101 skills · sorted by relevance
Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.
The English summary is being prepared. Showing the original description.
Set up large language model pretraining with TorchTitan and PyTorch. Configure training across multiple GPUs, combine four ways of splitting the workload, enable Float8 calculations on H100 hardware, and manage checkpoints for saving and resuming training.
Configure large language model pretraining with torchtitan across multiple GPUs. Covers ways to split training work, launch jobs on computing clusters, enable Float8 numerical processing, and save or resume training through distributed checkpoints.
Configure large language model pretraining with torchtitan across multiple GPUs. Covers ways to split training work, launch jobs on computing clusters, enable Float8 numerical processing, and save or resume training through distributed checkpoints.
Improve CircleCI build and test pipelines by identifying slow or unreliable jobs, then adjusting caching, data sharing between jobs, triggers, and parallel execution. Receive incremental configuration changes with expected effects, validation criteria, and rollback guidance.
Plan how to split AI models and workloads across multiple GPUs. Compare memory needs, computation, communication, and waiting time to estimate completion times and choose a layout suited to response time, throughput, or cost goals.
Compare computing devices for a defined AI model workload using memory capacity, computing speed, data transfer, energy use, and cost. Build a comparison that separates predicted performance from measurements and identifies the constraints that determine the choice.
Edit English and Chinese prose to reduce formulaic AI writing patterns, including filler, inflated claims, and vague attribution. Preserves the original meaning and intended tone while improving sentence variety, directness, and the writer’s voice.
Helps a QwenPaw Worker choose between direct work, short parallel checks, and temporary agents with custom roles. It guides dependent task execution, shared files, progress updates, result merging, and removal of temporary agents after the work ends.
Write and repair custom MSBuild targets, the steps that organize build tasks. Preserve existing build sequences, choose appropriate insertion points, distinguish returned results from rebuild checks, and register generated files for cleanup.
Establish repeatable performance measurements for MSBuild and .NET builds before optimizing. Compare clean-output, changed-source, and unchanged builds, record timing ranges, and assess output layouts, dependency references, and parallel build options against the measured baseline.
Diagnose why MSBuild builds across multiple projects fail to run efficiently in parallel. Use build logs and project dependencies to identify blocking chains, then recommend changes to references, scheduling, or parallel execution settings.
This skill should be used when the user asks to "remove AI writing patterns", "humanize this text", "make this sound more natural", "remove AI-generated traces", "fix robotic writing", or needs to eliminate AI writing patterns from prose. Supports both English and Chinese text. Based on Wikipedia's "Signs of AI writing" guide, detects and fixes inflated symbolism, promotional language, superficial -ing analyses, vague attributions, AI vocabulary, negative parallelisms, and excessive conjunctive phrases.
The English summary is being prepared. Showing the original description.
Operational guide for enabling hierarchical context parallelism in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
The English summary is being prepared. Showing the original description.
Guide for selecting and configuring distributed training strategies in NeMo AutoModel, including FSDP2, Megatron FSDP, DDP, and parallelism settings.
The English summary is being prepared. Showing the original description.
Techniques for reducing peak GPU memory in Megatron Bridge — expandable segments, PEFT + SP input re-gather, parallelism resizing, activation recompute, CPU offloading constraints, and common OOM fixes.
The English summary is being prepared. Showing the original description.
Recommend and customize Megatron Bridge library and benchmark recipes for a user's model, GPU count, hardware, sequence length, and pretrain/SFT/PEFT goal. Use when selecting a starting recipe, comparing library and benchmark configs, resizing parallelism for a GPU allocation, or distinguishing convergence changes, semantics-preserving execution tuning, and benchmark-only shortcuts.
The English summary is being prepared. Showing the original description.
Use this skill when baseline NeMo Relay instrumentation exists and the user wants to configure or evaluate adaptive plugin behavior, including telemetry, state, adaptive_hints, tool_parallelism, ACG, hint consumption, or measured rollout.
The English summary is being prepared. Showing the original description.
How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.
The English summary is being prepared. Showing the original description.
Optimize Anima code generation performance with caching, parallelism, and output tuning. Use when reducing generation latency, optimizing batch component generation, or improving generated code quality for production use. Trigger with: "anima performance", "anima slow", "anima optimization", "anima caching".
The English summary is being prepared. Showing the original description.
Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases. Credits: Original skill by @blader - https://github.com/blader/humanizer
The English summary is being prepared. Showing the original description.
Analyze and, when the task authorizes source optimization, refactor a concrete long-running Python entrypoint containing repeated independent work. Use for explicit Python speedup or parallelization requests and evidence-backed long-running entrypoints; skip ordinary Python edits, test execution alone, short scripts, and projects that merely contain Python files.
The English summary is being prepared. Showing the original description.
Optimize MATLAB design files for GPU Coder to generate faster CUDA code. Iteratively profiles, rewrites, and benchmarks until performance targets are met or diagnostics are resolved. Use when asked to: optimize for GPU Coder, improve GPU codegen performance, profile generated GPU/CUDA code, profile GPU MEX, fix gpuPerformanceAnalyzer diagnostics, speed up GPU MEX, reduce GPU memory transfers, improve kernel parallelism, rewrite MATLAB for CUDA, or run gpuPerformanceAnalyzer.
The English summary is being prepared. Showing the original description.
Deploy and manage vLLM for high-throughput LLM inference. Configure continuous batching, tensor parallelism, quantization, and OpenAI-compatible API endpoints for production LLM serving.
The English summary is being prepared. Showing the original description.