Shigangli / Ok-Topk
Ok-Topk is a scheme for distributed training with sparse gradients. Ok-Topk integrates a novel sparse allreduce algorithm (less than 6k communication volume which is asymptotically optimal) with the decentralized parallel Stochastic Gradient Descent (SGD) optimizer, and its convergence is proved theoretically and empirically.
☆23Updated last year
Related projects ⓘ
Alternatives and complementary repositories for Ok-Topk
- Hi-Speed DNN Training with Espresso: Unleashing the Full Potential of Gradient Compression with Near-Optimal Usage Strategies (EuroSys '2…☆15Updated last year
- ☆23Updated 2 years ago
- Artifact for OSDI'23: MGG: Accelerating Graph Neural Networks with Fine-grained intra-kernel Communication-Computation Pipelining on Mult…☆37Updated 8 months ago
- ☆14Updated 5 months ago
- Artifacts for our ASPLOS'23 paper ElasticFlow☆52Updated 6 months ago
- ☆14Updated 2 years ago
- ☆9Updated last year
- A Cluster-Wide Model Manager to Accelerate DNN Training via Automated Training Warmup☆34Updated last year
- Artifacts for our SIGCOMM'22 paper Muri☆40Updated 10 months ago
- ☆23Updated last year
- Cupcake: A Compression Scheduler for Scalable Communication-Efficient Distributed Training (MLSys '23)☆9Updated last year
- ☆18Updated 2 years ago
- Compiler for Dynamic Neural Networks☆43Updated last year
- ☆41Updated last year
- ☆33Updated last year
- ☆23Updated last year
- Proteus: A High-Throughput Inference-Serving System with Accuracy Scaling☆8Updated 8 months ago
- ☆13Updated 2 years ago
- Chimera: Efficiently Training Large-Scale Neural Networks with Bidirectional Pipelines.☆46Updated 11 months ago
- Artifact for PPoPP22 QGTC: Accelerating Quantized GNN via GPU Tensor Core.☆27Updated 2 years ago
- LLM serving cluster simulator☆81Updated 6 months ago
- ☆37Updated 3 years ago
- ☆8Updated 2 years ago
- ☆73Updated last year
- ☆69Updated last year
- DISB is a new DNN inference serving benchmark with diverse workloads and models, as well as real-world traces.☆54Updated 3 months ago
- SHADE: Enable Fundamental Cacheability for Distributed Deep Learning Training☆29Updated last year
- SOTA Learning-augmented Systems☆33Updated 2 years ago
- REEF is a GPU-accelerated DNN inference serving system that enables instant kernel preemption and biased concurrent execution in GPU sche…☆85Updated last year
- ☆38Updated 4 years ago