arjunkshah21
/

sc-keep-crossencoder-v4-large

Text Classification

context-compression

text-embeddings-inference

Model card Files Files and versions

sc-keep-crossencoder-v4-large

Production Neural Keep weights for SuperCompress.

This is the model currently served on the hosted API (api.supercompress.dev → Fly sc-neural-keep).

Architecture: ModernBERT-large cross-encoder (query-conditioned line keep)
Task: keep / drop each context line given the current user query
Weights: model.safetensors (1,583,351,632 bytes)
License: MIT

Inference threshold

See sc_meta.json (inference_threshold). Hosted Fly default may override via SC_NEURAL_THRESHOLD.

Related

Distill (smaller, not production): arjunkshah21/sc-keep-crossencoder-v4-base-distill-v2
Product: https://supercompress.dev

Downloads last month: 45

Safetensors

Model size

0.4B params

Tensor type

F32

·

Model tree for arjunkshah21/sc-keep-crossencoder-v4-large

Base model

answerdotai/ModernBERT-large

Finetuned

(413)

this model