Skip to content

Model Sources

Xberg downloads ML models from the HuggingFace Hub on first use and caches them under the platform cache directory (~/.cache/xberg/ on Linux/macOS, %LOCALAPPDATA%/xberg/ on Windows), or under XBERG_CACHE_DIR when set.

Models that Xberg hosts itself live under the xberg-io organization. Everything else is pulled from its upstream publisher. ONNX weights are SHA256-verified after download where a checksum is pinned.

Every model below is downloaded from a checksum-pinned xberg-io repository with weights unmodified from upstream. All are permissively licensed (Apache-2.0 or MIT).

Capability Repository Notes
PaddleOCR — detection, classification, recognition (per-script + unified), text-line / document orientation xberg-io/paddleocr-onnx-models PP-OCRv5 and PP-OCRv6 ONNX exports.
Table structure — SLANeXt (wired/wireless), SLANet+, table classifier xberg-io/paddleocr-onnx-models v2/table/*, v2/classifiers/*.
Layout detection — RT-DETR, TATR, PP-DocLayout-V3 xberg-io/layout-models
PaddleOCR-VL 1.6 (candle vision-language OCR) xberg-io/paddleocr-vl-1.6 Mirror of PaddlePaddle/PaddleOCR-VL-1.6 (Apache-2.0). SigLIP vision encoder + Ernie-4.5 text decoder; tasks ocr, table, formula, chart.
Named-entity recognition (GLiNER) xberg-io/gliner-models xberg-managed span-mode ONNX exports and tokenizer files. Source model lineage is gliner-community.
Dense embeddings — gte-modernbert-base, arctic-embed-m-v2.0, qwen3-embedding-0.6b xberg-io/embedding-models Mirrors of Alibaba-NLP/gte-modernbert-base (Apache-2.0), Snowflake/snowflake-arctic-embed-m-v2.0 (Apache-2.0), and Qwen/Qwen3-Embedding-0.6B (Apache-2.0).
Dense embeddings — potion-base-8M (lightweight preset, model2vec) xberg-io/embedding-models (potion-base-8m/) Mirror of minishlab/potion-base-8M (MIT). Pure-Rust static backend — no ONNX Runtime.
Sparse embeddings — opensearch-v3-distill xberg-io/sparse-embeddings Mirror of opensearch-project/opensearch-neural-sparse-encoding-doc-v3-distill (Apache-2.0).
Reranking — ettin-reranker-150m, qwen3-reranker-0.6b xberg-io/reranker-models Mirrors of cross-encoder/ettin-reranker-150m-v1 (Apache-2.0) and Qwen/Qwen3-Reranker-0.6B (Apache-2.0). qwen3-reranker-0.6b is a custom causal-LM ONNX export (generative reranker head).
Late interaction (ColBERT) — gte-moderncolbert xberg-io/late-interaction-models Mirror of lightonai/GTE-ModernColBERT-v1 (Apache-2.0).

For GLiNER, Xberg downloads only the exported artifacts listed in xberg-io/gliner-models. If the repository is private or not publicly readable, configure Hugging Face credentials supported by hf-hub before warming the cache or running inference.

Capability Repositories
Embeddings Xenova/all-MiniLM-L6-v2, Xenova/bge-base-en-v1.5, Xenova/bge-large-en-v1.5, intfloat/multilingual-e5-base
Reranking BAAI/bge-reranker-base, rozgo/bge-reranker-v2-m3, jinaai/jina-reranker-v1-turbo-en
Sparse embeddings (SPLADE) prithivida/Splade_PP_en_v1
Late interaction (ColBERT) answerdotai/answerai-colbert-small-v1
Transcription (Whisper) onnx-community/whisper-tiny, onnx-community/whisper-base, onnx-community/whisper-small, Xenova/whisper-medium, Xenova/whisper-large-v3
Tokenizers (token counting / chunk sizing) Xenova/gpt-4o, thenlper/gte-small

You can point the reranker and embedding Custom presets at any compatible ONNX repository on the Hub; see the reranking guide.