12/25 new papers

[AmberLJC]

add papers

HydraInfer: Hybrid Disaggregated Scheduling for Multimodal Large Language Model Serving
https://arxiv.org/abs/2505.12658

Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing
https://arxiv.org/abs/2512.17574

Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings
https://arxiv.org/abs/2509.10534

[AmberLJC]

@claude

[claude]

Claude finished @AmberLJC's task in 2m 3s —— View job • claude/issue-46-20251226-0739 • Create PR ➔

---

Task Complete

I've added 2 of the 3 papers to the repository. Here's what I did:

Papers Added ✅

Both papers were added to the Multi-Modal Serving Systems section (README.md:318-319):

• HydraInfer: Hybrid Disaggregated Scheduling for Multimodal Large Language Model Serving
• Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing

Paper Not Added ⚠️

• Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings (https://arxiv.org/abs/2509.10534)

Reason: This paper appears to focus on positional embedding architecture rather than systems research. According to the repository's CLAUDE.md guidelines, this repository focuses on "systems research, not pure ML or algorithm papers." All existing attention/architecture papers in the repo (FlashAttention, Ring Attention, etc.) have clear system efficiency implications (faster inference, memory management), whereas this paper appears to be primarily about model architecture without clear systems contributions.

If you believe this paper should be included, please let me know which section would be most appropriate, or if there's a systems angle I'm missing!

Next Steps

--- | Branch