Supermix Omni Collective V4 Frontier

Custom PyTorch checkpoint for the omni_collective_v4 frontier model.

Included files

omni_collective_v4_frontier.pth
omni_collective_v4_frontier_meta.json
omni_collective_v4_frontier_summary.json
omni_collective_v4_model.py
omni_collective_model.py
image_feature_utils.py
image_recognition_model.py
math_equation_model.py
train_omni_collective_v4.py

Model summary

Parameters: 19032281
Stage 1 rows: 8589
Stage 2 rows: 9447
Stage 2 validation score: 0.5176
Stage 2 intent accuracy: 0.8195
Stage 2 response accuracy: 0.1402
Stage 2 vision accuracy: 0.9020
Stage 2 domain accuracy: 0.7765

Training sources

3d: 190
bible: 180
books: 620
coding: 380
creative: 940
creative_100k: 280
delta_anchor: 171
delta_official: 32
dictionary: 180
finnegans: 120
hybrid: 860
quality_anchor: 2
qwen_teacher_league_total: 36
reasoning: 1400
science: 160
science_image: 420
science_novel: 120
supermix_plus: 2200
video_contact: 248
world_events: 66

Notes

This is a custom checkpoint, not a standard Transformers from_pretrained model. It adds sparse expert routing, per-depth modality-context routing, deeper memory, wider multimodal data, and Qwen teacher-league repair rows on top of the earlier omni line.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support