Starscream: Reimagining LLM Sharding on AMD Instinct GPUs with GPU Partitioning
cs.DC, cs.AR, cs.LG
Submitted: 2026-11-11
Updated: 2026-11-11
Code: https://github.com/NVIDIA/TensorRT-LLM
Terminology
Sources
- Striped Attention: Faster Ring Attention for Causal Transformers
- DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
- DeepSeek-V3 Technical Report
- USP: A Unified Sequence Parallelism Approach for Long Context Generative AI
- LoongTrain: Efficient Training of Long-Sequence LLMs with Head-Context Parallelism
- DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models
- GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding
- Helix Parallelism: Rethinking Sharding Strategies for Interactive Multi-Million-Token LLM Decoding
- Efficiently Scaling Transformer Inference
- Fast Transformer Decoding: One Write-Head is All You Need
- Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
- Eliminating Multi-GPU Performance Taxes: A Systems Approach to Efficient Distributed LLMs
Related papers
- Programmable Probabilistic Computer with 1,000,000 p-bits
- BatchGen: An Architecture for Scalable and Efficient Batch Inference
- Seamless Reconfiguration for DAG-BFT (Technical Report)
- ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
- iScheduler: Reinforcement Learning-Driven Continual Optimization for Large-Scale Resource Investment Problems
- TIRx: A Unified Tile Primitive ML Compiler Across Heterogeneous Backends