mvp-ai-lab/mvp-engine
24 skills
$npx skills add mvp-ai-lab/mvp-engineSkillInstalls
experiment-analysisAnalyze mvp-engine experiment outputs and produce a structuredfsdp2-prefetchingAdd, review, update, and validate recipe/model-local FSDP2gradient-checkpointingAdd, review, update, and validate gradient checkpointing throughliger-kernelDecide where and how to wire Liger Kernel into an MVP-Engine recipeliger-kernel-kitUse LigerKernelKit for reusable Liger Kernel integration before modelloss-spike-guardAdd, review, update, and validate loss spike guards for scalar-loss or per-token-loss training, including config thresholds, micro-batch skip wiring,…mfu-kitUse MFUKit for global model FLOPs utilization logging, includingmllm-data-kitUse and extend the MVP-Engine MLLM data kit with explicit specs,mllm-model-kitUse and extend MLLMModelKit for multimodal model loading, recipemodel-compileAdd, review, update, and validate torch model.compile support usingmodel-flops-utilizationAdd, review, update, and validate model FLOPs utilization usingmodel-migrationAdd, review, update, and validate migration of external model/configuration code into mvp-engine recipes with strict behavior parity, checkpoint compatibility,…new-recipe-templateCreate a new recipe scaffold under recipes/ with standard config,optim-kitUse OptimKit to build optimizers over trainable parameters andpr-feedbackTriage reviewer feedback on an open PR, implement scoped fixes,pr-gateRun a pre-PR or pre-push quality gate by inspecting the branch diff,recipe-merge-repairMerge an upstream branch into a recipe development branch bysequence-parallelAdd, review, update, and validate recipe-local sequence-paralleltensor-parallelAdd, review, update, and validate recipe-local tensor-paralleltoken-loss-kitUse TokenNormedLossKit for unreduced per-token loss patching,token-normalized-lossAdd, review, update, and validate token-normalized training lossvlm-data-pipelineAdd, review, update, and validate VLM/MLLM data pipelines usingvlm-freeze-policyAdd, review, update, and validate VLM freeze policy usingvlm-packingAdd, review, update, and validate VLM/MLLM packing behavior around