Capabilities
What the badges mean
| Badge | Meaning |
|---|---|
| Chat | General conversational / text generation. |
| Tools | Can participate in structured function/tool calling when model, template and runtime support it. |
| Vision | Can accept image input; local GGUF workflows may require a matching projector (mmproj) file. |
| Coding | Explicitly trained or tagged for software/code tasks. |
| Reasoning | Explicitly reasoning-oriented or supports a reasoning mode. |
| Embeddings | Produces numerical vector representations; not treated as a normal chat model. |
| Audio | Speech/audio capabilities where model and runtime support them. |
Where the labels come from
Badges are inferred from Hugging Face metadata: the pipeline tag (text-generation, feature-extraction, …), repository tags, and the model name. On the catalog this is marked as inferred from Hub tags.
Three evidence levels exist in the product model:
- Source-declared — the model card or tags declare the capability.
- Inferred — BotConnector derived it from name/tags/metadata.
- Runtime-verified — a load+probe confirmed the capability on your machine. (Not yet implemented; see status.)
Important limitation
important
How to verify a capability yourself
- Vision: load the model with its mmproj projector and send an image — Vision.
- Tools: send a
toolsarray to the API and check for a tool call in the response — Tool use. - Embeddings: call
POST /v1/embeddingsand confirm a vector is returned — Embeddings.
Missing badges
Absence of a badge is not proof of absence — some repositories are tagged sparsely on the Hub. Check the model card on huggingface.co for the authoritative description.