30건 · 관련도순
Achieve aggressive v3 performance targets: 2.49x-7.47x Flash Attention speedup, 150x-12,500x search improvements, 50-75% memory reduction. Comprehensive benchmarking and optimization suite.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Guide MNN Metal backend development and optimization, from finding performance bottlenecks to changing GPU computation, combining operators, and adjusting execution scheduling. A task-based document index directs work toward relevant implementation rules, benchmarks, and correctness checks.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Write landing page copy with attention to the hero, value proposition, social proof, objection handling, and conversion-focused CTAs. Use this skill whenever the user wants to write a landing page, sales page, hero section, or any conversion-focused web copy. Triggers on landing page, sales page, hero copy, value proposition, headline, subheadline, hero section, CTA copy, conversion copy, opt-in page, squeeze page. Also triggers when the user has a marketing campaign or product launch needing dedicated conversion copy. Use `cro-optimization` instead when the page already exists and the ask is to lift its conversion rate through testing rather than to write the copy.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Provides guidance for writing, optimizing, and benchmarking Triton kernels for Intel XPU GPUs (Battlemage/Arc Pro B50) using the Xe-Forge optimization framework. Includes an LLM-driven trial-loop workflow (analyze, validate, benchmark, profile, finalize), XPU-specific patterns (tensor descriptors, GRF mode, tile swizzling), KernelBench fused kernels, and Flash Attention.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
优化实际模型推理链路，将正确性对齐、分段 profiling、显存与数据搬运、TensorRT/ONNX/PyTorch 后端、attention/kernel、FP8/compile、缓存与少步采样、质量回归、GPU 成本和服务验收串成同一实验闭环。当用户要求推理提速、降低显存或 GPU 成本、复现模型效果、定位 GPU 利用率低、优化图像/视频/扩散模型或自托管 LLM 时使用，提供 model inference performance optimization 的脱敏案例。纯 API 调用、提示词或 token 用量优化，以及电脑卡顿，不进入本技能的模型推理实验流程。
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Acelere a inferência de LLMs usando especulative decoding, múltiplas cabeças Medusa e técnicas de lookahead decoding. Use ao otimizar velocidade de inferência (aceleração de 1,5-3,6×), reduzir latência em aplicações em tempo real ou fazer deploy de modelos com recursos computacionais limitados. Cobre modelos draft, atenção em árvore, iteração de Jacobi, geração paralela de tokens e estratégias de deploy em produção.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.