Upload sharded 2GB SFT+DPO final model with clean native tokenizer files

Delete tokenizer.json with huggingface_hub

Upload sharded (chunked) SFT+DPO final model and tokenizer files

Delete tokenizer.json with huggingface_hub

Upload sharded SFT+DPO model with original unbroken tokenizer

Delete tokenizer.json with huggingface_hub

Upload sharded (chunked) SFT+DPO final model and tokenizer files

Delete tokenizer.json with huggingface_hub

Upload merged SFT + DPO model

Delete tokenizer.json

Update: no_think SFT+DPO LoRA merged — eliminates <think> loop

Add files using upload-large-folder tool