TinyClick โ€” ONNX export used by PredaCore

Float32 ONNX conversion of TinyClick (based on Microsoft Florence-2-base), as loaded by PredaCore's native vision runtime. PredaCore downloads these files on first launch and verifies every file against export_manifest.json (size and SHA-256) before use. The model itself was developed by the TinyClick authors.

Files

File Purpose
vision_encoder.onnx image encoder
encoder_model.onnx text/encoder graph
decoder_model_merged.onnx decoder with explicit self-attention cache
tokenizer.json tokenizer
export_manifest.json sizes, SHA-256 checksums and export parameters
notices/ upstream licenses, model card and source checksums

Provenance

Conversion changes: separate image / text / decoder graphs, an explicit decoder self-attention cache, and restoration of the checkpoint's untied output head. CPU parity with the source checkpoint was verified at export time (verified_cpu_parity: true in the manifest).

License

MIT, following the upstream model. notices/TINYCLICK-LICENSE.txt preserves the Samsung R&D Poland notice and notices/FLORENCE-LICENSE.txt preserves Microsoft's notice; notices/sources.json records where each was obtained, with checksums.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for shubhp23/predacore-tinyclick-onnx

Quantized
(1)
this model

Paper for shubhp23/predacore-tinyclick-onnx