jaswanthsanjay88/rev-qwen3-4b

A Rev-style prefill-only decision model based on Qwen/Qwen3-4B-Base.

Features

  • Single forward pass: Parallel question evaluation with zero autoregressive decoding.
  • Prefix KV-Caching: Instant repeated evaluation on long document states.
  • Pointer readout head: Bilinear projection over option boundaries.
  • TypeSafe System One API: Native compatibility with /v1/systemone.

Files included

  • adapter_model.safetensors: Trained LoRA adapter
  • head.pt: PointerHead weights and base model metadata
  • Tokenizer configurations
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for jaswanthsanjay88/rev-4b

Adapter
(90)
this model