Sentence Similarity
ONNX
Safetensors
English
ogma
embeddings
dense-retrieval
matryoshka
rag
agents
mteb
semantic-search
text-embeddings
text-embedding
vector-search
document-retrieval
similarity-search
classification
clustering
edge-ai
on-device
local-inference
efficient-ai
rag-retrieval
custom_code
Eval Results (legacy)
Commit History
-
Remove coreml/ (the permissive open-ogma CoreML build sat under this repo's CC-BY-NC banner); point at the MIT successor
Verified signature33a8404
-
Add CoreML .mlpackage bundle for Cliphoard on-device inference
Verified signature080bcc7
-
docs: remove campaign-notes link (repo private)
Verified signature9d532ac -
docs: append MTEB(eng, v2) full-41 results (2026-07 eval campaign)
Verified signature5e32aa9
-
Upload README.md with huggingface_hub
SamVerified signaturec418f5f -
Upload README.md with huggingface_hub
SamVerified signaturec43d2c2 -
Upload README.md with huggingface_hub
SamVerified signature6d7c3fc -
Upload README.md with huggingface_hub
SamVerified signature8e7c584 -
fix(docs): retrieval uses task=qry for both queries and docs
Empirical sweep (commit 5aa2fe6) shows (QRY, QRY) is best on retrieval. The published registry scores were measured under (QRY, QRY). Use that.761deba -
docs: remove CPU inference benchmark table
9294030 -
fix(docs): correct retrieval example to symmetric encoding
Empirical sweep (commit 5aa2fe6) showed (QRY, DOC) asymmetric routing underperforms symmetric (SYM, SYM) on these checkpoints by ~4pp. Updated Usage to show task="sym" for both queries and documents in retrieval, and migrated examples to AutoModel + AutoTokenizer.4254e91 -
Add OgmaTokenizerFast + model.embed() high-level API
- tokenization_ogma.py: PreTrainedTokenizerFast subclass that shifts content token ids by N_SPECIAL=7 (training-matched tokenization, no manual offset needed by callers) - tokenizer_config.json: wires AutoTokenizer to OgmaTokenizerFast - ogma_model.py: adds TaskToken class attr + embed(texts, task=) method so callers need only AutoModel + AutoTokenizer, no sys.modules digging57aafa4
-
Remove hedging from public model card
Hephaestus9c3f997 -
Clean public model card provenance wording
Hephaestus7b0f31c -
Fix MTEB score and comparison hygiene
Verified signaturea43fe04
-
Enable AutoModel loading
Verified signature8540761 -
Enable AutoModel loading
Verified signature6682fef -
Enable AutoModel loading
Verified signature4636da5 -
Enable AutoModel loading
Verified signature3ee9a30 -
Enable AutoModel loading
Verified signatured14c9aa -
Enable AutoModel loading
Verified signature7f488b2 -
Use text embedding discovery tags
Verified signatureb3b7d58 -
Expand use cases and explain Ogma name
Verified signature1bdd461 -
Fix ogma-large MTEB status in model cards
Verified signature3dbc213 -
Add ONNX export with parity report
Verified signature285dd13 -
Add ONNX export with parity report
Verified signature186a356 -
Improve model card discoverability for embedding search
Verified signature75df39c
-
Expand safety eval: full toxicity + prompt injection tables (all classifiers, all metrics)
Verified signaturee33b8f6 -
Add comprehensive model card with MTEB benchmarks, MiniLM comparison, CPU inference benchmarks, and safety evaluation
Verified signature935b789 -
chore: update license from MIT to CC-BY-NC-4.0
Verified signaturefb388ad -
Fix: relative imports for Hub loading, corrected model cards with accurate numbers and usage examples
Verified signatureea854a7 -
Initial upload: ogma-small embedding model
Verified signaturee405320 -
initial commit
Verified signature0e8f50f