Zero-Shot Classification
Safetensors
PEFT
English
openjev
classification
decision-model
listwise
gemma4
research
Instructions to use bambamdevs/openjev-e4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use bambamdevs/openjev-e4b with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
File size: 1,076 Bytes
03223d7 | 1 2 3 4 5 6 7 8 | # Checkpoint provenance for OpenJEV E4B 1.0
The public 1.0 release uses the checkpoint identified internally as `v1.6 R7`. `RELEASE_METADATA.json` binds that source checkpoint to the public release, the pinned Gemma base, and the hashes of the copied files.
`checkpoint_json/arena_manifest.json` records the Arena generator, seeds, families, and generated-set hashes. `checkpoint_json/arena_training_meta.json` records the selected checkpoint's gate, score (0.024174), preservation flags, frontier, family weights, and training settings. The gate stored flags and the final score, but not every component value.
Other `checkpoint_json/` files describe earlier development stages. The public copies of `model/openjev_config.json` and `checkpoint_json/v151_selection.json` drop internal labeling notes; no settings, gates, or scores were changed. Files under `inherited_v132/` are inherited qualification records and are not evidence of the selected checkpoint's public evaluation. The Experiment 1 whitepaper explains the model's evolution and public evaluation separately.
|