Deploy and train LLM and VLM systems with version-aware configuration and validation.
npx skills add https://github.com/black-yt/skills --skill llm-deploy-training
Please help me install this Agent Skill. Skill: llm-deploy-training Source: https://github.com/black-yt/skills/tree/main/llm-deploy-training Command: npx skills add https://github.com/black-yt/skills --skill llm-deploy-training
This Skill helps you deploy, validate, troubleshoot, and train large language and vision-language models without breaking shared environments or relying on guesswork.
Ask for a version-aware deployment or training plan for your target model, including the correct vLLM or ms-swift commands, validation checks, and troubleshooting steps.
High-intent search queries and answers about installing and using this skill.
To deploy a multimodal LLM with vLLM, configure OpenAI-compatible endpoints with correct multimodal limits, tool calling, reasoning controls, and CUDA Graph settings, followed by remote access validation to ensure safe serving.
Configuring ms-swift for SFT and DPO requires version-aware hyperparameters, full-training settings, loss masking verification, long-context handling, and checkpoint hygiene to train models safely without breaking shared environments.
To verify loss masking and avoid OOM errors during LLM training, use dry runs plus logs to catch parser issues, masking errors, and memory limits early, validating labels and loss scales before executing full runs.
Yes, vLLM supports Qwen tool calling and reasoning control for multimodal inference by applying version-aware configuration and validation to manage CUDA Graph settings and multimodal limits correctly during serving.
You need version-aware configuration for LLM deployment and training to prevent unsafe environment edits, ensure compatibility across vLLM and ms-swift workflows, and validate hyperparameters for safe shared-environment usage.
Setting up GRPO workflows with ms-swift requires applying version-aware hyperparameters, full-training settings, and strict checkpoint hygiene to ensure safe shared-environment usage and prevent data loss during training.
ovachieverCurate multimodal LLM training data with GPU-accelerated filtering and deduplication.
ScientiaCapitalExport fine-tuned models to GGUF and deploy with vLLM or Ollama.
ovachieverDeploy OpenAI-compatible LLM APIs with vLLM for high-throughput inference.
FacuMTrain and fine-tune language models on Hugging Face Jobs with TRL.
