Modelpublic

ericlmtn/Qwen3.8-27B-Steiner-Style-LoRA

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face

0likes26downloads

Model Card

Qwen3.8-27B Steiner Style LoRA

Private rank-16 LoRA adapter for prose-style generation with Qwen3.8-27B.

Base model

—Model: Qwen/Qwen3.8-27B
—Revision: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0

Use that exact base revision for reproducible loading and evaluation.

Adapter

—Format: PEFT LoRA
—Rank: 16
—Alpha: 32
—Dropout: 0
—Task: causal language modeling
—Maximum training sequence length: 8,192 tokens
—Training library: PEFT 0.20.0 with Unsloth

The adapter targets the model's full-attention projections, Gated DeltaNet projections, and language-model MLP projections.

Privacy and limitations

The training corpus is private and is not included in this repository. This adapter is intended for private evaluation and inference. It may reproduce phrases or stylistic patterns from its training material, and detector scores are not evidence of authorship.

Files

—adapter_model.safetensors: LoRA weights
—adapter_config.json: PEFT configuration and pinned base revision
—SHA256SUMS: integrity checksums