Commit History
27 days ago
-
Update Hugging Face org references: llm-semantic-router → vllm-sr
Verified signature290318f
-
Expand encoder retrieval coverage with matching exports and measured scores
Verified signatureUpdate the encoder and matching exports, refresh only product-page scores, preserve the opt-in AMD helper, and retain complete measured trade-offs in the separate evaluation evidence.972c180 -
Update encoder with additional Chinese retrieval training and matching exports
Verified signatureUpdate the encoder and matching exports, refresh only product-page scores, preserve the opt-in AMD helper, and retain complete measured trade-offs in the separate evaluation evidence.bc1093b
-
Add optional qualified AMD SDPA runtime helper
Verified signature9709cdd -
Update retrieval-distilled encoder and matching exports
Verified signatureUpdate the encoder and matching exports, refresh only product-page scores, and retain detailed evaluation evidence separately. Small classification, retrieval and long-document point gains, alongside clustering and representation-grid regressions, remain visible in the measured results.75c25cd -
Refine retrieval and multi-resolution embeddings
Verified signatureUpdate the encoder and matching exports, refresh only product-page scores, and retain detailed evaluation evidence separately. Long-document, clustering and cross-language trade-offs remain visible in the measured results.9952e61 -
Improve multilingual classification and long-document retrieval
Verified signatureUpdate the encoder and matching exports, refresh only product-page scores, and retain detailed evaluation evidence separately. Short-retrieval regressions remain visible in the measured results.27cf105 -
Restore product model card and update evaluation scores
Verified signatureRestore the original product copy and quick start. Replace only the Evaluation section with the verified current scores against mmBERT. Detailed evaluation records remain separate.604c75c -
Publish R3 interim encoder: classification and long retrieval gains with disclosed tradeoffs
Verified signatureAtomic native weights, tokenizer/config, eight standard/CK ONNX graphs and shared weights, model card, and complete bounded evaluation evidence. Native FP32 scores compare primarily with mmBERT. Short retrieval and early-exit regressions remain visible; this is not a full-MMTEB or universal-superiority claim.ae3a92a
-
Improve multilingual retrieval and semantic matching
Verified signature1e57ceb -
Improve multilingual retrieval and semantic matching
Verified signature3ef3575 -
Add ROCm-accelerated ONNX representations
Verified signaturea72bbb7
-
Add portable FP32 ONNX models
Verified signatureaa234e9 -
Add original mmBERT evaluation comparison for Vela Embedding
Verified signature4cd1406 -
Simplify Vela quick starts and model cards
Verified signatureb8d1dbd -
Release Vela on the unified Encoder base
Verified signature4761bf4 -
Unify Vela model cards with Semantic Router branding
Verified signature4af13a0
-
Release Vela 1.0 Embedding
Verified signatureMultilingual retrieval with flexible representations and long-context support.5e639f1 -
initial commit
Verified signature9e11614