co-authored-by: Robert Shaw <rshaw@neuralmagic.com>
Co-authored-by: ZX <zx@lbx.dev>

Signed-off-by: Muralidhar Andoorveedu <muralidhar.andoorveedu@centml.ai>

Co-authored-by: rsnm2 <rshaw@neuralmagic.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-neuralmagic@users.noreply.github.com>

Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>

The 2nd PR for #4532.

This PR supports loading FP8 kv-cache scaling factors from a FP8 checkpoint (with .kv_scale parameter).

Currently we need to call rotary embedding kernel for each LoRA, which makes it hard to serve multiple long context length LoRA. Add batched rotary embedding kernel and pipe it through.

It replaces the rotary embedding layer to the one that is aware of multiple cos-sin-cache per scaling factors.

Follow up of https://github.com/vllm-project/vllm/pull/3095/files