metadata
license: apache-2.0
library_name: transformers
tags:
- token-classification
- text-compression
- context-compression
- modernbert
- yscompress
YSCompress v1
YSCompress v1 is an internal context/token compression model package.
Intended use
This package is intended for context compression and token-level keep/discard scoring in Decompute/Recompute inference systems.
Loading
from transformers import AutoConfig
cfg = AutoConfig.from_pretrained("decompute/yscompress-v1", token="hf_...")
print(cfg.model_type) # modernbert
Runtime note
model_type is intentionally retained as modernbert for Transformers compatibility. The deployment-facing architecture alias is YSCompressor.
Package contents
The package includes model weights, tokenizer assets, model configuration, optional ONNX assets if present, license, and third-party notices.