Integrate new LLMs or VLMs into LMDeploy's PyTorch backend.
npx skills add https://github.com/InternLM/lmdeploy --skill support-new-model
Please help me install this Agent Skill. Skill: support-new-model Source: https://github.com/InternLM/lmdeploy/tree/main/.claude/skills/support-new-model Command: npx skills add https://github.com/InternLM/lmdeploy --skill support-new-model
This guide enables developers to extend LMDeploy by adding a new LLM or VLM to the PyTorch backend, streamlining the end-to-end integration from model code to deployment configuration.
Follow the steps in the guide to add a new model to LMDeploy and verify it loads correctly.
High-intent search queries and answers about installing and using this skill.
To add a new LLM or VLM to LMDeploy, you create a model file, register it in the module_map, optionally supply a non-standard HF config builder, and add quantization mappings to ensure loadability and execution.
Integrating a VLM into LMDeploy requires creating the model file, registering the architecture in module_map, and adding necessary VLM preprocessors to ensure the model is configurable and executable for production deployment.
Yes, LMDeploy supports optional non-standard HF config builders during model integration, allowing you to properly configure and load new architectures alongside their specific weight loading requirements.
Yes, you can add quantization mappings when onboarding new LLM or VLM architectures into LMDeploy, ensuring the newly registered models are loadable and executable within your production deployment pipeline.
You need to register a model in the module_map when adding a new LLM or VLM to LMDeploy, which bridges the new model file with the PyTorch backend to enable proper architecture registration and weight loading.
open-thoughtsPublish HF-only SFT cells to Hugging Face without database registration.
DeepLink-orgGuide new LLM or VLM adaptation to dlinfer on domestic hardware.
bensonlee5Monitor robotic task progress with Qwen3-VL-4B VLM zero-shot reward model.
eyadsibaiDeploy and monitor machine learning models in production with Docker and Kubernetes.
