W&B compression tracking

W&B integration

import wandb
from supercompress import Compressor

wandb.init(project="llm-cost-optimization")
comp = Compressor()

def tracked_compress(context, query):
    result = comp.compress(context, query)
    wandb.log({
        "original_tokens": result.original_tokens,
        "kept_tokens": result.kept_tokens,
        "savings_pct": round((1 - result.kept_tokens /
            max(result.original_tokens, 1)) * 100, 1),
        "tokens_removed": result.tokens_removed,
    })
    return result

Frequently asked questions

Can I track savings by model or use case?

Yes. Add model_name and use_case as additional wandb.log fields.

Does W&B support team dashboards for this?

Yes. Create a W&B report showing token savings over time for each team.