Strip leaked W&B API key from training notebook

Drop 'small model' claim (training not yet converged)

Point to new 300-step 7B W&B run (axwh8hrb)

Drop 'small model' claim (training not yet converged)

Strip em-dashes; fix trap-library facts (4 held-out, 7 categories); explicit hardware-optimisation framing; cleaner 1x2 results grid

Drop 'small model' claim (training not yet converged)

Point to new 300-step 7B W&B run (axwh8hrb)

Point to new 300-step 7B W&B run (axwh8hrb)

Sync chunk: docs (diagrams + plots)

Sync chunk: training scripts + notebook

Sync chunk: top-level files (md + manifest + dockerfile + pyproject)

Point to new 300-step 7B W&B run (axwh8hrb)

Drop 'small model' claim (training not yet converged)

Drop 'small model' claim (training not yet converged)

Add training metrics grid; fix hardware-target claims (no RISC-V/Cortex-A78); add HF Jobs link

Push diagrams + updated README (blog assets, env explainer)

Auto-update blog with latest job results

Add HF job links (login-gated) to blog links section

Add public W&B run link for 300-step rerun

Upload training_runs/partial-20260426-061351/grpo_component_means.png with huggingface_hub

Upload training_runs/partial-20260426-061351/grpo_reward_curve.png with huggingface_hub

Upload training_runs/partial-20260426-061351/baseline_vs_trained_metrics.png with huggingface_hub

Upload training_runs/partial-20260426-061351/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260426-061351/summary.json with huggingface_hub

Upload training_runs/partial-20260426-060026/grpo_component_means.png with huggingface_hub

Upload training_runs/partial-20260426-060026/grpo_reward_curve.png with huggingface_hub

Upload training_runs/partial-20260426-060026/baseline_vs_trained_metrics.png with huggingface_hub

Upload training_runs/partial-20260426-060026/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260426-060026/summary.json with huggingface_hub

Sync latest submission updates (GRPO-only + blog markdown + bugfixes)

Upload training_runs/partial-20260425-211348/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260425-211348/summary.json with huggingface_hub

Upload training_runs/partial-20260425-210137/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260425-210137/summary.json with huggingface_hub

Upload training_runs/partial-20260425-205532/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260425-205532/summary.json with huggingface_hub

Improve training data quality, teacher policy, ablation, and reward-aligned stage

Upload training_runs/partial-20260425-161650/reward_distribution.png with huggingface_hub

Upload training_runs/partial-20260425-161650/summary.json with huggingface_hub

Add LoRA-enabled script runner with auth-safe snapshot download

Add eval progress prints for long runs

Improve notebook runner logging and timeout safety

Use LoRA adapter tuning to avoid T4 OOM

Sync latest notebook training fixes

Add notebook runner script for HF Jobs

Use pure fp16 weights without AMP scaler in SFT

Disable grad clipping for stable fp16 SFT on T4

Fix CUDA dtype and device handling for full training

Enable GPU-friendly full training config

Upload training_runs/notebooks/openenv_hackathon_training.partial.20260425-124204.ipynb with huggingface_hub