Classify phase 2 as continued pretraining refinement

Point instruction post-training to full SFT phase

Rename release as TR-HASH MoE and repair provenance links

Add matching 32k tokenizer files for standalone inference

Update model card with phase-2 progress and interim PIQA

Delete training/tensorboard

Add files using upload-large-folder tool

Upload SFT checkpoint step 8156

Rename download.json to config.json

Rename config.json to download.json

Update config.json

Upload config.json

Remove internal script/config paths, keep it reader-facing

Document optional phase-3 LoRA instruction post-training

Reserve repo for phase-2 SFT, plan card, no weights yet

initial commit