Loading...
Loading...
Found 110 Skills
Generate images and videos using fal.ai AI models. Production-grade catalogue covering Flux, SDXL, ideogram, and other community-hosted endpoints.
ONLY for OpenAI Triton (@triton.jit) kernel development. NEVER use for CUDA C++ kernels, TileIR, or profiling tools (ncu, nsys). The user's request must involve Triton explicitly. Covers Triton-specific patterns: fused elementwise, reductions (softmax, LayerNorm, RMSNorm), tiled GEMM with triton.autotune, and flash attention. Workflow: design, write, verify (with fast-path for explicit requests).
Use when importing a new model architecture into MAX from a Hugging Face model ID. Triggers on: "import a model into MAX", "add model to MAX", "bring up <HF model> in MAX". Workflow: inspect Hugging Face config and modeling code, scaffold from a similar MAX architecture, implement each graph layer to match HF, serve, then debug against the Hugging Face reference until outputs match.
Minimal text generation smoke test for Model Studio Qwen text models.
Guide for adding support for new LLM or VLM models in Megatron-Bridge. Covers bridge, provider, recipe, tests, docs, and examples.
Trigger this skill when building applications with Gemma or for general knowledge inquiries related to Gemma models (e.g. prompt structure, capabilities). Covers model selection, development workflows, and deployment best practices.
INVOKE THIS SKILL when creating, running, or analyzing Arize experiments. Covers experiment CRUD, exporting runs, comparing results, and evaluation workflows using the ax CLI.
Install, audit, repair, and manage a loopback-only CLIProxyAPI bridge that registers subscription-backed Codex, Grok, Antigravity/Gemini, or newly added CLI models in WorkBuddy. Use whenever the user asks to connect a local CLI or agent model to WorkBuddy, install CLIProxyAPI, add or repair a WorkBuddy custom model, preserve image/tool/reasoning/Fast capabilities, onboard another CLI provider, or diagnose a WorkBuddy model that stopped working. This Skill is macOS-first and should be used even when the user names only the model or says their local WorkBuddy API has expired.
Model-family-specific prompt craft for fal.ai endpoints. Trigger when the user mentions a specific model family by name and asks how to prompt it ("how do I prompt Kling", "GPT Image 2 prompt structure", "Happy Horse tips"), or when prompts to a routed endpoint keep coming back generic and the family's known nuances should be applied. For endpoint selection ("which model for X"), use `fal-models-catalog` instead. This skill is about how to talk to a model once it has been chosen.
Read this BEFORE launching any subagent (Task tool, background agents, parallel agents, best-of-N, delegating work to another agent). Hard model rules for subagents plus consensus principles for using them well. Triggers: launch a subagent, spawn agents, run agents in parallel, delegate to a subagent.
Selects, deploys, and customizes AI models on Amazon SageMaker. Fine-tuning (SFT, DPO, RLVR, RLAIF), model selection, dataset preparation, evaluation, deployment to SageMaker endpoints or Bedrock, and endpoint diagnostics. Covers the full lifecycle from planning through production. Use when fine-tuning models on SageMaker, selecting base models from SageMaker Hub, finding a model to deploy without fine-tuning, transforming datasets for training, checking data readiness, evaluating model quality, deploying to endpoints, setting up IAM roles and S3 buckets for training jobs, or managing a SageMaker Managed MLflow app. Also use to check endpoint health, diagnose failures, debug latency or errors, or view container logs and CloudWatch metrics. Covers Serverless Model Customization, Nova and OSS deployment paths, and PySDK v3 usage. NOT for Ground Truth labeling, Feature Store, or general-purpose AWS infrastructure.
Convert evaluation traces and production logs into SFT examples and preference pairs. Use when graded traces or failure examples exist and need to become training data, when applying rejection sampling to model outputs, or when building DPO pairs from passing and failing runs.