Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

feat: rebuild static_frontend from latest Next.js source

Merge branch 'main' of https://github.com/razancodes/Meta-Pytorch-Hackathon

fix: parsing params

fix: A100 compliant v2

fix: A100 compliant

fix: A100 compliant

switched model to llama8B

converted everything to trl grpo + openenv

perf: A100 optimizations — BF16, dual-model loading, larger batches

fix: add root / → /web/ redirect for HF Spaces (OpenEnv SDK has no root route)

feat: AGUI live state polling, rebuild static frontend with API fix

fix: preserve LR scheduler across Launderer model swaps

fix: include static_frontend/ for HF Spaces /web demo

deployment fixes + frontend integration (alpha) + cleanup

chore: remove deprecated files, update all documentation

fix: batch normalization, EMA baseline, investigation bonuses, entity regex

fix(launderer): stop truncating response_text so scoring loop can parse JSON

fix: PPO stagnation - launderer diversity, reward noise, KL logging, terminal dedup

Updated TRAINING.md + README.md + PROJECT_CONTEXT.md

Updated TRAINING.md

feat: add L4-optimized Colab self-play training script

fix: final audit — orchestrator scores, KL direction, grader typology alias

fix: get_scenario() always produces suspicious scenarios (force_clean=False)

debug: flush __pycache__ at test start, add grader breakdown output

fix: terminal actions skip grade_step to prevent reward double-count

fix: don't roll clean_ratio when typology is explicitly requested

feat: self-play AML simulator — clean scenarios, reward composition, OS metrics

audit: fix tool roster (10+3+5=18), align adversary default to local Llama, add accelerate/bitsandbytes to README deps

docs: rewrite PROJECT_CONTEXT, TRAINING, README — correct 18-tool count, add Phase 3 FinCEN tools, verified terminal weights, unified GRPO/PLR/adversarial coverage

feat: Replace GPT-4o-mini with local Llama-3.1 8B in Adversary Agent

updated markdown

feat: Add PLR curriculum engine, GRPO trainer, and 5th AGUI panel; merge frontend updates

fix(ui): patch trim typeerror, globe invisible arcs, and leaflet latency

feat(ui): phase 4 ui polish - fix cytoscape, globe, leaflet rendering and unify command center layout

chore: Finalize Phase 1-3 upgrades (Memex UI, 3D Globe, Adversarial Training, Reward Caps)

Fix: Sync client.py and demo_eval.py with Phase 3 SAR schemas

Merge muaz-backend: Phase 3 FinCEN 4-pillar expansion

feat: Phase 3 FinCEN 4-pillar expansion — industrial-grade AML environment

docs: update PROJECT_CONTEXT.md and README.md for OpenEnv deployment