“cuda”的搜索结果

Use when auditing the imp codebase for structural debt, dead code, god-objects/files, duplication, flag sprawl, or deciding whether a cleanup is worth shipping - "structure audit", "tech debt", "is this still used", "dead code", "refactor for clarity", "should we split this file", "remove this flag", "file size gate red", "alloc sites gate red". Do NOT use for build/test mechanics (building-and-testing), perf/kernel work (benchmark-cuda / sm120-cuda-expert), or output quality (check-degeneration).

中文简介正在准备中。当前显示原文说明。

Use when working on imp's quantization formats, loaders, or dequant paths - GGUF Q4_0...Q8_0/Q*_K/IQ4/MXFP4, NVFP4 two-level scaling (Modelopt vs compressed-tensors layouts), FP8 E4M3, StorageTier, decode cache, KV-cache dtypes (auto/fp16/fp8/int8/int4/nvfp4/mxfp4), "which quant should I use", "which KV dtype", scale-factor layout, dequant kernel wiring, imp-quantize / AWQ, judging quant quality (PPL). Do NOT use for writing/optimizing GEMM/GEMV kernels (sm120-cuda-expert) or measuring quant perf (benchmark-cuda).

中文简介正在准备中。当前显示原文说明。

Use when writing HIP kernels with hipcc, porting CUDA code through HIPIFY, profiling with rocprofv3, debugging with rocgdb, or optimizing for MI300X. Not for NVIDIA-side kernels: use cuda.

中文简介正在准备中。当前显示原文说明。

Use when writing, optimizing, or benchmarking a C++ CPU kernel with AVX2 or AVX512 intrinsics for the Hugging Face kernels ecosystem. Not for CUDA kernels: use cuda.

中文简介正在准备中。当前显示原文说明。

窓際族物語の動画を、チャプター毎のキーフレーム画像を作らずに「プロンプト＋キャラクターシート＋セリフ音声」だけでMiniMax H3（R2V）に生成させるワークフロー。ユーザーから「画像なしで動画を作って」「キーフレームなしで動画を作って」「シートだけで動画を作って」と指示されたときに必ず使用する。大まかな指示→チャプター分割案→チャプター毎の動きを人間に確認→全チャプター確定→プロンプト作成の順で進め、確定前にプロンプトを書かない。画風の注意文はチャプターに映るものだけを最小限で書き、モブキャラはラン専用のキャラクターシートを作って渡す。動画生成の実行先は/colab-video（Colab）または/local-videoステップ7（ローカルCUDA）。

中文简介正在准备中。当前显示原文说明。

帮助编写和检查PyTorch模型与训练代码，涵盖数据加载、实验可复现性、训练状态保存与恢复，并提供改善GPU显存占用和运行速度的方法。

• 编写与审查PyTorch训练代码
• 排查实验结果不一致的原因
• 改善加载速度与GPU显存占用

使用TRL指导语言模型进一步训练，使回答更符合指令或人的偏好。涵盖数据准备、训练方式选择、奖励模型训练与评估，以及内存不足和训练不稳定时的调整。

• 用回答示例训练模型遵循指令
• 用DPO学习回答偏好
• 构建使用奖励模型的RLHF流程

帮助在NVIDIA GPU上加速大语言模型的响应生成，并通过API提供服务。支持量化、批量处理和多GPU分布运行，以调整显存占用与服务性能。

• 通过聊天API提供语言模型服务
• 批量处理大量提示词
• 通过量化减少显存占用

用Python实现文字生成图像、现有图像风格转换和指定区域补全，并指导如何控制构图与风格、调整生成参数及减少内存占用。

• 根据文字生成图像
• 将现有图像转为水彩风格
• 补全照片中的指定区域

使用 NeMo Curator 整理大语言模型训练数据，去除重复和低质量内容，删除或替换个人信息。还涵盖处理流程的构建，以及图像、视频和音频数据的筛选整理。

• 将网页采集文本整理为训练数据
• 按完全匹配、近似和语义去重
• 删除或替换文本中的个人信息

分析开放权重语言模型的拒答机制，支持无需重新训练的权重修改实验，并根据计算环境选择模型与方法，比较修改前后的拒答率和回答质量。

• 研究语言模型的拒答机制
• 无需重新训练开展权重修改实验
• 按计算环境选择模型与方法

根据歌词及曲风、乐器标签在本地生成歌曲，指导安装、模型下载、输入整理和节省显存的设置，并协助将生成结果保存为MP3文件。

• 将自写歌词生成歌曲
• 通过标签尝试曲风和乐器
• 在本地生成音乐

指导使用 AudioCraft 选择模型并实现从文字生成音乐、音效和环境声，涵盖旋律及参考音频引导、立体声输出和批量生成。

• 根据文字生成音乐
• 制作音效和环境声
• 使用旋律或参考音频生成音乐

Plan and review PufferLib reinforcement learning workflows, where agents learn through rewards. Covers version-specific environment checks, parallel execution, training and evaluation plans, plus checkpoint inspection before loading saved models.

