-
Clean repo: remove non-research content, fix paths, update docs
Removed:
- convergent-ai.md (blog post, not research)
- MANIFEST.md (stale v27, contained Kaggle account details)
- TRAINING_PLAN.md (referenced internal .ava-private paths, stale 6.9M config)
- architecture.png + models/ (empty LFS pointers with no content)
Fixed:
- Hardcoded /home/adam/workspace in v27, v29 training scripts → env vars
- Hardcoded path in plug.py → env var
- .ava-lab internal paths in experiment logs → relative paths
- MODEL_CARD.md: 601M → 476M (actual trained params)
- MODEL_CARD.md: removed reference to deleted convergent-ai paper
- GLADIUS_TRAJECTORY.md: removed internal .ava-lab paths
README rewritten as clean model repo index.
-
GLADIUS v5.0 — Cognitive kernel with Synthase depth attention, PUP uncertainty, Memory V2, multi-tokenizer architecture
