Blog: The Digital Traffic Jam

Remove old blog file

Update README with latest

Blog: The Digital Traffic Jam - How We Gave Linux a 160-IQ Brain

Remove old BLOG.md

Blog: The Digital Traffic Jam - How We Gave Linux a 160-IQ Brain

Updated blog: The Digital Traffic Jam - engaging storytelling + technical depth

Add detailed project blog post

Benchmark: benchmark_results.json

Benchmark: benchmark_summary.png

Benchmark: benchmark_action_distribution.png

Benchmark: benchmark_rolling_reward.png

Benchmark: benchmark_reward_distribution.png

Benchmark: benchmark_cumulative_reward.png

Add training plot: grpo_training.png

Add training plot: strategist_warmstart_training.png

Add training plot: world_model_training.png

Add training plot: training_summary.png

Fix: keep Colab defaults, install only trl+peft with --no-deps, handle all TRL versions

Fix: let pip resolve all versions, uninstall old transformers, auto-restart if needed

Fix: uninstall old transformers first, pin exact versions, restart runtime

Fix: reinstall torchvision first, pin transformers 4.47-4.49, no runtime restart

Fix: use Colab default torch, auto-detect TRL API, no version pins

Fix: keep Colab default torch, only install sympy fix + training libs

Fix: restart runtime after pip install to resolve sympy/torch conflict

Fix: pin sympy + all deps to avoid Colab version conflicts

Add training notebook (runnable on Colab/HF)

Tokenizer

Merged strategist (warm-start + GRPO)

GPU training script for HF

Training pipeline scripts

LoRA adapter (warm-start SFT)

initial commit