中文简介正在准备中。当前显示英语简介。

• Choose APIs for a specific PufferLib version
• Check environment behavior before training
• Create training and evaluation plans with separate seeds

Generate candidate positions and orientations for small molecules binding to proteins with DiffDock and DiffDock-L. Prepare single or batch docking runs, organize predicted poses by model confidence, and review validation needs without treating confidence as binding affinity.

中文简介正在准备中。当前显示英语简介。

• Predicting binding poses for a protein and small molecule
• Preparing and validating batch docking inputs
• Comparing poses across receptor conformations

Organize SGLang GPU operations when adding or moving code. It guides API placement, operation registration, test organization, and migration checks to preserve existing behavior, references, and continuous integration coverage.

中文简介正在准备中。当前显示英语简介。

• Choose where to place a new GPU operation
• Split or move existing operation files
• Review API registration and test placement

Build and debug sglang tests comparing token probabilities during prompt processing and step-by-step generation. Separate numerical differences from reused-state bugs, choose test thresholds, and locate the first operation that diverges.

中文简介正在准备中。当前显示英语简介。

• Adding probability consistency tests for a model
• Choosing or justifying a kl_div threshold
• Separating numerical differences from cache state bugs

Investigate SGLang serving failures by collecting server state, preserving failing requests, and replaying them on a clean target. Summarize evidence and likely causes, then choose the next debugging tool for crashes, stalls, or performance regressions.

中文简介正在准备中。当前显示英语简介。

• Investigate failed health checks and growing queues
• Reproduce request-specific timeouts or crashes
• Investigate output changes after deployment

Investigate phishing emails that hide malicious links in QR codes. Decode destinations, analyze suspicious URLs, configure email detection rules, and plan mobile protections and awareness training for people who scan codes on their phones.

中文简介正在准备中。当前显示英语简介。

• Investigating suspicious emails containing QR codes
• Configuring QR detection for email images and PDF attachments
• Protecting phones from malicious QR destinations

Convert LLMs to GGUF and reduce model size through quantization, which stores weights at lower numerical precision. Prepare models for CPU or Apple Silicon inference, compare compression levels, and integrate them with local AI tools.

中文简介正在准备中。当前显示英语简介。

• Preparing an LLM for CPU-only use
• Running a local model on Apple Silicon
• Comparing model size and quality across quantization levels

Guide language model training with TRL, from learning to follow instructions to matching human preferences. Covers method selection, training data, reward models that score responses, evaluation, and adjustments for memory limits or unstable training.

中文简介正在准备中。当前显示英语简介。

• Train a model on instruction and response examples
• Align responses using preferred and rejected answer pairs
• Build a training pipeline using human feedback

Build image generation workflows in Python with Stable Diffusion and HuggingFace Diffusers. The guide covers text prompts, transforming existing images, replacing selected regions, controlling composition, applying learned styles, and adjusting memory use during generation.

中文简介正在准备中。当前显示英语简介。

• Generating images from text descriptions
• Changing the style of an existing image
• Replacing a selected region of an image

Configure LLM inference and serving on NVIDIA GPUs with TensorRT-LLM. Covers quantization, which reduces numerical precision, batching requests during generation, and distributing models across GPUs to improve processing capacity and response times.

中文简介正在准备中。当前显示英语简介。

• Setting up an LLM server on NVIDIA GPUs
• Configuring FP8 or INT4 quantization
• Distributing large models across multiple GPUs

Use SAELens to train and analyze sparse autoencoders, which separate neural network activity into features that may represent recognizable concepts. Explore pretrained features, evaluate reconstruction quality, and investigate how individual features influence language model outputs.

中文简介正在准备中。当前显示英语简介。

• Exploring features with a pretrained SAE
• Training and evaluating a custom SAE
• Studying overlapping concepts in model activity