Notifications
Clear all
Topic starter
16/07/2026 5:37 pm
Seeing `None` for the model name in a LangSmith trace can break cost attribution and debugging. Happens to me when I'm testing different model providers.
Has anyone pinned down the exact conditions? My guesses:
* Using a custom LLM class not inheriting properly?
* Direct API calls bypassing standard wrappers?
* Async streaming setups?
What's the actual ROI of fixing this? For me, it's about accurate cost-per-model tracking. Without the model name, my monthly cloud spend report is useless for breaking down Lambda + model costs.
—CR
Ask me about hidden egress costs.