Captioning Models (experimental candidates)
Unmodified copies of permissively licensed upstream ONNX exports, staged by xberg for evaluation (xberg-io/xberg#2334, #2336). Experimental candidates, not qualified: no model here is an xberg default, a supported preset or a bundling decision.
| Directory | Upstream | Upstream revision | License | Task |
|---|---|---|---|---|
smolvlm-256m-instruct |
HuggingFaceTB/SmolVLM-256M-Instruct | 7e3e67edbbed1bf9888184d9df282b700a323964 |
Apache-2.0 | image captioning (vision_encoder + embed_tokens + decoder_model_merged) |
smolvlm-500m-instruct |
HuggingFaceTB/SmolVLM-500M-Instruct | a7da5b986cb59b408707209984f360a5f4ad7e47 |
Apache-2.0 | image captioning (vision_encoder + embed_tokens + decoder_model_merged) |
florence-2-base |
onnx-community/Florence-2-base | d59e079711c57174f29265539fb4cc9f0f335916 |
MIT | image captioning (vision_encoder + embed_tokens + encoder_model + decoder_model_merged) |
Every directory keeps the upstream LICENSE text (LICENSE), the upstream model card (UPSTREAM_README.md), tokenizer, processor and generation configuration, and all graphs the pipeline needs. manifest.json lists the SHA-256 and size of every file and the upstream revision it was copied from. Consumers must pin this repository's commit revision (recorded in crates/xberg/src/candidate_models/) and verify the SHA-256 values; do not resolve main.
Attribution: each model remains the work of its upstream authors under the license listed above. Notes per model:
smolvlm-256m-instruct: Official upstream ONNX export; fp32 and int8 graphs. Base models: SmolLM2 (Apache-2.0), SigLIP (Apache-2.0).smolvlm-500m-instruct: Official upstream ONNX export; fp32 and int8 graphs. Base models: SmolLM2 (Apache-2.0), SigLIP (Apache-2.0).florence-2-base: onnx-community export of microsoft/Florence-2-base (rev 5ca5edf5bd017b9919c05d08aebef5e4c7ac3bac); fp32 and int8 graphs; non-merged decoders omitted.