support-new-model

To adapt a new LLM for Ascend hardware, you provide the model name and target vendor to analyze integration gaps and generate an end-to-end implementation plan via dlinfer.

Integrating a VLM with dlinfer involves gathering model information, analyzing capability gaps against the target backend, and outlining step-by-step implementation artifacts for the vendor-specific hardware.

Yes, models already supported by the LMDeploy CUDA backend can be expanded to MACA backends by using dlinfer to determine the necessary artifacts and implementation steps for the domestic chip ecosystem.

Yes, dlinfer supports model integration across vendor-specific backends including Ascend, Camb, and MACA by guiding information gathering and gap analysis for each target environment.

You need to provide the model name and target vendor to start the adaptation process, which then determines the necessary implementation artifacts and integration steps.

The adaptation process assumes the model is already supported by the LMDeploy CUDA backend; if not, you must ensure baseline compatibility before using dlinfer to generate the implementation plan for the Camb backend.