0 views collected around this technical thread.
This article explains how to dramatically speed up PyTorch model training using code optimizations, mixed‑precision, torch.compile, distributed data parallelism, and DeepSpeed, presenting benchmark results that show up to 11.5× acceleration on multiple GPUs while maintaining high accuracy.
The DeepSeek open‑source week introduced seven breakthrough technologies—FlashMLA, DeepGEMM, DeepEP, DualPipe, EPLB, 3FS, and Smallpond—that together overhaul data flow, algorithmic complexity, hardware utilization, MoE communication, and resource balancing, dramatically improving large‑model training efficiency and lowering entry barriers for the AI industry.
This article presents a comprehensive overview of EasyRec’s recommendation system architecture, detailing training and inference optimizations, embedding parallelism, CPU/GPU placement strategies, online learning pipelines, and network compression techniques that together improve scalability, latency, and cost efficiency.
EasyRec, Alibaba Cloud’s modular recommendation framework, unifies configurable data, embedding, dense, and output layers on MaxCompute, EMR, and DLC, and speeds training with deduplication, EmbeddingParallel sharding, lock‑free hash tables, GPU embeddings, and AMX BF16, while inference benefits from operator fusion, low‑precision AVX/AMX kernels, compact caches, batch merging, and network compression, enabling real‑time online learning and delivering higher recommendation quality at lower cost in e‑commerce.
This article presents a comprehensive overview of EasyRec's recommendation system architecture, detailing training and inference optimizations, distributed deployment strategies, operator fusion techniques, online learning pipelines, and network-level improvements to enhance performance and scalability.
This article presents a comprehensive approach to enhancing large language models' mathematical reasoning by reviewing model architectures, introducing mixed CoT‑PoT instructions, generating and filtering synthetic data, and applying multi‑stage training optimizations such as RFT, PPO, and DPO, with detailed experimental results and Q&A.
GBA (Global Batch Gradients Aggregation) introduces a training mode that seamlessly switches between synchronous and asynchronous learning for search‑advertising models by keeping a constant global batch size, using token‑controlled gradient aggregation and staleness management to retain synchronous‑level accuracy while preserving asynchronous efficiency and eliminating manual hyperparameter tuning.