Push checkpoint to HF Hub instead of download (Dexifried/tiny-router-checkpoint)

fix: skip ONNX export on HF, download checkpoint zip, export locally

fix: eval uses --data-file, export uses --output-dir

fix: use --encoder-lr and --head-lr instead of --lr

fix: suggested_hardware must be h200

Replace with tiny-router trainer (ZeroGPU/H200)

Add ZeroGPU-compatible spaces hook

Replace old neural bake trainer with polymarket paper trading lab

Neural Update: 2026-03-10 19:07:51

Architecture: Stripped deprecated pixel constraints, implemented fluid Vega-Lite containers for iOS

Architecture: Upgrading to Gradio 6.0 API, Vega-Lite charting, and Mobile UX

UX/Infra: Advanced mobile Mission Control & Model Repo redirection

Deployment: Adding HfApi upload_folder logic for weight persistence

Fix: Removing redundant tokenizer argument from Trainer for compatibility

Fix: Updating deprecated evaluation_strategy to eval_strategy

Hardware: Injecting @spaces.GPU for A100 allocation

Neural Update: 2026-03-10 07:53:15

Neural Update: 2026-03-10 05:42:00

Neural Update: 2026-03-09 21:09:57

Neural Bake V2: 900+ Gold Standard Rows

Auto-detect pushed CSV

Fix Gradio error handling

Baking the Brain on H200

initial commit