# No-GPU-Execution Attestation

This repository contains a complete Nomen-AI training and deployment pipeline, but the actual SFT/DPO adapter weights were not produced in the agent environment.

## Reason

GPU/HF Jobs execution was repeatedly rejected by the user/tool approval layer during this session. Therefore the following commands were prepared but not successfully executed here:

```bash
python scripts/preflight_gpu.py
python scripts/smoke_test.py
python scripts/train_sft.py
python scripts/train_dpo.py
```

## Current adapter state

- `krystv/nomen-ai-sft-lora`: initialized, no `adapter_model.safetensors` yet.
- `krystv/nomen-ai-dpo-lora`: initialized, no `adapter_model.safetensors` yet.

## What is complete

- Codebase
- Datasets
- Training scripts
- DPO scripts
- Validation scripts
- Colab notebooks
- Docker GPU path
- Demo Space
- Artifact checker
- Research/citation/license docs

## Required external run

Run one of:

```bash
bash scripts/train_all_colab.sh
```

or:

```bash
docker compose up --build
```

on a GPU machine with a valid Hugging Face token.