agurung/colar-gemma-3-4b-ff-sft

CoLAR Gemma 3-4B Flawed Fictions SFT

This repository stores CoLAR exports in a Hugging Face-compatible layout. The repo root works for standard Transformers loading, and extra_state.pt preserves the latent head for latent decoding.

Current Revision

• Current tag: last-ew6ob26g
• Stage: supervised fine-tuning final checkpoint
• Task: Flawed Fictions continuity error detection
• Compare slug: gemma3_colar_sft_last_ew6ob26g

Tagged Checkpoints

Tag	Local reference	Status
best-epoch02-step35-val_loss=2.5829	best step35 export	tagged checkpoint
last-ew6ob26g	canonical last export	current commit

Files

• HF model files at repo root for standard decoding
• extra_state.pt for CoLAR latent decoding
• export_meta.json from the local export
• latent_metadata.json with archival provenance

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('agurung/colar-gemma-3-4b-ff-sft', revision='last-ew6ob26g', torch_dtype='auto', device_map='auto')
tokenizer = AutoTokenizer.from_pretrained('agurung/colar-gemma-3-4b-ff-sft', revision='last-ew6ob26g')

For latent decoding, download the same revision and use extra_state.pt together with the repo root model files.

Notes

• This is the canonical Gemma3 CoLAR post-SFT export used in the ablation materials.