Murali Andoorveedu issues

Results 6 issues of


                                            Murali Andoorveedu

[RFC] Make more operations inplace (GELU, BatchNorm, LayerNorm)

### 🚀 The feature, motivation and pitch # Make more operations inplace (GELU, BatchNorm, LayerNorm) ## **Summary** Hi PyTorch team, We would like to enable users to make the following...

module: nn

triaged

enhancement

needs research

[Distributed] Add P2P versions of *object_list operations

This PR adds `send_object_list` and `recv_object_list` to `distributed_c10d.py`. This is extending functionality already present in PyTorch with `broadcast_object_list` that I noticed was missing and decided to upstream. With this change,...

oncall: distributed

module: cpu

triaged

module: mkldnn

open source

module: amp (automated mixed precision)

release notes: quantization

release notes: distributed (c10d)

module: inductor

module: dynamo

module: distributed_checkpoint

[Core] Pipeline Parallel Support

Adds initial pipeline parallelism support to vLLM. ToDo: Milestone 1: POC Prototype - [x] Make changes to support multiple schedulers and cache engines in `worker.py`, `llm_engine.py`, `async_llm_engine.py` and block managers....

[Docs] Docs update for Pipeline Parallel

This adds docs for pipeline parallel. cc: @simon-mo @youkaichao @njhill

[Models] Add remaining model PP support

Add all other possible models for PP FIX #7684

ready

[V1][Tests] Adding additional testing for multimodal models to V1

Adding tests for V1 multimodal abort (as requested by @WoosukKwon) as well as load.

ready