omnijev-work is an open-weight model from Fnruha0921. Its published files total 912.0 MB.
Code, logs, results and training checkpoints of OmniJev / MM-Jev: Jev-style typed decisions (noul / choice / score) on Gemma 3n E4B with text, image, audio and video states.
Model Card
Code, logs, results and training checkpoints of OmniJev / MM-Jev: Jev-style typed decisions (noul / choice / score) on Gemma 3n E4B with text, image, audio and video states. This is a scratch workspace synced by the training job; the finished model is published separately. code/ – training / evaluation code (MM-Jev architecture: option tree in one pass, KV-share truncation, modality exit + latent memory, sibling bridge, MatFormer width slicing) run/ – pipeline log, results.json, checkpoints (LoRA + decision heads + latents) Base model: google/gemma-3n-E4B-it (Gemma terms of use apply to derivatives).
Excerpt from the card by Fnruha0921.
Identity and Version
- Repository
- fnruha0921/omnijev-work
- Publisher
- Fnruha0921
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- d1ecf475d768172db7b1d838203d79a864da7533
- First published
- 2026-09-28
- Last updated
- 2026-09-29
Files and Weights
41 files, 912.0 MB in total. The weights are 4 files totalling 908.9 MB in pt.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| run/adapter_final.pt | Weights | 113.7 MB | 631ea72a09f8 |
| run/ckpt/last.pt | Weights | 340.7 MB | c9f79ae0102f |
| run_mm/adapter_final.pt | Weights | 113.7 MB | b717a61c3433 |
| run_mm/ckpt/last.pt | Weights | 340.9 MB | 007f26df63ad |
| code/build_data.py | Configuration | 21.2 KB | — |
| code/build_mm_hf.py | Configuration | 15.6 KB | — |
| code/build_text2.py | Configuration | 7.1 KB | — |
| code/build_text3.py | Configuration | 6.2 KB | — |
| code/build_text4.py | Configuration | 10.0 KB | — |
| code/build_text5.py | Configuration | 11.8 KB | — |
| code/demos.py | Configuration | 16.1 KB | — |
| code/media.py | Configuration | 2.7 KB | — |
| code/mmjev.py | Configuration | 44.1 KB | — |
| code/mmjev_eval.py | Configuration | 9.4 KB | — |
| code/pipeline.py | Configuration | 17.0 KB | — |
| code/train.py | Configuration | 16.5 KB | — |
| run/results.json | Configuration | 8.2 KB | — |
| run/state.json | Configuration | 489 B | — |
| run_mm/results.json | Configuration | 20.0 KB | — |
| run_mm/state.json | Configuration | 390 B | — |
| README.md | Documentation | 692 B | — |
| code/vast_entry.sh | Other | 2.1 KB | — |
| run/pipeline.log | Other | 562.4 KB | — |
| run_mm/assets/aokvqa_1.png | Other | 309.7 KB | 1b7434cbf703 |
| run_mm/assets/aokvqa_2.png | Other | 410.6 KB | 286fa34be77e |
| run_mm/assets/esc50_choice_1.png | Other | 46.3 KB | — |
| run_mm/assets/esc50_choice_2.png | Other | 48.4 KB | — |
| run_mm/assets/mvbench_mini_1.mp4 | Other | 154.1 KB | 3a06ae503bf5 |
| run_mm/assets/scienceqa_img_1.png | Other | 92.5 KB | — |
| run_mm/assets/snake_seed11.mp4 | Other | 258.1 KB | fbe43df7fb1a |
| run_mm/assets/snake_seed21.mp4 | Other | 252.4 KB | 3361b7479e93 |
| run_mm/assets/snake_seed3.mp4 | Other | 53.8 KB | — |
| run_mm/assets/speech_commands_1.png | Other | 59.6 KB | — |
| run_mm/assets/tempcompass_1.mp4 | Other | 212.5 KB | 764542e9db1a |
| run_mm/assets/tempcompass_2.mp4 | Other | 222.2 KB | cb87af444c47 |
| run_mm/assets/ucf101_1.mp4 | Other | 86.0 KB | — |
| run_mm/assets/urbansound8k_1.png | Other | 88.6 KB | — |
| run_mm/pipeline.log | Other | 20.3 KB | — |
| run_mm/setup.log | Other | 1.2 KB | — |
| run_mm2/pipeline.log | Other | 985 B | — |
| .gitattributes | Repository | 2.0 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 908.9 MB
Released by Fnruha0921 through its official repository on Hugging Face.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 908.9 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.