cloudrift-aiOfficial·6 Agent Skills Included

emmy

Compile, benchmark, and deploy LLMs on any GPU

Compiles PyTorch models into fused CUDA kernels, autotunes them per GPU, and deploys LLM serving stacks via Docker locally, over SSH, or on cloud VMs. Eliminates manual kernel tuning, guesswork in GPU sizing, and slow stock inference engines. Includes guided workflows for model discovery, onboarding, benchmarking, and publishing prebuilt serving images.

75|8

↗

npx skills add cloudrift-ai/emmy --all -g -y

Available:

Instructs the agent on Emmy's architecture, CLI commands, environment variables, and testing rules so it can route user requests to the correct compile, tune, deploy, or benchmark workflow.

All Skills in This Repository (6)

Pure Emerald Level Indicators

📦 In Repo

cloudrift-ai

tune-model

Autotune compiled LLM kernels and generate root-cause performance findings reports.

collect-node-data

Automate per-GPU kernel node data collection and merging for LLM compiler autotuning.

start-remote-server

Provision cloud GPU VMs on GCP or CloudRift for emmy compiler workloads.

benchmark-new-model

Automate onboarding of HuggingFace models into emmy with deployment validation and benchmarking.

tune-golden

Re-tune and validate GPU matmul golden configuration YAMLs via the emmy CLI.

discover-new-models

Rank newly released open-weight LLMs by demand and VRAM compatibility.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install Emmy?▼

Run `npx skills add cloudrift-ai/emmy --all -g -y` in your terminal to install all skills in this suite globally.

What does Emmy do for LLM inference?▼

Emmy compiles PyTorch models into fused CUDA kernels, autotunes them per GPU, and benchmarks them against eager and torch.compile. It can outperform stock vLLM on supported models.

Can Emmy deploy models to cloud GPUs?▼

Yes. It provisions GPU VMs on CloudRift or GCP, deploys models via Docker Compose, runs benchmarks, and tears everything down automatically.

How do I add a new model to Emmy?▼

The discover-models and onboard-model skills find trending open models, check VRAM fit, create a serving recipe, and benchmark it on your target GPU.

Do I need compiler expertise to tune kernels?▼

No. The tune-kernels skill runs a guided hybrid search that proposes, measures, and verifies kernel configurations on your GPU automatically.