# Manually running diffusion evaluation pipeline

We have a wandb run: https://wandb.ai/suno/diff_dpo/runs/l5ev9y81/overview

With checkpoint: `/app/suno/checkpoints/2025-05-23_05-21-32_s6113/last_ckpt_infer.pt`

1. Copy ckpt to s3 and overwrite the evaluation checkpoint
   
```
aws s3 cp /app/suno/checkpoints/2025-05-23_05-21-32_s6113/last_ckpt_infer.pt \
s3://suno-data/christian/checkpoints/diffusion/modal_eval_ckpt.pt
```

3. Launch model script passing the wandb run name
```
modal run sunoDiff/modal_eval.py --run-name "v2_infill_d3_t10"
```

6. Pull down results from s3 to cluster (generated audio and metadata)
```
aws s3 sync s3://suno-data/christian/outputs/auk-clips-up-u-2/ \
/app2/suno/data/eval_outputs/auk-clips-up-u-2
```

8. Run the notebook to generate plots: `diffusion_eval.ipynb`