Model Certification
This page explains what model certification means and how a model becomes certified. For the models certified on specific hardware, refer to Certified Models by Hardware.
Certification Overview
A certified model is a large language model (LLM) that Spectro Cloud has validated to run correctly on the listed GPU configuration for PaletteAI Inference Launchpad. This validation is based on Spectro Cloud's own testing rather than public benchmarks, and it can extend across equivalent GPU configurations. When you deploy a certified model on a configuration where it is certified, you can be confident it loads and serves requests without hardware surprises.
Model Use Cases
Spectro Cloud chooses the certified models with coding assistants as the primary use case. If your primary need is a coding assistant, start with the certified models.
Other models may serve other use cases better, so if your primary need is not a coding assistant, contact Spectro Cloud to discuss the right models for your use case.
The certified list is not exclusive. You can bring your own model as long as it fits within the GPU resources available on your appliance. Refer to Bring Your Own Model. If you want Spectro Cloud to certify a model for your hardware, contact Spectro Cloud.
For how to add a certified model to a running appliance, refer to Upload a Model and Deploy a Model. To bring a model that is not certified, refer to Bring Your Own Model. To let a certified text-only model answer questions about images, refer to Enable Vision Preprocessing.
Certification Compared with Bringing Your Own Model
Both paths place a model on the same appliance and serve it with the same inference engines. What differs is who writes the serving recipe and who establishes that the model works on the hardware.
| Aspect | Certified model | Your own model |
|---|---|---|
| Metadata file | Spectro Cloud authors it, and you download it from Artifact Studio. | You author it. |
| Hardware fit | Validated on the GPU configurations listed in Certified Models by Hardware. | You state what the model needs, and you confirm it fits your GPUs. |
| Serving configuration | Tuned per GPU configuration by Spectro Cloud. | You choose the inference engine, the tensor-parallel width, and the context length. |
| Validation | Spectro Cloud tests that the model loads and serves on the listed configurations. | You test the model on your hardware. |
Bringing your own model does not change how the appliance handles the model once it is serving. The gateway routes requests, meters tokens, and enforces quotas the same way it does for a certified model.
How Certification Differs from Model as a Service
PaletteAI Inference Launchpad is not a Model as a Service. Rather than offering the widest possible catalog of models on demand, Spectro Cloud certifies a focused set of models for on-premises inference on your own hardware. For multi-tenant AI factory use cases, PaletteAI is the better fit. For a full comparison of the two products, refer to What is PaletteAI Inference Launchpad?.