Tagged articles
training acceleration
11 articles · Page 1 of 1
BigMac introduces a dependency‑safe nested pipeline that keeps the LLM schedule intact while inserting encoder and generator work, achieving 1.08‑1.9× speedup and stable GPU memory usage across large‑scale multimodal LLM training workloads.
BigMacPipeline Parallelismmemory efficiency
0 likes · 17 min read


Didi TechMay 14, 2026 · Artificial Intelligence
This article analyzes the latency bottlenecks of large language models in multi‑round AI Agent scenarios, introduces SpecForge‑based speculative decoding and Unified Sequence Parallelism (USP) techniques applied to the EAGLE-3 model, and presents benchmark results showing over two‑fold Accept‑Len gains and 35‑44% reductions in P95 token‑level latency while enabling 128K context training on an 8‑GPU node.
Agent AIEAGLE-3Unified Sequence Parallelism
0 likes · 26 min read

NVIDIA, Hong Kong University, and MIT introduced the Sol‑RL framework, which uses reinforcement‑learning‑guided sampling to cut diffusion model training time by several‑fold without sacrificing image quality, potentially lowering entry barriers for small teams and shifting the AIGC industry toward an efficiency‑driven competition.
AIGCNVIDIASol-RL
0 likes · 6 min read

The article details MiniMax’s Forge system, an asynchronous native Agent‑RL architecture that standardizes Agent‑LLM interaction, introduces engineering optimizations, novel scheduling, prefix‑tree merging and reward designs, enabling million‑sample daily throughput, stable reward growth and up to 40‑fold training acceleration for the MiniMax M2.5 model.
Agent ArchitectureMixed SchedulingReward Design
0 likes · 17 min read

This technical report details the end‑to‑end workflow for integrating an AIGC video generation module into a social app, covering requirement analysis, model and hardware selection, dataset construction, LoRA and full‑parameter training, multiple acceleration techniques such as Sage Attention, TeaCache, XDiT, gradient‑checkpointing offload, tiled VAE, and quantization, followed by extensive performance evaluation and metric‑based ranking of the final models.
AI Video GenerationLoRA fine-tuningSage Attention
0 likes · 38 min read

SPEC‑RL introduces speculative rollouts that reuse verified historical rollouts as prefixes, cutting rollout time by 2–3× while maintaining or improving performance across various math and reasoning benchmarks, and works seamlessly with PPO, GRPO, DAPO and other on‑policy algorithms.
AI Efficiencylarge language modelsreinforcement learning
0 likes · 8 min read

The article details the design and deployment of 360's AI Compute Center, covering GPU server selection, high‑performance networking, Kubernetes‑based cluster management, advanced scheduling, training and inference acceleration techniques, and a comprehensive AI development platform with visualization and fault‑tolerance features.
AI infrastructureGPU clusterKubernetes
0 likes · 21 min read


58 TechJun 3, 2024 · Artificial Intelligence
This article systematically analyzes popular parameter‑efficient fine‑tuning (PEFT) techniques for large language models—including Adapter Tuning, Prefix Tuning, LoRA, QLoRA, AdaLoRA, and SoRA—detailing their principles, implementation code, experimental results on NLU tasks, and practical acceleration using the Unsloth library.
AdaLoRALoRAPEFT
0 likes · 39 min read

The article explains the scaling challenges of ever‑larger LLMs, introduces the MFU performance metric, surveys industry parallelism and memory‑saving techniques, and details Baidu’s AIAK‑LLM suite—including resource, component and acceleration layers—as well as concrete training and inference optimizations that raise MFU by 30‑60% and cut deployment costs.
AI infrastructureMFUParallelism
0 likes · 25 min read

QQGC, Tencent’s two‑stage text‑to‑image model that separates CLIP‑based Prior mapping from a Stable Diffusion Decoder, leverages T5‑enhanced text embeddings and a suite of efficiency tricks—including FP16, flash attention, ZeRO and GPU‑RDMA—to train over‑2 B‑parameter models on 64 GPUs, achieving state‑of‑the‑art FID and CLIP scores while supporting image variation, semantic img2img, precise CLIP‑vector edits and unsafe‑content filtering, and now powers the company’s Magic Painting Room.
AI paintingCLIP embeddingdiffusion model
0 likes · 12 min read

The MetaSpore team detailed a three‑pronged optimization—sparse model tuning, training‑pipeline acceleration, and low‑level framework tweaks—that boosted DeepRec CTR model training efficiency by over three times without sacrificing AUC, securing first place in the global AI competition.
AI competitionCTRDeepRec
0 likes · 9 min read
