Meddies/meddies-pii-v2-onnx
Token Classification • Updated
Meddies development model candidates. Production pins exact revisions and digests during preload. Git stores no weights.
Note Selected for /v1/embedding. Canonical RAG model is jinaai/jina-embeddings-v5-text-small-retrieval; Q8 parity and EC2 latency remain gated.
Note Candidate for /v1/ocr; Q8 artifact and prompt contract remain quality and CPU benchmark-gated.
Note Candidate for /v1/asr; exact quantization and OpenMOSS prompt contract remain quality and CPU benchmark-gated.
Note Selected canonical model for /v1/rerank. CPU production uses native Transformers; this GGUF stays catalog-only until upstream llama.cpp support.