dharshanrai's picture
Upload README.md with huggingface_hub
d77afcf verified
|
Raw
History Blame Contribute Delete
1.08 kB
---
license: cc-by-4.0
pretty_name: TraceLab coding-agent traces (DuckDB mirror)
tags:
- agents
- code
- llm
- cost-analysis
---
# TraceLab coding-agent traces — DuckDB mirror
Unmodified mirror of `syfi_coding_trace.duckdb` from the
[TraceLab](https://github.com/uw-syfi/TraceLab) project by the University of
Washington SYFI lab: ~4,300 real Claude Code and Codex sessions (357K LLM
rounds) with per-round token accounting (`prefix_tokens`, `append_tokens`,
`cache_creation_input_tokens`, `input_tokens`, `output_tokens`, timing events).
- **Upstream**: https://github.com/uw-syfi/TraceLab (GitHub release asset)
- **License**: [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) —
credit the TraceLab authors (UW SYFI lab) when using this data.
- **SHA-256**: `97715265367cc72376475f5d444c8e1900b88cab1482aa7b9a742894d9f15619`
Mirrored here to back the
[summarize-at-cache-cold](https://huggingface.co/spaces/dharshanrai/summarize-at-cache-cold)
Space, which replays these sessions to model prompt-cache-miss costs and
summarize-on-cache-cold savings.